cancers Review Advances in the Use of Deep Learning for the Analysis of Magnetic Resonance Image in Neuro-Oncology Carla Pitarch 1,2, * , Gulnur Ungan 3,4 , Margarida Julià-Sapé 3,4 and Alfredo Vellido 1,4 1 Department of Computer Science, Universitat Politècnica de Catalunya (UPC BarcelonaTech) and Intelligent Data Science and Artificial Intelligence (IDEAI-UPC) Research Center, 08034 Barcelona, Spain; alfredo.vellido@upc.edu 2 Eurecat, Digital Health Unit, Technology Centre of Catalonia, 08005 Barcelona, Spain 3 Departament de Bioquímica i Biologia Molecular and Institut de Biotecnologia i Biomedicina (IBB), Universitat Autònoma de Barcelona (UAB), 08193 Barcelona, Spain; gulnur.ungan@autonoma.cat (G.U.); margarita.julia@uab.cat (M.J.-S.) 4 Centro de Investigación Biomédica en Red (CIBER), 28029 Madrid, Spain * Correspondence: carla.pitarch@upc.edu Simple Summary: Within the rapidly evolving landscape of Machine Learning in the medical field, this paper focuses on the forefront advancements in neuro-oncological radiology. More specifically, it aims to provide the reader with an in-depth exploration of the latest advancements in employing Deep Learning methodologies for the classification of brain tumor radiological images. This review meticulously scrutinizes papers published from 2018 to 2023, unveiling ongoing topics of research while underscoring the main remaining challenges and potential avenues for future research identified by those studies. Beyond the review itself, the paper also underscores the importance of placing the image data modelling provided by Deep Learning techniques within the framework of analytical pipeline research. This means that data quality control and pre-processing should be correctly coupled with modelling itself, in a way that emphasizes the importance of responsible data utilization, as well as the critical need for transparency in data disclosure to ensure trustworthiness and reproducibility of findings. Citation: Pitarch, C.; Ungan, G.; Abstract: Machine Learning is entering a phase of maturity, but its medical applications still lag Julià-Sapé, M.; Vellido, A. Advances behind in terms of practical use. The field of oncological radiology (and neuro-oncology in particular) in the Use of Deep Learning for the is at the forefront of these developments, now boosted by the success of Deep-Learning methods for Analysis of Magnetic Resonance the analysis of medical images. This paper reviews in detail some of the most recent advances in the Image in Neuro-Oncology. Cancers 2024, 16, 300. use of Deep Learning in this field, from the broader topic of the development of Machine-Learning- https://doi.org/10.3390/ based analytical pipelines to specific instantiations of the use of Deep Learning in neuro-oncology; cancers16020300 the latter including its use in the groundbreaking field of ultra-low field magnetic resonance imaging. Academic Editor: Shinji Kawabata Keywords: machine learning; neuro-oncology; radiology; deep learning; data analysis pipeline; Received: 9 November 2023 ultra-low field magnetic resonance imaging Revised: 28 December 2023 Accepted: 8 January 2024 Published: 10 January 2024 1. Introduction Although Machine Learning (ML) is entering a phase of maturity, its applications in the medical domain at the point of care are still few and tentative at best. This paradoxical Copyright: © 2024 by the authors. Licensee MDPI, Basel, Switzerland. contradiction has been explained according to several different factors. One of them is the This article is an open access article lack of experimental reproducibility, a requirement in which ML models in health have been distributed under the terms and reported to fare badly in comparison to other application areas [1]. One main reason to conditions of the Creative Commons explain this is the mismatch between a data-centered (and often data-hungry) approach Attribution (CC BY) license (https:// and the scarcity of publicly available and properly curated medical databases, combined creativecommons.org/licenses/by/ with a nascent but insufficient data culture at the clinical level [2]. Another factor has to do 4.0/). with regulatory issues of ML (and Artificial Intelligence in general) in terms of both lack of Cancers 2024, 16, 300. https://doi.org/10.3390/cancers16020300 https://www.mdpi.com/journal/cancers Cancers 2024, 16, 300 2 of 55 maturity and geographical heterogeneity [3]. Further elements hampering ML-based tools adoption include data leakage, dataset shift, required model recalibrations, analytical pipeline maintenance failures, or changing medical practice patterns, to name a few [4]. The field of oncological radiology (and neuro-oncology in particular) is arguably at the forefront of the practical use of ML in medicine [5], now boosted by the success of Deep- Learning (DL) methods for the analysis of medical images [6,7]. Unfortunately, though, imaging does not escape the challenges and limitations summarized in the previous para- graph. Central to them, what has been called the “long-tail effect” [8]: pathologies for which only small and scattered datasets exist due to the scarcity of clinical data management strategies (technically complex and expensive) at levels beyond the local (regional, national, international). Associated with this, we must account for the difficulty of achieving stan- dardized labeling (annotation) of imaging databases. An example of how to deal effectively with these problems is Federated Learning, which was used in [9] to gather data from 71 sites from 6 continents, analyzed using ML to address a problem of tumor boundary detection for glioblastoma brain tumors. Please note that the resulting database includes 6314 cases, which is impressive for this medical domain but still modest from an ML perspective. The success of ML in oncological radiology, as summarily stated in [10], will depend on its ability to create value in the delivery of medical care in terms of “increased diagnostic certainty, decreased time on task for radiologists, faster availability of results, and reduced costs of care with better outcomes for patients”. This paper surveys some of the most recent advances in the use of ML for the analysis of magnetic resonance imaging (MRI) data in neuro-oncology without trying to make an all-encompassing review out of it. Instead, we focus on the most rapidly developing area, which involves the use of methods from the DL family. The variety of approaches sprouting from this family of methods has shaken the standards of data pre-processing or feature engineering before modeling as such. For this reason, we proceed to address the review hierarchically, starting in Section 3 with the broader topic of the development of ML-based analytical pipelines, which addresses the data analysis process beyond specific models and in which we will provide examples from two promising feature engineering approaches, namely source extraction in the form of independent component analysis (ICA) and nonnegative matrix factorization (NMF), and radiomics. The review of DL methods for image data analysis as such is delivered in Section 4. As an addition to this section, we will discuss the potential uses of DL in the groundbreaking field of ultra-low field (ULF) MRI [11]. Before all this, the following section will provide some contextual basic definitions of neuro-oncology concepts and a description of the main challenges and open issues concerning the use of ML in this domain. 2. Open Problems in AI Applied to MRI Analysis The open problems for the use of ML-based analytical processes in the field of MRI in neuro-oncology can be seen from different perspectives. The first one is the analytical problem itself, according to which the main division is into categorization and segmentation problems. The latter is commented on later in this section. Categorization can, in turn, be split into diagnosis and prognosis. In diagnosis, the correlation between neuroimaging classifications and histopathological diagnoses was assessed in [12] based on the 2000 version of the WHO classification of brain tumors and in [13] based on the 2007 version. In both studies, the main finding was that the sensitivity was variable among classes, whereas specificity was in the range of 0.85–1. The most difficult categories to diagnose were the glioma subtypes. The study based on the 2000 classification [12] reported a sensitivity of 0.14 for low-grade astrocytoma and 0.15 for low-grade oligodendroglioma. In the study based on the 2007 classification [13], increased sensitivity for low-grade astrocytoma (0.56) was found, but sensitivity was still low for other low-grade gliomas (LGG) such as oligodendroglioma (0.26), or for anaplastic gliomas (astrocytoma, 0.17 or ependymoma, 0.00), and other classes in the long-tail such as meningiomas of grade II and III in aggregate (0.17), or subependymomas and choroid Cancers 2024, 16, 300 3 of 55 plexus papilloma (0.33 for both). The recently released 2021 WHO classification [14], which incorporates the genetic alterations, opens the door to the reevaluation of these baseline results to accurately estimate the added value of any clinical decision support system (CDSS) based on ML or radiogenomics, over the limits of radiological interpretation of imaging findings. It is reasonable to foresee that the problematic tumor categories will remain so, or even more challenging, given the enhanced stratification of the glial category (e.g., different mutations of IDH1/2, ATRX, TP53, BRAF, H3F3A, CDKN2A/B, TERT and MGMT promoters, EGFR amplification, GFAP, 1p/19q codeletion, etc.). On the other hand, regarding follow-up, there is no standard of care for recurrent high- grade gliomas, and the currently accepted criteria to assess response are those established by the Response Assessment in Neuro-Oncology Working Group (RANO) [15], which deals with the pseudoprogression phenomenon, defined as the appearance of contrast- enhancing lesions during the first 12 weeks after the end of the concomitant treatment or when the lesion developed within the first 3–6 months after radiation therapy, if it is in the radiation field (inside the 80% isodose line), and especially if it presents as a pattern of enhancement related to radiation-induced necrosis enhancement [16]. Also, with pseudoresponse in those patients treated with antiangiogenics in countries where these are approved [17,18]. Antiangiogenic agents, like bevacizumab, are designed to block the VEGF effect. The mechanism of action may be related to decreased blood supply to the tumor and normalization of tumor vessels, which display increased permeability. These agents are associated with high radiologic responses if we evaluate only the contrast enhancement. The recently published RANO 2.0 criteria [19] refine the former RANO, distinguishing between high-grade and low-grade gliomas. RANO 2.0 also takes into account the IDH status to decide whether the surrounding non-enhancing region should be taken into account or not. In this sense, ML-based pipelines should ideally be designed to allow the evaluation of their added value with respect to medical guidelines for clinical decision-making. Another viewpoint to approach the open problems in the field has to do with the fact that ML-based analysis is strongly dependent on data pre-processing and the post-processing of results. A fundamental prerequisite for the successful application of DL models in brain tumor classification is the pre-processing of the MRI data. The key pre-processing steps for the harmonization of MRI data are as follows. • Resampling: MRI scans can exhibit variations in resolution and voxel sizes depending on the acquisition system. Resampling standardizes the resolution across the MRI images to ensure uniform dimensions. • Co-registration: entails the alignment of MRI scans with a standardized anatomical template with the purpose of situating different scans within the same anatomical coordinate system. • Skull-stripping: The main objective of the skull-stripping step is to efficiently isolate the cerebral region of interest from non-cerebral tissues, which enables DL models to focus exclusively on those brain tissues. • Bias Field Correction: aims to rectify intensity inhomogeneities that are pervasive in MRI scans to guarantee uniformity in intensity values. The technique of choice for bias field correction is N4ITK (N4 Bias Field Correction) [20], which is an improved variant of the N3 (non-parametric nonuniform normalization) retrospective bias correction algorithm [21]. • Normalization: a technique adopted to rescale intensity values of MRI scans to a numeric range, rendering them consistent across the dataset. This process mitigates scale-related disparities. Two prominent approaches commonly applied to MRI data as input for DL models are min-max normalization and z-score normalization. Min-max achieves its goal by rescaling intensity values within MRI scans, spanning their range between 0 and 1. In contrast, z-score, often referred to as standardization, transforms Cancers 2024, 16, 300 4 of 55 the distribution of intensity values by centering it around a zero mean and standard deviation of value 1. • Tumor identification: A critical and optional pre-processing step before the classifica- tion task involves identifying the tumor region of interest (ROI) through segmentation or by defining a bounding box that encompasses the tumor. Popular DL architectures, such as UNet [22], Faster-RCNN [23], and Mask-RCNN [24] are often employed to perform such segmentation or detection tasks. The post-processing of results must often address the fact that the DL family of methods is, by their nature, an extreme case of black-box approach, a characteristic that may strongly hamper their medical applicability [25]. This limitation can be addressed using explainability and interpretability strategies; for further details on these, the reader is referred to [25]. 3. Ml-Based Analytical Pipelines and Their Use in Neuro-Oncology Ultimately, the whole point of using ML methods for data-based problems in the area of neuro-oncology is to provide radiologists with evidence-based medical tools at the point of care that can assist them in decision-making processes, especially with ambiguous or borderline cases. This is why it makes sense to embed these methods in Clinical Decision Support Systems (CDSS). A thorough and systematic review of intelligent systems-based CDSS for brain tumor analysis based on magnetic resonance data (spectra or images) is presented in this same Special Issue of Cancers [26]. It reports their increasing use over the last decade, addressing problems that include general ones such as tumor detection, type classification, and grading, but also more specific ones such as physicians’ alerting of treatment change plans. At the core of ML-based CDSS, we need not just ML methods, models, and techniques but, more formally, ML pipelines. An ML pipeline goes beyond the use of a collection of methods to encompass all stages of the data mining process, including data pre-processing (data cleaning, data transformations potentially including feature selection and extraction, but also other aspects of data curation such as data extraction and standardization, missing data imputation and data clinical validation [27]) and models’ post-processing, potentially including evaluation, implementation and the definition of interpretability and explain- ability processes [25]. Pipelines can also accommodate specific needs, such as those related to the analysis of “big data”, with their corresponding challenges of standardization and scalability. As described in [28], in a clinical oncology setting, this may require a research infrastructure for federated ML based on the findable, accessible, interoperable, and reusable (FAIR) principles. Alternatively, we can aspire to automate the ML pipeline definition using Automated ML (AutoML) principles, as in [29], where Su and co-workers used a Tree-based Pipeline Optimization Tool (TPOT) in the process of selecting radiomics features predictive of mutations associated with midline gliomas. An example of an ML pipeline for the specific problem of differentiation of glioblas- tomas from single brain metastases based on MR spectroscopy (MRS) data can be found in [30]. In this same issue of Cancers, Pitarch and co-workers [31] describe an ML pipeline for glioma grading from MRI data with a focus on the trustworthiness of the predictions generated by the ML models. This entails robustly quantifying the uncertainty of the models regarding their predictions, as well as implementing procedures to avoid data leakage, thus avoiding the risk of unreliable conclusions. All of these can be seen as part of a quest to avoid the pitfalls of implementation of ML-based CDSS that result in the problems of limited reproducibility of analytical results in clinical practice that have been reported in recent studies [1]. As previously explained, the first stages of an ML pipeline, prior to the data modeling itself, involve data pre-processing, and this task may, in turn, involve many sub-problems. As an example of the potential diversity and complexity of this landscape, we comment here on a few recently selected contributions to the problem of feature engineering and extraction Cancers 2024, 16, 300 5 of 55 following just two particular and completely different approaches: statistical image feature engineering using radiomics and source extraction using ICA- and NMF-based methods. Radiomics is an image transformation approach that aims to extract either hard-coded statistical or textural features based on expert domain knowledge or feature representations learned from data, often using DL methods. The former may include first-order statistics, size and shape-based features, image intensity histogram descriptors, image textural infor- mation, etc. The use of this method for the pre-processing of brain tumor images prior to the use of ML has been recently and exhaustively reviewed in [32]. From that review, it is clear that the predominant problem under analysis is diagnosis, with only a limited number of studies addressing prognosis, survival, and progression. The types of brain tumors under investigation are dominated by the most frequent classes. In particular, glioblastoma, either on its own or combined with metastasis as a super-class of aggressive tumors, is the subject of many studies, with some others also including other frequent super-classes such as low-grade glioma or meningioma, while minority tumor types and grades are only considered in a limited number of studies. Importantly, and related to our previous comments concerning scarce data availability, most of the studies reported in [32] work with very small sample sizes, often not reaching the barrier of 100 cases. The challenge posed by data scarcity is compounded by the fact that most of the studies extract Radiomic features in the hundreds if not the thousands. This means that the ratio of cases-to-features is extremely low, making the use of conventional ML classifiers very difficult. To alleviate this problem, most of the reviewed papers resort to different strategies for qualitative and quantitative feature selection. Image modalities under analysis are dominated by T1, T2, and FLAIR, with few exceptions (PET, or Diffusion- and Perfusion-Weighted Imaging). Most studies are shown to resort to the Area Under the ROC Curve (AUC) as a performance metric, which is a safe choice, as it is far more robust than plain accuracy for small and class-imbalanced datasets. The use of radiomics as a data transformation strategy in pre-processing is facilitated by the existence of off-the-shelf software such as the open-source PyRadiomics package [33]. Source extraction methods have a very different analytical rationale for data dimen- sionality reduction as a pre-processing step. They do not achieve it through plain feature transformation, as in radiomics. Instead, they aim to find the underlying and unobserved sources of observed radiological data. In doing so, they achieve dimensionality reduction as a byproduct of a process that may provide insight into the generation of the images themselves. The ICA technique [34] has a long history in medical applications, most notoriously for the analysis of electroencephalographic signals. Source extraction is natural in this context as a tool for spatially locating sources of the EEG from electric potentials measured in the skull surface. In ICA, we assume that the observed data can be expressed as a linear combination of sources that are estimated to be statistically independent or as independent as possible. This technique has mostly been applied to brain tumor segmentation, but some alternative recent studies have extended its possibilities to dynamic settings, such as that in [35], where dynamic contrast-enhanced MRI is analyzed using temporal ICA (tICA), and in [36], where probabilistic ICA is used for the analysis of dynamic susceptibility contrast (DSC) perfusion MRI. The NMF technique [37], on the other hand, was originally devised for the extraction of sources from images and assumes data non-negativity but does not assume statistical independence. Data are still approximated by linear combinations of factors. Although NMF and variants of this family of methods have extensively been used for the pre- processing and analysis of MRS and MRS imaging (MRSI) signal [38,39], they have only scarcely been used for the pre-processing of MRI. Some outstanding exceptions include the work in [40] with hierarchical NMF for multi-parametric MRI and the recent proposal of a whole new architecture based on NMF called Factorizer [41], constructed by replacing the self-attention layer of a Vision Transformer (ViT, [42]) block with NMF-based modules. Cancers 2024, 16, 300 6 of 55 The technical details of ICA and NMF and their manifold variants are beyond the scope of this review and can be found elsewhere in the literature. 4. Deep Learning in Neuro-Oncology Data Analysis: A Review In this section, we review existing recent literature to gather evidence about the advantages, challenges, and potential future directions in the use of DL techniques for supervised problems in neuro-oncology. Furthermore, we aim to provide insights into the current state-of-the-art methodologies, address their limitations, and identify areas for further research. Ultimately, our objective is to facilitate the development of robust, responsible, and applicable DL solutions that can effectively contribute to the field of neuro-oncology. 4.1. Overview of the Main DL Methods of Interest Recent advances in the DL field have brought about new possibilities in medical imaging analysis and diagnosis. One of its arguably most successful models is Convolu- tional Neural Networks (CNNs), a widely used type of DL algorithm, well known for its ability to capture spatial correlations within image pixel data hierarchically. They have shown promise in medical imaging tasks [43–45], enabling improved tumor detection, classification, and prognosis assessment. The input data of a CNN is represented as a tensor with dimensions in the format of (channels, depth, height, width). Notably, the “depth” dimension is specific to 3D images and not applicable to 2D data, and “height” and “width” correspond to the image’s spatial dimensions. In practical terms, the number of channels for color images is translated into three, representing Red, Green, and Blue (RGB) components, while gray-scale images consist of a single channel. The most characteristic operation in a CNN is called convolution, which gives the name to the convolutional layers. These layers capture spatial correlations by applying a set of filters or kernels across all areas of the input image data and compute the weighted sum, resulting in the generation of a feature map as an output. This feature map contains essential characteristics extracted by the actual layer and serves as the input for subsequent layers of processing. Another useful layer used in CNNs is the pooling layer. The pooling operation consists of downsampling the feature maps obtained from the convolution operation. The idea is to reduce the dimensionality without losing significant information. There are mainly two kinds of pooling: max-pooling and average-pooling. The outputs of convolutional layers are often passed through activa- tion functions to introduce non-linearity. The most popular activation functions are ReLU, which inactivates negative values in the output through the formula f ( x ) = max (0, x ); Sigmoid, which maps output values between 0 and 1 using the equation f ( x ) = 1+1e−x ; and SoftMax, which is the extension of Sigmoid for multi-class problems. CNNs often consist of multiple layers that work together to learn hierarchical high- level image features. These layers progressively extract more abstract and complex in- formation from the input image data. In the final step, the last feature map is passed through a fully connected layer, resulting in a one-dimensional vector. To obtain the class probabilities, Sigmoid or SoftMax are applied to this vector. several networks have made significant contributions to the world of DL. AlexNet [46], GoogLeNet [47], InceptionNet, VGGNet [48], ResNet [49], DenseNet [50], and Efficient- Net [51] are among the most widely used CNNs to extract patterns from medical imaging. DL models are considered data-hungry since they require substantial amounts of data for effective training. In the realm of medical data analysis, a primary challenge, as previously mentioned, is the inherent data scarcity and class imbalance. Common solutions to address this challenge include the application of data-augmentation (DA) methods and transfer-learning (TL) techniques. Data Augmentation techniques are a crucial strategy to mitigate the challenge of limited annotated data in medical image analysis. These methods encompass a range of transformations applied to existing images, effectively expanding the dataset in terms of both size and diversity. Former approaches involve a wide range of geometric modifications Cancers 2024, 16, 300 7 of 55 such as rotation, scaling, flipping, cropping, zooming, or color changes. Beyond traditional augmentations, advanced methods like Generative Adversarial Networks (GANs) [52] are used to generate new synthetic and realistic examples. The idea behind TL is to leverage pre-trained models, typically trained in large and diverse datasets, and adapt them for the specific task at hand, for which we might not have such a representative sample. Widely used pre-trained CNNs, such as ImageNet [53] or MS-COCO [54], have been originally developed from 2D large-scale datasets. However, a notable challenge when dealing with medical image data is the limited availability of large and diverse 3D datasets for universal pre-training [55]. Transferring the knowledge acquired from the 2D to the 3D domain proves to be a non-trivial task, primarily due to the fundamental differences in data structure and representation between these two contexts. To tackle this challenge and address the limitation of limited data, a broadly used strategy is to decompose 3D volumes into individual 2D slices within a determined anatomical plane. However, the decomposition of 3D volumes into individual 2D slices introduces a potential data leakage concern. This issue arises when 2D slices from the same individual inadvertently end up in both the training and testing datasets in an analytical pipeline. Such data leakage can lead to overestimations of model performance and affect the validity of experimental results. In addition, it is important to note that this approach comes with the trade-off of losing the 3D context present in the original data. Recent efforts have aimed at overcoming these challenges. Banerjee et al. [56] classified low-grade glioma (LGG)and high-grade glioma (HGG) multi-sequence brain MRIs from TCGA and BraTS2017 data using multiple slice-based approaches. In their work, they provided a comparison of the performance obtained with CNNs trained from scratch on 2D image patches (PatchNet), entire 2D slices (SliceNet), and multi-planar slices through a final ensemble method that averages the classification obtained from each anatomical view (VolumeNet). The classification obtained with these models is also compared with pre-trained VGGNet and ResNet on ImageNet. The multi-planar method outperformed the rest of the approaches with an accuracy of 94.74%, and the lowest accuracy (68.07%) was obtained with pre-trained VGGNet. Unfortunately, TCGA and BraTS data share some patient data, which could involve an overlap between training and testing samples and hence be prone to data leakage. Ding et al. [57] combined radiomics and DL features using 2D pre-trained CNNs using single-plane images and performing a subsequent multi-planar fusion. VGG16, in combination with radiomics and RF, achieved the highest accuracy of 80% when combining the information obtained from the three views. Even though the multi-planar approach processes the information gathered from the axial, coronal, and sagittal views, it is still essentially a 2.5D approach, weak at fully capturing 3D contexts. Zhuge et al. [58] presented a properly native 3D CNN for tumor segmentation and subsequent binary glioma grade classification and compared it with a pre-trained 2D ResNet50 on ImageNet with previous tumor detection, employing a Mask R-CNN. The results of the 3D approach were slightly higher than the 2D ones, reporting 97.10% and 96.30% accuracy, respectively. In their study, Chatterjee et al. [59] explored the role of (2+1)D, mixed 2D–3D, and native 3D convolutions based on ResNet. This study highlights the effectiveness of mixed 2D–3D convolutions, achieving an accuracy of 96.98%, surpassing both the (2+1)D and the pure 3D approaches. Furthermore, the use of pre-trained networks demonstrated enhanced performance in the spatial models, yet, intriguingly, the pure 3D model performed better when trained from scratch. A study conducted by Yang et al. [55] introduced ACS convolutions, a novel approach that facilitates TL from models pre-trained on large publicly accessible 2D datasets. In this method, 2D convolutions are divided by channel into three parts and applied separately to the three anatomical views (axial, coronal, and sagittal). The effectiveness of this approach was demonstrated using a publicly available nodule dataset. Subsequently, Baheti et al. [60] further advanced the application of ACS convolutions by showcasing their enhanced performance on 3D MRI brain tumor data. Their study provides evidence of notable improvements in both segmentation and radiogenomic classification tasks. Cancers 2024, 16, 300 8 of 55 4.2. Publicly Available Datasets Access to large and high-quality datasets plays a crucial role in the development and evaluation of robust DL classification algorithms. This section aims to provide a comprehensive review of several publicly accessible datasets that have been widely used in brain tumor classification tasks and DL research. These datasets encompass diverse tumor types, imaging modalities, and annotated labels, facilitating the advancement of computational methods for accurate tumor classification. Table 1 provides a detailed overview of the most frequently used datasets in the literature. Cancers 2024, 16, 300 9 of 55 Table 1. An overview of publicly available MRI datasets for brain tumor classification benchmarking. Dataset Categories Dim. Sample Size MRI Modalities 2020 3D 369 (LGG: 76, HGG: 293) T1, T1c, T2, FLAIR 2019 3D 335 (LGG: 76, HGG: 259) T1, T1c, T2, FLAIR Low-Grade Glioma (LGG) 2018 3D 284 (LGG: 75, HGG: 209) T1, T1c, T2, FLAIR BraTS [61] 2017 3D 285 (LGG: 75, HGG: 210) T1, T1c, T2, FLAIR High-Grade Glioma (HGG) 2015 3D 274 (LGG: 54, HGG: 220) T1, T1c, T2, FLAIR 2013 3D 30 (LGG: 10, HGG: 20) T1, T1c, T2, FLAIR 2012 3D 30 (LGG: 10, HGG: 20) T1, T1c, T2, FLAIR Astrocytoma (AS) IDH-mutant Oligodendroglioma (OG) IDH-mutant Training: 221 (AS: 54, OG: 34, GB: 133) CPM-RadPath [62] 3D T1, T1c, T2, FLAIR 1p/19q codeletion [unseen sets] Val: 35, Test: 73 Glioblastoma (GB) IDH-wildtype Meningioma (MN), Glioma (GL), Figshare [63] 2D 233 (MN: 82, GL: 89, PT: 62) T1c Pituitary (PT) IXI [64] Healthy 3D 600 T1, T2, PD, DW Kaggle-I [65] Healthy (H), Tumor (T) 2D 3000 (H: 1500, T: 1500) - Healthy (H), Meningioma (MN), 3264 (H: 500, MN: 937, GL: 926, PT: Kaggle-II [66] 2D - Glioma (GL), Pituitary (PT) 901) Kaggle-III [67] Healthy (H), Tumor (T) 2D 253 (H: 98, T: 155) - Radiopaedia [68] - - - - Oligodendroglioma (OG), 111 (OG: 21, AS: 47, GB: 44) Astroctyoma (AS), Glioblastoma (GB) REMBRANDT [69] 3D T1, T1c, T2, FLAIR Grade II (G.II), Grade III (G.III), Grade 109 (G.II: 44, G.III:24, G.IV: 44) IV (G.IV) TCGA-GBM [70] Glioblastoma 3D 262 T1, T1c, T2, FLAIR Grade II (G.II), Grade III (G.III) 197 (G.II: 100, G.III: 96, discrepancy: 1) Astroctyoma (AS), TCGA-LGG [71] 3D T1, T1c, T2, FLAIR Oligodendroglioma (OG), 197 (AS: 64, OG: 86, OAS: 47) Oligoastrocytoma (OAS) DW: Diffusion-weighted, FLAIR: Fluid Attenuated Inversion Recovery, PD: Proton Density, T1c: contrast-enhanced T1 weighted. Cancers 2024, 16, 300 10 of 55 The Brain Tumor Segmentation Challenge (BraTS) and The Computational Precision Medicine: Radiology-Pathology Challenge on Brain Tumor Classification (CPM-RadPath) datasets were created for two popular challenges held at the MICCAI (Medical Image Computing and Computer Assisted Intervention) Conference. The BraTS Challenge [61] was initially developed in 2012 to benchmark tumor segmen- tation methods distinguishing glioblastoma from “lower grades”. Notably, this challenge provides not only MRI data but also clinical labels, including a binary classification of glioma grades. Even though their definition does not fully align with WHO’s terminology, they include grades 2 and 3 when referring to “lower grades”. Throughout the years, the BraTS Challenge has continually evolved, expanding to include additional tasks and diverse datasets. In 2017, the dataset was enriched by integrat- ing data from the TCIA repository, specifically including samples from the TCGA-LGG [71] and TCGA-GBM [70] datasets. It is worth noting that TCGA-LGG data provides labels to differentiate between gliomas of grades 2 and 3. Although the primary focus of the BraTS Challenge has traditionally centered on automated brain tumor segmentation, it has grown to become a widely adopted resource for brain tumor grade classification. Recent challenges have included tasks such as survival prediction and genetic classification, and the 2023 challenge even included image synthesis tasks. CPM-RadPath [62] from 2019 was designed to evaluate brain tumor classification algorithms in three classes, taking into account the WHO classification of 2016: A (astrocy- tomas grades II and III, IDH-mutant), O (oligodendroglioma grades II and III, IDH-mutant, 1p/19q codeleted) and G (Glioblastoma and diffuse astrocytic glioma with molecular fea- tures of glioblastoma, IDH-wildtype (Grade IV)), interestingly grouping the anaplasic with the low grades in the A and O classes. This challenge provides participants with paired radiology scans and digitized histopathol- ogy images. It is worth noting that the data provided by these challenges are distributed after pre-processing, involving co-registered to the same anatomical template, interpolated to a consistent resolution of 1 mm3, and skull-stripped. The datasets under consideration encompass a variety of MRI modalities. Specifically, BraTS, CPM-RadPath, REMBRANDT, and TCGA comprise images from four key modal- ities: T1, T1 post-contrast weighted (T1c), T2-weighted, and Fluid Attenuated Inversion Recovery (FLAIR). The IXI dataset provides not only T1 and T2 but also Proton Density (PD) and Diffusion-weighted (DW) images. Notably, images on Figshare are limited to the T1c modality, while datasets from Kaggle and Radiopaedia lack this information. The images in the BraTS, CPM-RadPath, IXI, REMBRANDT, and TCGA datasets are stored in 3D structures using widely used medical image formats, specifically NIfTI or DICOM. In contrast, datasets sourced from Kaggle consist of 2D images in PNG format. Notably, Figshare contains 2D images in MATLAB data format. In the Figshare data reposi- tory, images are provided alongside a 5-fold CV split at the patient level to mitigate the risk of data leakage. The use of this split ensures that no patient is inadvertently present in both training and testing sets, thus preventing leakage. Moreover, this dataset comprises multiple 2D slices from the same patient in the three distinct anatomical perspectives. Con- versely, the datasets sourced from the Kaggle repository lack patient identifier information, making it challenging to ascertain if images are from unique patients or to trace the origin of the data. Figure 1 summarizes the prevalence of dataset usage in the reviewed literature, includ- ing public and private datasets. Datasets that appear in two or fewer papers are grouped under the “Others” category. It is worth highlighting that over 85% of the papers reviewed in this analysis make use of public datasets. It is essential to acknowledge that the sample sizes of the datasets, in general, are roughly in the hundreds range. This limited sample size can pose challenges in drawing robust and generalizable conclusions, which is a notable concern within the ML healthcare domain. Addressing the need for larger and more diverse datasets, as previously discussed, is an ongoing challenge in this field. Cancers 2024, 16, 300 11 of 55 Figure 1. Dataset usage prevalence across the reviewed literature. 4.3. Literature Review Various online repositories of scientific research articles, including PubMed, Google Scholar, and Scopus, were utilized to collect pertinent papers for this review section. The selection was restricted to the years 2018–2023. More specifically, only articles published prior to 30 June 2023 were taken into consideration. The document type was restricted to journal or conference papers. The focal keywords were centered on classifying brain tumors from pre-operative MRI images using DL techniques. While refining our choices, we excluded publications with ambiguous data explanations or lacking methodology details, as the utmost priority was placed on guaranteeing the strength and acuity of our conclusions. An initial identification process yielded a total of 555 papers, with 146 papers remaining after the screening procedure. Figure 2 depicts the distribution of these papers across the years under review, shedding some light on the temporal evolution of research in this domain. Figure 2. Yearly inclusion of articles in this review that focus on classifying brain tumors using DL and MRI scans. In the subsequent analysis, we provide comprehensive insights into the data sources and methodologies employed in the examined papers. Table A1 offers a detailed overview of the datasets, focusing on essential aspects such as the dimensionality of the images, sample size, MRI details, and pre-processing methods used. Table A2 delves into the specifications of the employed DL models, highlighting the brain tumor classification task, data partitioning, architecture, and the reported performance metrics. These tables contribute to a comprehensive understanding of the methodologies employed in the re- Cancers 2024, 16, 300 12 of 55 viewed literature. Tables A1 and A2 exclusively display the information available from the original authors in the analyzed papers. Any omissions in the table reflect the absence of such details as provided by the original authors in the surveyed papers. Notice that several papers are marked with an asterisk (*), which denotes that not all models have been reported in our tables due to the extensive array of results reported by the authors. Especially in these cases, we recommend readers refer to the original papers for a compre- hensive overview of findings. In this review, we focus on the differentiation of primary brain tumor types, with particular attention to gliomas due to their aggressive nature. Among the 146 examined papers in this section, some address multiple tasks concurrently. Specifically, 77 focused on distinguishing primary brain tumor types, while 27 aimed to identify tumorous images from images of healthy patients. Furthermore, the pursuit of accurate glioma grade classification is assessed in 66 papers, with 41 of them focusing on the binary distinction between low (grades II and III) and high-grade (glioblastoma, grade IV) gliomas. Note that the question asked by these 41 works does not correspond to any of the canonical releases of the WHO classification of brain tumors, as III is, in fact, high-grade. In this sense, such a grouping would facilitate the achievement of good performance results by grouping entities that are more prone not to show contrast enhancement, in contrast to glioblastoma, which always will show contrast enhancement [72]. Additionally, 12 studies delve into the distinction of glioma subtypes. As previously highlighted, pre-processing techniques are pivotal in medical image pro- cessing. Among the 146 papers analyzed in this review, a substantial 80% of them provide insights into the specific pre-processing methodologies that were employed. Within this subset, it was observed that 35% employed registration techniques involving registration to a common anatomical template and co-registration to the same MRI modality. Furthermore, 40% employed segmentation as a critical step to isolate the brain from the surrounding skull structures. Notably, nearly half of the papers embraced normalization techniques to standardize the intensity of the image data before it was fed to the models. Additionally, 30% of the papers undertook the task of brain tumor extraction through methods such as bounding box delineation or tumor segmen- tation. Moreover, 15% of the papers employed pre-processing integrated image enhancement techniques to improve the contrast and visibility of crucial anatomical structures. In several studies [73–75], researchers investigated the advantages of utilizing the tumor area as opposed to the entire image, highlighting the significant benefits of concentrating on the tumor region rather than the entire image. In the realm of medical research, the size and diversity of the training data sample stand as fundamental factors that substantially influence the performance, generalizability, and robustness of ML models. Several studies have explored the impact of varying the size of the training data sample on model performance [76–84]. Their findings highlight the value of ensuring that a substantial volume of data is available for training, as it significantly contributes to the model’s ability to make more accurate and reliable predictions. Regarding addressing the limitation of data scarcity, approximately 60% of the ex- amined studies employed DA techniques, and 40% incorporated TL in a 2D domain as a viable solution. Several of these investigations [85–92] have demonstrated the advantages of increasing both the quantity and variability of the samples through the inclusion of augmented images. Applying traditional DA techniques, such as geometric variations from original images, was the most widely used strategy, while only a few studies opted for the use of DL generative models [89,93–95]. Several studies [73,92,96–99] have inte- grated DA as an oversampling technique to address the problem of imbalanced data in the context of brain tumor classification. Furthermore, other works have explored the inclusion of multi-view 2D slices from axial, coronal, and sagittal planes, in addition to employing image flipping and rotations to augment the dataset [100]. Pre-trained models have demonstrated performance enhancements in the classification of glioma grades in several studies [79,101,102]. However, it is noteworthy that not all investigations have reported equivalent advantages when employing pre-trained models to discriminate be- Cancers 2024, 16, 300 13 of 55 tween healthy and tumorous samples [103] or to differentiate tumor types [104]. These variations in findings underscore the complexity of the observed performance disparities, which may not be solely ascribed to the classification task itself but may also be influenced by intrinsic dataset variations. The ability of CNNs to automatically extract meaningful features from brain MRI images, as opposed to the conventional need for manual feature engineering in certain ML algorithms like RF, GrB, and SVM, has been emphasized by numerous studies. These studies underscore the potential of CNNs in revolutionizing the landscape of MRI feature extraction for enhanced accuracy and efficiency in brain tumor classification [105–111]. Most of the reviewed papers (approximately 60%) utilized established state-of-the-art CNN architectures to obtain brain tumor classification. Among these, ResNet and VGGNet backbones were the most prevalent choices, closely followed by AlexNet, GoogLeNet, and Inception. In contrast, the remaining 40% of the papers concentrated on enhancing brain tu- mor classification by introducing novel model architectures. The inherent black-box nature of CNNs highlights the importance of delving into the comprehension of their predictions, especially in a medical context. Several studies within our review [74,112–117] have ap- plied post-processing explainability tools to validate that the network’s decision-making process aligns with the intended diagnostic criteria, therefore enhancing the reliability of CNN-based medical applications. Additionally, selected studies [57,118,119] explored the synergies of ensemble learning by combining the outputs of radiomics and DL models. Another interesting area of research has considered the opportunity of incorporating ML classifiers as the final layer in CNNs, effectively bypassing the traditional SoftMax layer [76,96,99,103,119–128]. The integration of information from various data sources has garnered growing interest in the medical field. Brain tumors, due to their distinct features both at the histopathological and radiological level, have motivated numerous studies to explore the synergy between whole slide imaging (WSI) and MRI data [97,129–132]. These investigations consistently highlight the richer information content in WSI as compared to MRI. However, they also re- veal that combining data from both sources leads to improved overall performance in brain tumor characterization. Ensemble learning methods have shown promise in not only inte- grating information from diverse data types but also in combining predictions from multiple DL models on MRI to improve overall performance [75,82,91,107,108,116,122,133–139]. As brain tumor diagnosis and prognosis are significantly linked to genetic factors, several stud- ies have undertaken efforts to explore the capabilities of DL models in extracting meaningful MRI features for the classification of these genetic frameworks [56,83,93,100,117,140,141]. Although brain MRIs inherently capture 3D data, a notable observation is that over 80% of the studies conducted their analyses within a 2D domain, focusing on 2D MRI slices. Nonetheless, some investigations have actively explored the significance of incorporating 3D volumetric information into the realm of brain tumor classification [56,58,59,74,97,98, 100,112,117,129–132,135,140–148]. Although 3D volumes inherently capture information from the three anatomical planes, 2D slices are restricted to a specific view. Notably, among the studies that adopted a 2D approach, only 44% provided details about the chosen anatomical plane. Among this subset, more than 50% utilized axial, coronal, and sagittal views, while over 40% exclusively employed axial views. Similarly, close to 70% of the reviewed studies disclosed the MRI modalities utilized for the analysis. Among these, close to 50% exclusively employed the T1c sequence, while 26% used a combination of T1c, T1, T2, and FLAIR sequences, 12% used three sequences, and the rest chose one sequence. Various strategies were employed to integrate information from multiple modalities. The prevalent method involved fusing them as input channels, comparable to the treatment of channels in RGB images. In their study, Ge et al. [100] evaluated the sensitivity of T1c, T2, and FLAIR modalities in glioma grade classification. Their investigation highlighted the T1c sequence as the most informative among these modalities. To further enhance the classification performance, they incorporated informa- tion from each source using an aggregation layer within the network architecture. Subse- Cancers 2024, 16, 300 14 of 55 quently, similar ensemble learning approaches were adopted by Gutta et al. [106], Hussain et al. [148], Rui et al. [149]. Notably, Guo et al. [150] directly compared the performance of a modality-fusion approach, where the four MRI modalities were concatenated as a four-channel input, with a decision-fusion approach, where final predictions were derived through a linear weighted sum from the probabilities obtained through four independent pre-trained unimodal models. This study reinforced the notion of the T1c modality’s significance in glioma subtype classification. Moreover, it revealed that any multimodal ap- proach consistently outperformed unimodal models, with the decision-ensemble approach emerging as the most effective strategy. As previously discussed, decomposing 3D volumes into individual 2D slices may introduce the potential for data leakage. Maintaining the reliability of the analysis is crucial for obtaining robust and trustworthy findings. It is worth noting, however, that only a limited number of studies that use multiple 2D slices [56,76,77,79,93,100,101,106, 107,115,118,126,135,149,151–157], explicitly detailed their approach to data splitting at the patient level, addressing this critical concern. Remarkably, an insightful comparison was carried out in the work of Badža and Barjaktarović [158] between data-splitting strategies at the patient and image levels. The findings elucidate that an image-wise approach yields accuracy results as high as 96% for brain tumor type classification, while a patient-level split demonstrates a higher degree of reliability with an accuracy of 88%. These results underscore the critical importance of utilizing a patient-wise training approach to assess the model’s generalization capacity. Similarly, Ghassemi et al. [85], Ismael et al. [159] also provided evidence of superior performance when using an image-wise split, further reinforcing the importance of thoughtful data splitting. It is also important to note that 3D models operate on complete 3D volumes and are inherently structured at the patient level. This approach substantially reduces the likelihood of data leakage, therefore enhancing the reliability of the analysis and ensuring that the results faithfully represent the model’s performance. This aspect may provide a valuable perspective when interpreting differences in accuracy between 3D and 2D models. The predominant approach for data partitioning in the examined papers involves the use of hold-out validation with training and validation sets. This was followed by the adoption of K-fold cross-validation, which enhances the robustness of model evaluation. A less frequently employed method was the three-way split, which includes training, validation, and testing sets. In total, only 36% of the studies assessed their final results using an independent test set. Decuyper et al. [140], Gilanie et al. [152], Alanazi et al. [160] took a step further by assessing the generalizability and robustness of their models using external validation sets. Although the authors of the Figshare dataset thoughtfully included a 5-fold CV setup alongside the data to promote comparability and reproducibility, it is still important to remark here that a substantial majority of studies continue to prefer custom data partitioning methods. 5. Machine Learning Applications to Ultra-Low Field Imaging A completely different area of application of ML to neuroradiology has recently emerged with the availability of ultra-low field magnetic resonance imaging devices for point-of-care applications, typically with <0.1 T permanent magnets [161–163]. In the 0.055 T implementation described by Liu et al. [11], DL was used to improve the quality of the acquisition by detecting and canceling external electromagnetic interference (EMI) signals, eliminating the need for radio-frequency shielded rooms. They compared the results of the DL EMI cancelation in 13 patients with brain tumors, both in the 0.055 T and in another 3 T machine, on same-day acquisitions, finding that it was possible to identify the different tumor types. Please note that these processes are, in fact, a completely different use of DL for data pre-processing to those reviewed in Section 3. Another example is the Hyperfine system, which received FDA clearance in 2020 for brain imaging and in 2021 (K212456) for DL-image reconstruction to enhance the quality of the generated images. In particular, DL is used as part of the image reconstruction pipeline of T1, Cancers 2024, 16, 300 15 of 55 T2, and FLAIR images. There are two DL steps: the first one is a so-called DL gridding, where the undersampled k-space data are transformed into images not by Fourier transformations but with DL. The transformed images are then combined, and a final post-processing DL step is applied to eliminate noise. However, no details about the specific algorithms are provided. Although the main application seems to be in the neurocritical setting [164], this system is beginning to be compared with the imaging quality at higher fields at different stages, with a particular interest in the early post-operative monitoring after surgical resection (e.g., [165,166]). It is to be expected that Hyperfine brain tumor applications will emerge soon, for example, through the partnership with The Brain Tumor Foundation, to provide the general population with free brain scans. 6. Conclusions Neuro-oncological radiology relies on non-invasive data acquisition, which makes it the ideal target of data-centered analytics and places it at the forefront of ML-applied developments. In this review paper, we have focused on the most successful instantiation of ML currently, namely DL, and its use for the analysis of imaging data. Emphasis has been put on the fact that DL methods must be seen as only part of analytical pipelines, in which data pre-processing plays a key role. Promoting the responsible utilization of clinical data is of utmost importance when striving to establish trustworthy conclusions. A fundamental step in this endeavor is the comprehensive disclosure of both the data used and the analytical procedures undertaken. Such transparency not only fosters greater trust in research outcomes but also amplifies the generalizability and reproducibility of the findings. This, in turn, plays a pivotal role in advancing AI-driven solutions in the clinical pathway. Most DL-based analytical solutions depend, to a great extent, not only on data quality but also quantity. For this reason, we argue that the main challenge facing the use of DL in the radiological imaging setting is precisely the creation of sizeable curated image databases for the different problems at hand. Author Contributions: All authors (C.P., G.U., M.J.-S. and A.V.) contributed to the different phases of article development, including conceptualization, methodology, investigation, and writing. M.J.-S. and A.V. are responsible for supervision and funding acquisition. All authors have read and agreed to the published version of the manuscript. Funding: This research was funded by H2020-EU.1.3.—EXCELLENT SCIENCE—Marie Skłodowska- Curie Actions, grant number H2020-MSCA-ITN-2018-813120; Proyectos de investigación en salud 2020, grant number PI20/00064. PID2019-104551RB-I00; Centro de Investigación Biomédica en Red en Bioingeniería, Biomateriales y Nanomedicina (CIBER-BBN (http://www.ciber-bbn.es/en, accessed on 3 November 2023), CB06/01/0010), an initiative of the Instituto de Salud Carlos III (Spain) co-funded by EU Fondo Europeo de Desarrollo Regional (FEDER); Spanish Agencia Española de Investigación (AEI) PID2022-143299OB-I00 grant; XartecSalut 2021-XARDI-00021. Carla Pitarch is a fellow of Eurecat’s “Vicente López” Ph.D. grant program. Institutional Review Board Statement: Not applicable. Informed Consent Statement: Not applicable. Data Availability Statement: Data sharing is not applicable. Conflicts of Interest: The authors declare no conflict of interest. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript; or in the decision to publish the results. Cancers 2024, 16, 300 16 of 55 Abbreviations The following abbreviations are used in this manuscript: AUC Area Under the ROC Curve CDSS Clinical Decision Support System CNN Convolutional Neural Network DA Data Augmentation DL Deep Learning FAIR Findable, Accessible, Interoperable, Reusable GAN Generative Adversarial Network HGG High-Grade Glioma ICA Independent Component Analysis LGG Low-Grade Glioma ML Machine Learning MRI Magnetic Resonance Imaging MRS Magnetic Resonance Spectroscopy MRSI Magnetic Resonance Spectroscopy Imaging NMF Nonnegative Matrix Factorization RF Random Forest RGB Red, Green, Blue ROI Region of Interest SVM Support Vector Machine TCGA The Cancer Genome Atlas TCIA The Cancer Imaging Archive tICA Temporal ICA TL Transfer Learning TPOT Tree-based Pipeline Optimization Tool ULF Ultra-Low Field ViT Vision Transformer WHO World Health Organization WSI Whole Slide Imaging Appendix A. Literature Review Summary of Deep-Learning Sources This appendix comprises Tables A1 and A2, offering a detailed overview of the datasets and models employed in the papers summarized in Section 4.3. Table A1 provides key details about the datasets used, including image dimensionality, number of patients in the study and images included for the analysis, anatomical plane, MRI modalities, pre-processing procedures, and data-augmentation techniques. On the other hand, Table A2 summarizes analytical aspects, such as the classification task, data-splitting methodology, network architecture, overall performance metrics, and class-specific performance. Hyphens (-) in certain cells indicate that the information was not provided in the original paper. For papers marked with an asterisk (*), not all outcomes are included in the table due to the extensive range of results reported by the authors. We encourage readers to consult the original papers for a more comprehensive understanding of the findings. Cancers 2024, 16, 300 17 of 55 Table A1. Data Overview: Comprehensive overview of the datasets examined within the DL literature review centered on brain tumor classification tasks and MRI data. Essential information regarding dimensionality, sample size, anatomical plane, MRI modalities, and pre-processing methods are summarized. Sample Size Data Augmentation No. Reference Dim. Dataset Plane MRI Modality Pre-Processing Patients Images (Augmentation Factor) Ge et al. [100] Tumor mask Multi-view images (ax, sag, 1 2D BraTS2017 285 (Table 1) - Ax, Sag, Cor T1c, T2, Flair (2018) enhancement cor), rotation, flipping None 1 2 Ge et al. [73] (2018) 3D BraTS2017 285 (Table 1) 285 (Table 1) Ax, Sag, Cor T1c Tumor mask (LGG: 2) Flipping enhancement 2 BFC, z-score Sagittal flipping, rotation, Pereira et al. 3 3D BraTS2017 285 (Table 1) 285 (Table 1) Ax, Sag, Cor T1, T1c, T2, Flair normalization exponential intensity [74] (2018) (inside brain mask) transformation Z-score (14) HE, random rotation, Yang et al. 113 (LGG: 52, HGG: 867 (LGG: 368, HGG: 4 2D Private Ax T1c normalization, zooming, adding noise, [101] (2018) 61) 499) tumor ROI flipping Abd-Ellah 1800 (H: 600, LGG: 5 et al. [167] 2D Brats2017 - - - - - 600, HGG:600) (2019) IXI 600 Ax T1 REMBRANDT 130 - T1c 16,000 (H: 8000, G.II: 2000, Anaraki et al. [168] TCGA-GBM 199 - T1c Normalization, Rotation, translating, 6 2D G.III: 2000, G.IV: 4000) (2019) TCGA-LGG 299 - T1c resizing scaling, flipping Private 60 - T1c Figshare 233 989 Ax T1c Deepak and Min-max 7 Ameer [76] 2D Figshare 233 (Table 1) 3064 (Table 1) Ax, Sag, Cor T1c normalization, Rotation, flipping (2019) resizing Hemanth 8 et al. [169] 2D Private - 220 - T1, T2, Flair None None (2019) Kutlu and 9 Avcı [120] 2D Figshare 233 (Table 1) 3064 Ax, Sag, Cor T1c None None (2019) Lo et al. [102] 134 (G.II: 30, G.III: 134 (G.II: 30, G.III: Normalization, CE, 10 2D TCIA Ax T1c (56) AutoAugment [170] (2019) 43, G.IV: 57) 43, G.IV: 57) tumor segmentation 557 (G.I: 130, G.II: Muneer et al. Skull-stripping, 11 2D Private 20 169, G.III: 103, G.IV: Ax T2 Resize, reflection, rotation [171] (2019) tumor segmentation 155) IXI 600 - - - - - REMBRANDT 130 - - - - - 12 Rajini [172] (2019) 2D TCGA-GBM “around 200” - - - - - TCGA-LGG 299 - - - - - Figshare 233 - - T1c - - Cancers 2024, 16, 300 18 of 55 Table A1. Cont. Sample Size Data Augmentation No. Reference Dim. Dataset Plane MRI Modality Pre-Processing Patients Images (Augmentation Factor) Rahmathunneesa 760 (G.I: 198, G.II: Skull-stripping, Resizing, rotation, translation, 13 and Muneer 2D Private - 205, G.III: 172, G.IV: Axial T2 resizing reflection [173] (2019) 185) 121 (G.I: 36, G.II: 32, Radiopaedia - - - G.III: 25, G.IV: 28) (30) Rotation, flipping, skewness, Sajjad et al. [174] BFC, Segmentation, 14 2D shears, gaussian blur, sharpening, (2019) Figshare 233 3064 (Table 1) - T1c Z-score normalization edge detection, emboss Sultan et al. [175] Figshare 233 (Table 1) 3064 (Table 1) (5) Rotation, flipping, 15 2D Ax, Sag, Cor T1c Resizing (2019) 73 (G.II: 33, G.III: 19, 516 (G.II: 205, G.III: mirroring, noise REMBRANDT G.IV: 21) 129, G.IV: 182) Min-max Swati et al. 16 2D Figshare 233 (Table 1) 3064 (Table 1) Ax, Sag, Cor T1c normalization, - [77] (2019) resizing Rotation, flipping, Toğaçar et al. 17 2D Kaggle-III - 253 (Table 1) - - - brightening, CE, shifting, [176] (2019) scaling BraTS2012 1 25 (LGG: 5, HGG: 10) - Amin et al. [177] BraTS2013 2 30 (Table 1) - Noise Removal, 18 (2020) 2D BraTS2013 (LB) 3 25 (LGG: 4, HGG: 21) - - T1, T1c, T2, Flair tumor enhancement, - BraTS2015 4 274 (Table 1) - MRI modality fusion BraTS2018 5 284 (Table 1) - Afshar et al. 19 2D Figshare 233 3064 - - - - [178] (2020) Badža and Normalization, 20 Barjaktarović 2D Figshare 233 3064 (Table 1) Ax, Sag, Cor T1c Rotation, flipping Resizing [158] (2020) TCGA-GBM 262 Banerjee et al. [56] 21 TCGA-LGG 199 1590 (LGG:750, Ax 1 Ax, Sag, Cor 2 T1, T1c, T2, Flair - Rotation, shifts, flipping (2020) 2D BraTS2017 285 (Table 1) HGG:840) Bhanothu 2406 (MN: 694, GL: Min-max 22 et al. [179] 2D Figshare 233 - T1c - 805, PT: 907) normalization (2020) Çinar and 23 Yildirim [180] 2D Kaggle-III - 253 (Table 1) Ax, Sag, Cor - - - (2020) Cancers 2024, 16, 300 19 of 55 Table A1. Cont. Sample Size Data Augmentation No. Reference Dim. Dataset Plane MRI Modality Pre-Processing Patients Images (Augmentation Factor) Ge et al. [93] 24 2D BraTS2017 285 (Table 1) - Ax, Sag, Cor T1, T1c, T2, Flair - GAN (2020) Ghassemi Normalization 25 et al. [85] 2D Figshare 233 3064 (Table 1) Ax, Sag, Cor T1c Rotation, flipping (−1,1) (2020) Rotation, flipping, shifting, Ismael et al. 26 2D Figshare 233 (Table 1) 3064 (Table 1) - T1c Resizing, cropping zooming, ZCA whitening, [159] (2020) shearing, brightening Khan et al. 27 2D Kaggle-III - 253 (Table 1) - - Brain cropping Flipping, rotation, brightness [181] (2020) Ma and Jia Z-score Cropping, rotation, zooming, 28 3D CPM-RadPath2019 329 (Table 1) 329 Ax, Sag, Cor T1, T1c, T2, Flair [129] (2020) normalization translation, color changes Mohammed 1258 (H: 286, MN: Rotation, scaling, reflection, 29 and Al-Ani 2D Radiopaedia 60 (15 per class) 380, E: 311, Med: Ax, Sag, Cor - Resizing, denoising translating, cropping [182] (2020) 281) Min-max Mzoughi et al. 284 (LGG: 75, HGG: 30 3D BraTS2018 285 Ax, Sag, Cor T1c normalization, CE, flipping [142] (2020) 209) resizing Cropping, Naser and 108 (G.II: 50, G.III: 815 (G.II: 400, G.III: normalization Rotation, zooming, 31 Deen [183] 2D TCGA-LGG - T1, T1c, Flair 58) 415) (−1,1), resizing, shifting, flipping (2020) segmentation Noreen et al. 32 2D Figshare 233 3064 (Table 1) T1c Normalization - [184] (2020) Noise reduction, Pei et al. [143] z-score 33 3D CPM-RadPath2020 270 (Table 1) 270 Ax, Sag, Cor T1, T1c, T2, Flair Rotation, scaling (2020) normalization, tumor segmentation Rehman et al. 34 2D Figshare 233 3064 (Table 1) T1c CE Rotation, flipping [104] (2020) Saxena et al. Brain cropping, 35 2D Kaggle-III - 253 (Table 1) - - (20) not specified [185] (2020) resizing 1 30 (Table 1) BraTS 2013 36 Sharif et al. 2D BraTS2015 2 274 (Table 1) - - T1, T1c, T2, Flair CE, tumor segmentation - [186] (2020) BraTS2017 3 285 (Table 1) BraTS2018 4 284 (Table 1) Cancers 2024, 16, 300 20 of 55 Table A1. Cont. Sample Size Data Augmentation No. Reference Dim. Dataset Plane MRI Modality Pre-Processing Patients Images (Augmentation Factor) 2132 (H: 1041, T: 1091) 2156 (H: 1041, LGG: 484, HGG: 631) 2156 (H: 1041, AS: 557, OG: 219, GB: 339) Tandel et al. 1115 (AS-II: 356, 37 2D REMBRANDT 112 (Table 1) Ax, Sag, Cor T2 Skull-stripping Rotation, scaling [105] (2020) AS-III: 201, OG-II: 128, OG-III: 91, GB: 339) 2156 (H: 1041, AS-II: 356, AS-III: 201, OG-II: 128, OG-III: 91, GB: 339) Toğaçar et al. 38 2D Kaggle-III - 253 (Table 1) - - - Oversampling [96] (2020) Vimal Kurup 39 et al. [187] 2D Figshare 233 3064 (Table 1) - T1c Resizing Rotation, cropping (2020) BraTS2018 284 (Table 1) 284 (Table 1) Inhomogeneity 2D Ax, Sag, Cor T1c, T2, Flair (23) - AutoAugment [170] TCGA-LGG 30 30 correction, z-score Zhuge et al. [58] 40 normalization, min-max (2020) normalization, tumor BraTS2018 284 (Table 1) 284 (Table 1) 3D Ax, Sag, Cor T1c, T2, Flair segmentation Rotation, scaling, flipping TCGA-LGG 30 30 Alaraimi et al. HE, z-score Rotation, cropping, flipping, 41 2D Figshare 233 3064 (Table 1) - - [78] (2021) normalization scaling, translation Figshare 1 233 (Table 1) 3064 (Table 1) Ax, Sag, Cor T1c - 121 (MN G.I: 36, GL Ayadi et al. 2 (17) - Rotation, flipping, 42 2D Radiopaedia - G.II: 32, GL G.III: 25, - - - [86] (2021) gaussian blur, sharpen GB: 28) 112 (AS-II: 30, AS-III: REMBRANDT 3 17, OG-II: 14, OG-III: - - - - 7, GB: 44) IXI 582 (healthy) - Bashir-Gonbadi Flipping, mirroring, BraTS2017 285 - Skull-stripping, 43 and Khotanlou [188] 2D - - shifting, scaling, Figshare - 3064 resizing (2021) rotation Private - 230 BraTS 2018 43 LGG 43 Co-registration, BraTS2019 335 (Table 1) 335 resampling, Chakrabarty et al. LGG-1p19q 145 159 44 3D Ax, Sag, Cor T1c skull-stripping, - [144] (2021) 1234 (MET: 710, MN: z-score normalization, Private 143, AN: 158, PA: 82, 1234 resizing H: 141) Cancers 2024, 16, 300 21 of 55 Table A1. Cont. Sample Size Data Augmentation No. Reference Dim. Dataset Plane MRI Modality Pre-Processing Patients Images (Augmentation Factor) TCGA 285 (LGG: 121, HGG: 164) 285 Rotation, Flipping, Decuyper et al. TCGA-1p19q 141 141 45 3D Ax, Sag, Cor T1, T1c, T2, Flair Tumor segmentation Intensity scaling, [140] (2021) BraTS2019 202 202 Elastic transform GUH dataset 110 110 Díaz-Pernas Z-score 46 et al. [151] 2D Figshare 233 3064 (Table 1) Ax, Sag, Cor T1c (2) Elastic transforms normalization (2021) (12) PGGAN 1 Gab Allah et al. 47 2D Figshare 233 3064 (Table 1) Ax, Sag, Cor T1c Normalization (−1,1) (9) Rotation, mirroring, [94] (2021) flipping 2 180 (AS-1: 50, AS-II: 30,240 (AS-1: 8400, Gilanie et al. BFC, normalization, 48 2D Private 40, AS-III: 40, AS-IV: AS-II: 6720, AS-III: T1 & Flair: Ax, T2: Ax, Sag T1, T2, Flair Rotation [152] (2021) tumor Segmentation 50) 6720, AS-IV: 8400) Gu et al. [189] REMBRANDT 1 130 110,020 - - - - 49 2D (2021) Figshare 2 - 3064 (Table 1) - T1c - - CE, tumor ROI, Guan et al. 50 2D Figshare 233 (Table 1) 3064 (Table 1) Ax, Sag, Cor T1c min-max (3) Rotation, flipping [153] (2021) normalization 1425 (LGG: 375, BraTS2018 1 - - T1, T1c, T2, Flair HGG: 1050) Grayscaling, Gull et al. [154] 2 1675 (LGG: 380, 51 2D BraTS2019 - - T1, T1c, T2, Flair median filtering, - (2021) HGG: 1295) skull-stripping 2470 (LGG: 645, BraTS2020 3 - HGG: 1435, - T1, T1c, T2, Flair unknown: 390) Resampling, Gutta et al. 237 (G.I: 17, G.II: 59, 660 (G.I: 27, G.II: 144, co-registration, 52 2D Private - T1, T1c, T2, Flair - [106] (2021) G.III: 46, G.IV: 115) G.III: 184, G.IV: 305) skull-stripping, tumor segmentation Hao et al. [79] 6700 (20 random 53 2D BraTS2019 335 (Table 1) Ax, Sag, Cor T1c, T1, T2 - - (2021) slices per patient) (total) 2990 (H: 1350, T: 1640) 3950 RIDER 19 (G.IV) (H: 850, MN: 700, T1c, Flair Irmak [190] GL: 950, PT: 700, 54 2D - - - (2021) REMBRANDT 130 MT: 750) 4570 T1c, Flair TCGA-LGG 199 (G.II: 1676, G.III: 1218, T1c, Flair Figshare 233 G.IV: 1676) T1c Cancers 2024, 16, 300 22 of 55 Table A1. Cont. Sample Size Data Augmentation No. Reference Dim. Dataset Plane MRI Modality Pre-Processing Patients Images (Augmentation Factor) BraTS2012 - 1000 - - Noise removal, Kader et al. [191] BraTS2013 - 1000 - - 55 2D tumor segmentation, - (2021) BraTS2014 - 800 - - resizing BraTS2015 - 700 - - Kader et al. 56 2D Private - 17,600 - T1, T2, Flair - Yes, not specified [192] (2021) Kakarla et al. Resizing, min-max 57 2D Figshare 233 3064 (Table 1) - T1c - [193] (2021) normalization, CE 1 - 253 (Table 1) - - Kaggle-III Kang et al. [133] 58 (2021) 2D Kaggle-II 2 - 3264 (Table 1) - - Brain cropping, resizing Rotation, flipping Kaggle-I 3 - 3000 (Table 1) - - Z-score (20) Rotation, zooming, Khan et al. 59 2D BraTS2015 274 169,880 Ax, Sag, Cor T1, T1c, T2, Flair normalization, geometric transforms, [87] (2021) tumor segmentation sharpening, noise addition, CE Kumar et al. 60 2D Figshare 233 3064 (Table 1) T1c - Rotation [88] (2021) Masood et al. Figshare 233 3064 (Table 1) - T1c 61 2D BFC, CE, tumor ROI - [194] (2021) Kaggle-III - 253 (Table 1) - - Noreen et al. Min-max 62 2D Figshare 233 3064 (Table 1) - - - [134] (2021) normalization Multiple-cropping, (20) Rotation, zooming, Özcan et al. 104 (G.II: 50, G.IV: 63 2D Private 518 Ax, Sag, Cor Flair z-score shearing, flipping, elastic [155] (2021) 54) normalization gaussian transforms Pei et al. [97] BFC, z-score 64 3D CPM-RadPath2020 256 (Table 1) 256 (Table 1) Ax, Sag, Cor T1, T1c, T2, Flair (oversampling) (2021) normalization Sadad et al. 65 2D Figshare 233 (Table 1) 3064 (Table 1) Ax, Sag, Cor - CE, tumor detection Rotation, flipping [195] (2021) 2156 (H: 1041, T: 1091) 557 (AS-II: 356, Tandel et al. AS-III: 201) 66 [107] (2021) 2D REMBRANDT 130 (H: 18, T: 112) Ax, Sag, Cor T2 - Rotation, scaling 219 (OG-II: 218, OG-III: 91) 1115 (LGG: 484, HGG: 631) Toğaçar et al. Rotation, scrolling, 67 2D Figshare 233 3064 (Table 1) - T1c - [80] (2021) brightening Cancers 2024, 16, 300 23 of 55 Table A1. Cont. Sample Size Data Augmentation No. Reference Dim. Dataset Plane MRI Modality Pre-Processing Patients Images (Augmentation Factor) Yamashiro 68 et al. [145] 3D BraTS2018 284 (Table 1) 285 (Table 1) Ax, Sag, Cor T1c Tumor segmentation Flipping, scaling, shifting (2021) Tumor segmentation, Yin et al. [130] Brightness, CE, saturation, 69 3D CPM-RadPath2020 256 (Table 1) 256 (Table 1) Ax, Sag, Cor T1, T1c, T2, Flair resizing, z-score (2021) hue, flipping, rotation normalization CE, min-max Aamir et al. 70 2D Figshare 233 (Table 1) 3064 (Table 1) Ax, Sag, Cor T1c normalization, (2) Rotation, flipping [156] (2022) tumor ROI Ahmad et al. Resizing, CDA: Rotation, scaling 71 2D Figshare 233 3064 (Table 1) - T1c [89] (2022) normalization GDA: VAE, GAN 3000 (H: 1500, T: Kaggle-I - Noise removal, Alanazi et al. 1500) 72 2D - - cropping, z-score - [160] (2022) Kaggle-II 1 - 3264 (Table 1) normalization, resizing Figshare 2 233 3064 (Table 1) 2870 (H: 395, MN: Almalki et al. Kaggle-II 1 - - - Brain cropping, - 73 2D 822, GL: 826, PT: 827) [121] (2022) denoising, resizing Figshare 2 233 3064 (Table 1) - - - Min-max Amou et al. 74 2D Figshare 233 3064 (Table 1) Ax, Sag, Cor T1c normalization, None [81] (2022) resizing Figshare 233 3064 (Table 1) - - Rotation, flipping, Aurna et al. [82] Kaggle-II - 3264 (Table 1) - - 75 2D Resizing zooming, shifting, (2022) 4292 (H: 681, MN: scaling Kaggle [196] - 1318, GL: 1038, PT: - - 1255) 332 (LGG: 73, BraTS2019 332 Skull-stripping, Chatterjee et al. HGG: 259) 2D: Ax, 76 2D-3D T1c normalization (0.5,99.5), Affine, flipping [59] (2022) 3D: Ax, Sag, Cor IXI 259 259 resampling Chitnis et al. 77 2D Kaggle-II - 3264 (Table 1) - - Resizing Autoaugment [197] (2022) Histogram & Rotation, deformations, BraTS2018 284 (Table 1) Coupet et al. min-max normalization shearing, zooming, 78 2D-3D 50,812 Ax T1, T1c, T2, Flair [135] (2022) BraTS2020 369 (Table 1) flipping Segmentation, Dang et al. gamma correction, 79 3D BraTS2019 335 (Table 1) 335 (Table 1) Ax, Sag, Cor T1, T1c, T2, Flair (oversampling) Rotation [98] (2022) window setting optimization Cancers 2024, 16, 300 24 of 55 Table A1. Cont. Sample Size Data Augmentation No. Reference Dim. Dataset Plane MRI Modality Pre-Processing Patients Images (Augmentation Factor) Z-score 3D 707 (G.I: 189, 707 Ax, Sag, Cor T1c normalization, - Danilov et al. 80 Private G.II: 133, G.III: 127, resampling [146] (2022) G.IV: 258) ImageNet Rotation, scaling, 2D 17,730 Ax, Sag, Cor T1c standardization mirroring 101 (LGG: 58, HGG: Ding et al. [57] Private Tumor ROI, 81 2D-3D 43) 3 slices as channels Ax, Sag, Cor T1c - (2022) normalization, 50 (LGG: 25, HGG: TCIA + Private resizing 25) BraTS2015 - - - Resizing, normalization, (total) 4000 Shifting, Rotation, Ekong et al. denoising, BFC, 82 2D IXI - (H: 1000, MN: - 10,000, - T1, T2 Brightening, Image [198] (2022) registration, tumor GL: 1000, PT: 1000) enlargement, Flipping segmentation Figshare - - T1c Z-score Gao et al. 83 3D Private 39,210 39,210 Ax, Sag, Cor T1, T2, T1c normalization, - [112] (2022) resampling Gaur et al. 84 2D Kaggle-II - 2870 - - Resizing Gaussian Noise [199] (2022) BFC, skull-stripping, Guo et al. Rotation, resizing, scaling, 85 3D CPM-RadPath2020 221 (Table 1) 221 Ax, Sag, Cor T1, T1c, T2, Flair co-registration, [150] (2022) gaussian noise, CE tumor segmentation Gupta et al. 86 2D Kaggle-II - 3264 (Table 1) - - CE CycleGAN [95] (2022) Gurunathan 260 (LGG: 156, HGG: Resizing, tumor Rotation, shifts, reflection, 87 and Krishnan 2D BraTS - Ax, Sag, Cor T1, T2 104) segmentation flipping, scaling, shearing [200] (2022) Haq et al. [90] 88 2D Figshare 233 (Table 1) 3064 (Table 1) - T1c Resizing (2) Zooming (2022) BraTS2020 369 (Table 1) 369 Sampling patches, Hsu et al. [131] Rotation, flipping, 89 3D Ax, Sag, Cor T1, T1c, T2, Flair z-score normalization, (2022) affine translation CPM-RadPath2020 270 (Table 1) 270 tumor segmentation Isunuri and Resizing, 90 Kakarla [201] 2D Figshare - 3064 (Table 1) - T1c - Normalization (2022) Cancers 2024, 16, 300 25 of 55 Table A1. Cont. Sample Size Data Augmentation No. Reference Dim. Dataset Plane MRI Modality Pre-Processing Patients Images (Augmentation Factor) Jeong et al. 1445 (largest slice ± Resizing, z-score 91 2D BraTS2017 285 (Table 1) Ax, Sag, Cor T1, T1c, T2, Flair Rotation, flipping [113] (2022) 8) normalization 1500 (MN: 1000, GL: Kazemi et al. Figshare 1 233 - T1c 92 2D 800, PT: 600) Resizing - [108] (2022) TCIA 2 20 8798 - T1c Khazaee et al. 26,904 (LGG: 13,671, 93 2D BraTS2019 - - T1c, T2, Flair - Rotation, flipping [202] (2022) HGG: 13,233) Min-max Kibriya et al. (5) Rotation, flipping, 94 2D Figshare 233 (Table 1) 3064 (Table 1) - - normalization, [122] (2022) mirroring, adding noise resizing Koli et al. [203] 2D Kaggle-III - 253 (Table 1) - - - 95 Rotation (2022) Figshare - 3064 (Table 1) - - Lakshmi and 96 Rao [204] 2D Figshare - 3064 - T1c - - (2022) CE, tumor Maqsood et al. 2D Figshare 233 3064 (Table 1) - T1c 97 segmentation, z-score - [114] (2022) - BraTS2018 284 (Table 1) - - - normalization Murthy et al. Median filtering, CE, 98 2D Kaggle-III - 253 (Table 1) - - - [205] (2022) tumor segmentation Noise removal, Nayak et al. 3260 (196 H, 3064 gaussian blurring, (21) Rotation, Shifting, 99 2D Figshare - Ax, Sag, Cor T1c [206] (2022) Table 1) min-max Zooming normalization Rajinikanth 2000 (GL = 1000, GB 100 et al. [124] 2D TCIA - Ax - - - = 1000) (2022) Rasool et al. 101 2D Figshare 233 3064 (Table 1) Ax, Sag, Cor T1c - Yes, not specified [125] (2022) Raza et al. 102 2D Figshare 233 (Table 1) 3064 (Table 1) Ax, Sag, Cor T1c Resizing - [207] (2022) 230 (MN: 81, GL: 90, 3061 (MN: 707, GL: Rizwan et al. Figshare Ax, Sag, Cor T1c 103 2D PT: 59) 1425, PT: 929) Noise, cropping, resizing (5) Salt-noise, grayscaling [208] (2022) 70 (G.II: 32, G.III: 18, 513 (G.II: 204, G.III: REMBRANDT - T1c G.IV: 20) 128, G.IV: 181) Samee et al. 236 (MN: 83, GL: 90, 3075 (MN: 708, GL: (16) Rotation, zooming, 104 2D Figshare Ax, Sag, Cor T1c Grayscaling [209] (2022) PT: 63) 1427, PT: 940) brightening Cancers 2024, 16, 300 26 of 55 Table A1. Cont. Sample Size Data Augmentation No. Reference Dim. Dataset Plane MRI Modality Pre-Processing Patients Images (Augmentation Factor) Resizing, denoising, Samee et al. 65 (LGG: 14, HGG: 1056 (LGG: 176, 105 3D BraTS2015 - T1, T1c, T2, Flair CE, tumor - [147] (2022) 51) HGG: 880) segmentation Sangeetha Min-max 106 et al. [210] 3D Private 45 45 Ax, Sag, Cor T2 (14) Rotation, translation normalization (2022) Saravanan et al. BRATS 274 1200 - - 107 2D Resizing - [109] (2022) REMBRANDT 135 - - - Min-max Sekhar et al. 108 2D Figshare 233 (Table 1) 3064 (Table 1) Ax, Sag, Cor T1c normalization, Yes but not specified [126] (2022) resizing (H: 11, MN: 5, GL:6, PT: 5) Denoising, min-max Senan et al. 3060 (H: 396, MN: Rotation, cutting, zooming, 109 2D Kaggle-II - Ax, Sag, Cor - normalization, [99] (2022) 937, GL: 826, PT: 901) patching, padding, resizing, CE brightening Brain cropping, Scaling, cropping, resizing, Srinivas et al. 256 (Benign: 158, z-score 110 2D Kaggle - - - flipping, rotation, geometric [211] (2022) Malignant: 98) normalization, transforms resizing None1 , Tandel et al. 112 (LGG = 44, HGG 111 2D Rembrandt - Ax T1, T2, Flair Skull-stripping2 , Scaling, rotation [75] (2022) = 68) tumor ROI3 Tripathi and 322 (LGG:159, HGG: 7392 (LGG: 5088, Skull-stripping, Rotation, flipping, scaling, 112 Bag [83] 2D TCIA - T2 163) HGG: 2304) segmentation cropping, translation (2022) BraTS2019 202 Co-registration, Tripathi and Bag TCGA-GBM 158 (total) 617 (LGG: 331, skul-stripping, Flipping, shifting, 113 3D Ax, Sag, Cor T1c, T2, Flair [141] (2022) TCGA-LGG 119 HGG: 286) resampling, rotation, cropping LGG-1p19qdeletion [212] 138 tumor segmentation Tummala 114 et al. [136] 2D Figshare 233 3064 (Table 1) Ax, Sag, Cor T1c - - (2022) Vankdothu Grayscaling, 115 et al. [213] 2D Kaggle-II - 3264 - - rotation, denoising, - (2022) tumor ROI Wang et al. Resizing, brain Rotation, flipping, scaling, 116 3D CPM-RadPath2020 270 270 Ax, Sag, Cor T1, T1c, T2, Flair [132] (2022) cropping jittering Cancers 2024, 16, 300 27 of 55 Table A1. Cont. Sample Size Data Augmentation No. Reference Dim. Dataset Plane MRI Modality Pre-Processing Patients Images (Augmentation Factor) Resampling, skull-stripping, Xiong et al. 211 (AS: 54, OG: 67, z-score 117 2D Private 633 Ax, Sag, Cor ADC, T1c, Flair - [115] (2022) GB: 90) normalization, min-max normalization BFC, skull-stripping, Xu et al. [118] registration, z-score 118 2D BraTS2020 369 (Table 1) 369 Ax T1c, T1, T2 - (2022) normalization 1 tumor ROI 2 Yazdan et al. 119 2D Kaggle-II - 3264 (Table 1) - T1, T2, Flair Denoising None [214] (2022) Zahoor et al. 1 - 1994 (H) - - Rotation, sharing, Kaggle 120 2D Resizing [103] (2022) Figshare 2 3064 Ax, Sag, Cor - scaling, reflection Figshare - 2880 (H: 396, - AlTahhan et al. 121 2D Kaggle-II - MN: 825, GL: 829, - T1c - - [127] (2023) Kaggle-I - PT: 830) - Al-Zoghby 122 et al. [137] 2D Figshare 233 3064 (Table 1) Ax, Sag, Cor T1c Resizing - (2023) Anagun [215] Brain cropping, HE, (9) Flipping, rotation, 123 2D Figshare - 3064 (Table 1) Ax, Sag, Cor T1c (2023) denoising shifting, zooming Anand et al. 124 2D TCGA-LGG 110 3929 - Flair - Flipping [91] (2023) Apostolopoulos Kaggle [217] 26,249 (H: 2000, MN: 7866, 125 2D - - - - - et al. [216] (2023) Kaggle [218] GL: 8208, PT: 8175) Asif et al. 126 2D Figshare 233 (Table 1) 3064 (Table 1) - - Resizing, denoising - [138] (2023) Athisayamani Denoising, 127 et al. [110] 2D Figshare - - - - skull-stripping, Rotation, flipping (2023) brain segmentation BraTS2013 65 (40) Resizing, cropping, Bairagi et al. BraTS2015 - 327 rotation, reflection, 128 2D - T1, T2, Flair Resizing [111] (2023) OPEN-I NLM 229 shear, translation Deepa et al. BraTS2018 1 - - - Min-max normalization, Flipping, translation, 129 2D - [84] (2023) Figshare 2 - 3064 (Table 1) T1c tumor segmentation rotation, brightening, CE, gaussian noise Cancers 2024, 16, 300 28 of 55 Table A1. Cont. Sample Size Data Augmentation No. Reference Dim. Dataset Plane MRI Modality Pre-Processing Patients Images (Augmentation Factor) El-Wahab 130 et al. [219] 2D Figshare 233 3064 (Table 1) Ax, Sag, Cor T1c - - (2023) Hossain et al. (4) Rescaling, shearing, 131 2D Kaggle-II - 3264 - - - [116] (2023) zooming, flipping Hussain et al. T1, T1c, T2, Flair, Denosing, tumor 132 3D BraTS2020 369 (Table 1) 369 (Table 1) Ax, Sag, Cor - [148] (2023) Segmentation segmentation Kibriya et al. 1 - 253 (Table 1) - - - - Kaggle-III 133 2D [119] (2023) Kaggle-I 2 - 3000 (Table 1) - - - - Krishnapriya (Oversampling) Rotation, 134 and Karuna 2D Kaggle-III - 253 (Table 1) - - Brain cropping shifting, rescaling, [92] (2023) mirroring Resizing, Kumar et al. ACRIN-DSC-MR- 135 2D - 1731 - T1 grayscaling, CE, - [128] (2023) BRAIN tumor segmentation Kaggle-II - 3264 (Table 1) Mahmud et al. Normalization, Mirroring, rotation, 136 2D CPTAC-GB 189 - - - [220] (2023) smoothing shifting, zooming ACRIN-FMISO- 45 - BRAIN Muezzinoglu Resizing, patch 137 et al. [221] 2D Kaggle-II - 3264 (Table 1) - - - division (2023) Özkaraca Kaggle [223] total: 7021 (H: 2002, 138 et al. [222] 2D (combines Figshare, - MN: 1627, GL: 1623, - - - - (2023) Kaggle-I, Kaggle-II) PT: 1769) Özkaya and Tumor segmentation, (undersampling 139 Şağıroğlu 2D BraTS2020 369 Ax T1c, T2, Flair min-max - slices HGG) [224] (2023) normalization Rasheed et al. Resizing, 140 2D Figshare 233 3064 (Table 1) Ax, Sag, Cor T1c None [225] (2023) normalization Rui et al. 42 (G.II: 18, G.III: 10, 1176 (G.II: 504, G.III: Brain cropping, 141 2D Private Ax T1c, T2, Flair - [149] (2023) G.IV: 14) 280, G.IV: 392) normalization Shirehjini 1061 (T1c: 229, T1: 58 (G.I: 8, G.II: 16, Resizing, min-max 142 et al. [123] 2D Private 251, T2: 299, Flair: Ax, Sag, Cor T1, T1c, T2, Flair - G.III: 10, G.IV: 22) normalization (2023) 282) Cancers 2024, 16, 300 29 of 55 Table A1. Cont. Sample Size Data Augmentation No. Reference Dim. Dataset Plane MRI Modality Pre-Processing Patients Images (Augmentation Factor) Srinivasan Denoising, tumor 143 et al. [226] 2D REMBRANDT - 3100 - - - segmentation (2023) Tandel et al. 112 (LGG: 44, HGG: 144 2D REMBRANDT - Ax, Sag, Cor T1, T2, Flair Resizing Rotation, scaling [139] (2023) 68) Erasmus MC [227] 816 Haaglanden Medical 279 Center BraTS 168 Registration, resampling, REMBRANDT 109 van der Voort (total) 1412 (G.II: 277, BFC, skull-stripping, (2) Cropping, rotation, 145 3D CPTAC-GBM 51 Ax, Sag, Cor T1, T1c, T2, Flair et al. [117] (2023) G.III:173, G.IV: 962) brain cropping, brightening, CE Ivy GAP 39 z-score normalization Amsterdam UMC 20 Brain-tumor- 20 progression University Medical 6 Center Utrecht TCGA-LGG 107 TCGA-GBM 133 Z-score Wu et al. 326 (LGG:76, HGG: Rotation, translation, 146 2D BraTS2019 slices with tumor - T1, T1c, T2, Flair normalization, [157] (2023) 250) clipping center-cropping AS: Astrocytoma, Ax: Axial, BFC: Bias Field Correction, CE: Contrast Enhancement, Cor: Coronal, DA: Data Augmentation, E: Ependymoma, Flair: Fluid Attenuated Inversion Recovery, GL: Glioma, GAN: Generative Adversarial Network, GB: Glioblastoma, GDA: Generative Data Augmentation, H: Healthy, HE: Histogram Equalization, HGG: High-Grade Glioma, LGG: Low-Grade Glioma, Med: Medulloblastoma, MN: Meningioma, MT: Metastasis, OG: Oligodendroglioma, PT: Pituitary, ROI: Region of Interest, Sag: Sagittal, T: Tumor, T1c: T1 post-contrast weighted, VAE: Variational Auto-Encoder. Numerical superscripts link datasets with models in Table A2 when different data sources yield individual results. Cancers 2024, 16, 300 30 of 55 Table A2. Model Overview: Comprehensive summary of the DL architectures employed across the reviewed papers. The table outlines key information, including the brain tumor classification task, data partitioning, architecture, and the reported performance metrics. Data Split No. Reference Classification Task Architecture Acc% AUC% F1% Class Performance % Method Ratio Level [T1c] CNN 83.73 - - LGG = 82.54, HGG = 84.92 [T2] CNN 69.74 - - LGG = 59.52, HGG = 80.15 1 Ge et al. [100] (2018) LGG vs. HGG Three-way 60:20:20 Patient [Flair] CNN 75.40 - - LGG = 76.19, HGG = 74.60 [Modality-ensemble] CNN 90.87 - - LGG = 90.48, HGG = 91.27 Custom CNN 2 Ge et al. [73] (2018) LGG vs. HGG Three-way 60:20:20 Patient 1 84.21 - - - [whole image] 2 89.47 - - LGG = 90.48, HGG = 86.67 [tumor ROI] Custom CNN ROI: brain, Std.: image 89.50 88.57 86.45 LGG = 80.00, HGG = 92.90 3 Pereira et al. [74] (2018) LGG vs. HGG Three-way 60:20:20 Patient ROI: brain, Std.: brain 89.50 89.13 86.43 LGG = 80.00, HGG = 92.86 ROI: tumor, Std.: image 87.70 88.41 85.08 LGG = 86.67, HGG = 88.10 ROI: tumor, Std.: brain 92.98 98.41 90.96 LGG = 86.67, HGG = 95.24 TL GoogLeNet 94.50 96.80 - - TL AlexNet 92.70 96.60 - - 4 Yang et al. [101] (2018) LGG vs. HGG 5-fold CV, Test 80:20 Patient GoogLeNet 90.90 93.90 - - AlexNet 85.50 89.40 - - Abd-Ellah et al. [167] 5 H vs. LGG vs. HGG Three-way 65:10:25 - Parallel CNNs 97.44 - - (R) 97.00, (S) 98.00 (2019) H vs. G.II vs G.III vs. H = 99.80, G.II = 88.40, G.III = 93.10 - - Anaraki et al. [168] G.IV 86.80, G.IV = 97.40 6 Hold-out 80:20 - Custom CNN + GA (2019) G.II vs. G.III vs. G.IV 90.90 - - - MN = 87.80, GL = 98.30, PT = MN vs. GL vs. PT 94.20 - - 96.5 TL GoogLeNet Deepak and Ameer [76] - KNN 98.00 - - - 7 MN vs. GL vs. PT 5-fold CV Patient (2019) MN = 96.00, GL = 97.90, PT = - SVM 97.80 - 97.00 98.90 - SoftMax 92.30 - - - Hemanth et al. [169] MT = 94.00, MN = 93.00, GL = 8 MT vs. MN vs. GL vs. AS - - - Custom CNN 96.40 - - (2019) 93.00, AS = 89.00 TL AlexNet-DWT Kutlu and Avcı [120] - LSTM 98.66 99.00 - B = 99.33, M = 98.66 9 Benign vs. Malignant 5-fold CV 70:30 - (2019) - SVM 92.09 - - B = 96.04, M = 92.08 - KNN 85.91 - - B = 92.95, M = 85.91 G.II = 96.90, G.III = 96.80, G.IV = TL AlexNet 97.90 99.91 - 10 Lo et al. [102] (2019) G.II vs. G.III vs. G.IV 10-fold CV - 99.10 AlexNet 61.42 82.22 - - Muneer et al. [171] TL VGG19 94.64 - 93.71 - 11 G.I vs. G.II vs. G.III vs. G.IV Hold-out 70:30 - (2019) Wndchrm 92.86 - 92.32 - Cancers 2024, 16, 300 31 of 55 Table A2. Cont. Data Split No. Reference Classification Task Architecture Acc% AUC% F1% Class Performance % Method Ratio Level H vs. G.II vs G.III vs. H = 99.80, G.II = 89.20, G.III = 96.77 95.65 93.54 12 Rajini [172] (2019) G.IV Hold-out 80:20 - Custom CNN 85.27, G.IV = 98.00 MN = 93.69, GL = 99.15, PT = MN vs. GL vs. PT 98.16 97.93 97.21 99.13 G.I = 96.67, G.II = 93.44, G.III = - TL AlexNet 92.98 - 96.06 92.31, G.IV = 89.09 Rahmathunneesa and 13 G.I vs. G.II vs. G.III vs. G.IV Hold-out 70:30 G.I = 86.67, G.II = 98.36, G.III = Muneer [173] (2019) TL GoogLeNet 85.96 - 91.71 63.46, G.IV = 92.73 G.I = 76.67, G.II = 93.44, G.III = TL InceptionV3 86.84 - 91.62 90.38, G.IV = 87.27 G.I = 93.33, G.II = 91.80, G.III = TL ResNet50 96.05 - 97.76 100.00, G.IV = 100.00 TL VGG-19 G.I = 90.03, G.II = 89.91, G.III = w/o DA 87.38 - - 14 Sajjad et al. [174] (2019) G.I vs. G.II vs. G.III vs. G.IV Three-way 50:25:25 - 84.11, G.IV = 85.50 G.I = 95.54, G.II = 92.66, G.III = w/ DA 90.67 - - 87.77, G.IV = 86.71 MN = 90.22, GL = 93.12, PT = w/o DA - - - MN vs. GL vs. PT 89.08 MN = 94.05, GL = 96.14, PT = w/ DA 94.58 - - 93.21 MN = 95.50, GL = 94.40, PT = MN vs. GL vs. PT 96.13 - - 15 Sultan et al. [175] (2019) Hold-out 68:32 - Custom CNN 93.40 G.II = 100, G.III = 95.00, G.IV = G.II vs. G.III vs. G.IV 98.70 - - 100.00 GL = 95.97, MN = 89.98, PT = Block-wise TL VGG19 94.82 - 91.73 96.81 16 Swati et al. [77] (2019) MN vs. GL vs. PT 5-fold CV Patient Block-wise TL VGG16 94.65 - 91.50 (R) 93.51, (S) 94.56 TL AlexNet 89.95 - 86.83 (R) 89.10, (S) 89.84 Custom CNN 96.05 98.00 94.12 H = 96.00, T = 96.08 GoogLeNet 89.66 - 90.32 H = 84.85, T = 96.00 17 Toğaçar et al. [176] (2019) H vs. T Hold-out 70:30 - AlexNet 87.93 - 88.52 H = 84.38, T = 92.31 VGG16 84.48 - 85.25 H = 81.25, T = 88.46 1 - - H = 97.00, T = 97.00 97.00 2 - - H = 99.00, T = 95.00 98.00 18 Amin et al. [177] (2020) H vs. T Hold-out 50:50 - Custom CNN 3 - - H = 100.00, T = 100.00 100.00 4 - - H = 98.00, T = 92.00 96.00 5 - - H = 99.00, T = 93.00 97.00 MN = 75.35, GL = 96.85, PT = 19 Afshar et al. [178] (2020) MN vs. GL vs. PT Hold-out 80:20 Custom CNN 92.45 98.00 - 98.90 Cancers 2024, 16, 300 32 of 55 Table A2. Cont. Data Split No. Reference Classification Task Architecture Acc% AUC% F1% Class Performance % Method Ratio Level Custom CNN MN = 62.70, GL = 90.20, PT = Badža and Barjaktarović [158] Patient w/o DA 84.45 - 81.86 20 MN vs. GL vs. PT 10-fold CV 60:20:20 91.30 (2020) MN = 71.60, GL = 92.80, PT = Patient w/ DA 88.48 - 86.97 95.00 MN = 89.80, GL = 96.20, PT = Image w/o DA 95.40 - 94.93 98.40 MN = 90.20, GL = 98.00, PT = Image w/ DA 96.56 - 96.11 99.20 2 94.74 - - LGG = 94.29, HGG = 96.00 VolumeNet 1 85.96 - - LGG = 80.00, HGG = 88.10 SliceNet 21 Banerjee et al. [56] (2020) LGG vs. HGG Hold-out Patient 1 82.45 - - LGG = 74.67, HGG = 85.24 PatchNet 1 72.30 - - LGG = 72.06, HGG = 71.43 TL ResNet 1 68.07 - - LGG = 69.33, HGG = 67.62 TL VGGNet Bhanothu et al. [179] (P) GL = 75.18, MN = 68.18, PT = 22 MN vs. GL vs. PT Hold-out 80:20 - F-RCNN + VGG16 - - - (2020) 97.28 Custom CNN 97.01 - 96.90 H = 94.70, T = 100.00 ResNet50 92.54 - 93.33 H = 89.74, T = 96.40 Çinar and Yildirim [180] DenseNet201 91.04 - 92.30 H = 85.71, T = 100.00 23 H vs. T - - - (2020) AlexNet 89.55 - 90.05 H = 87.17, T = 92.85 InceptionV3 88.07 - 81.81 H = 81.81, T = 100.00 GoogLeNet 71.64 - 66.03 H = 66.03, T = 92.85 Modality-ensemble Semi-supervised CNN 24 Ge et al. [93] (2020) LGG vs. HGG Hold-out 60:20:20 Patient w/o DA 89.53 - - LGG = 78.26, HGG = 93.65 w/ DA 90.70 - - LGG = 84.35, HGG = 93.01 Custom CNN MN = 79.86, GL = 94.96, PT = Patient w/o pre-training 91.70 - 90.54 25 Ghassemi et al. [85] (2020) MN vs. GL vs. PT 5-fold CV 95.67 MN = 84.82, GL = 94.92, PT = Patient w/ GAN pre-training 93.01 - 92.10 96.92 MN = 89.98, GL = 96.83, PT = Image w/ GAN pre-training 95.60 - 95.10 97.93 MN = 93.00, GL = 99.00, PT = Patient 97.82 - 97.00 26 Ismael et al. [159] (2020) MN vs. GL vs. PT Hold-out 80:20 ResNet50 99.00 MN = 98.00, GL = 99.00, PT = Image 99.34 - 99.00 100.00 Custom CNN 100.00 100.00 100.00 H = 100.00, T = 100.00 VGG16 96.00 96.00 97.00 H = 92.85, T = 100.00 27 Khan et al. [181] (2020) H vs. T Three-way 70:20:10 - ResNet50 89.00 89.00 90.00 H = 85.71, T = 92.86 InceptionV3 75.00 75.00 74.00 H = 76.92, T = 73.33 Cancers 2024, 16, 300 33 of 55 Table A2. Cont. Data Split No. Reference Classification Task Architecture Acc% AUC% F1% Class Performance % Method Ratio Level [WSI] 2D ResNet50 83.33 - 91.40 - 28 Ma and Jia [129] (2020) AS vs. OG vs. GB Three-way 70:10:20 Patient [MRI] 3D DenseNet121 71.10 - 82.90 - [WSI-MRI] Ensemble 2D-3D 88.90 - 94.30 - Mohammed and Al-Ani 29 H vs. EP vs. MN vs. MB Three-way 70:10:20 - Custom CNN 96.00 - - - [182] (2020) Mzoughi et al. [142] 30 LGG vs. HGG - - Patient Custom CNN 96.49 - - - (2020) Naser and Deen [183] 31 G.II vs. G.III 5-fold CV - TL VGG16 95.00 97.00 - G.II = 98.00, G.III = 93.00 (2020) MN = 99.00, GL = 100.00, PT = InceptionV3 99.34 99.00 - 32 Noreen et al. [184] (2020) MN vs. GL vs. PT Hold-out 80:20 - 100.00 MN = 99.00, GL = 100.00, PT = DenseNet201 99.51 100.00 - 99.00 Pei et al. [143] 33 AS vs. OG vs. GB Three-way 67:11:22 Patient Custom CNN 63.90 - - - (2020) AlexNet 97.39 - - - GoogLeNet 98.04 - - - VGG16 98.69 - - - 34 Rehman et al. [104] (2020) MN vs. GL vs. PT Three-way 70:15:15 - TL AlexNet 95.77 - - - TL GoogLeNet 95.44 - - - TL VGG16 89.79 - - - TL ResNet50 95.00 95.00 95.20 - 35 Saxena et al. [185] (2020) H vs. T Three-way 70:20:10 - TL VGG16 90.00 90.00 90.90 - TL InceptionV3 55.00 55.00 68.90 - 70:30 1 - - - 98.30 2 - - - 36 Sharif et al. [186] (2020) * LGG vs. HGG 10-fold CV, Test - Ensemble TL InceptionV3-DRLBP 97.80 3 - - - 96.90 4 - - - 92.50 H vs. T 100.00 - 100.00 100.00 H vs. LGG vs. HGG 2-fold CV, 95.97 - 94.80 94.85 37 Tandel et al. [105] (2020) * H vs. AS vs. OG vs. GB 5-fold CV, - TL AlexNet 96.65 - 94.78 94.17 AS-II vs. AS-III vs. OG-2 10-fold CV 87.14 - 86.89 84.40 vs. OG-3 vs. GB H vs. AS-II vs. AS-III vs. 93.74 - 91.97 91.51 OG-2 vs. OG-3 vs. GB Cancers 2024, 16, 300 34 of 55 Table A2. Cont. Data Split No. Reference Classification Task Architecture Acc% AUC% F1% Class Performance % Method Ratio Level Ensemble TL 96.77 - 96.77 (R) 97.83, (S) 95.74 AlexNet-VGG16-RFE-SVM 38 Toğaçar et al. [96] (2020) * H vs. T Hold-out 70:30 - TL AlexNet 90.32 - 89.89 (R) 95.24, (S) 86.27 TL VGG16 87.10 - 87.23 (R) 87.23, (S) 86.96 Vimal Kurup et al. [187] GL = 96.00, MN = 94.00, PT = 49 MN vs. GL vs. PT Hold-out 80:20 - Custom CNN 92.60 96.33 93.33 (2020) 94.00 TL 2D ResNet50 w/o DA 89.10 - - (R) 86.40, (S) 91.70 40 Zhuge et al. [58] (2020) LGG vs. HGG 5-fold CV, Test 60:20:20 Patient w/ DA 96.30 - - (R) 93.50, (S) 97.20 3D ConvNet 97.10 - - (R) 94.70, (S) 96.80 TL VGG16 100.00 98.60 - - 41 Alaraimi et al. [78] (2021) MN vs. GL vs. PT Hold-out 80:20 - TL GoogLeNet 98.50 98.10 - - TL AlexNet 94.40 97.60 - 1 MN = 89.68, GL = 94.46, PT = MN vs. GL vs. PT 94.74 - 94.19 99.03 2 G.I = 88.23, G.II = 93.33, G.III = G.I vs. G.II vs. G.III vs. 90.35 - 90.38 G.IV 84.00, G.IV = 96.00 3 100.00 - 100.00 H = 100.00, T = 100.00 H vs. T 3 Custom CNN[w/o DA] H = 100.00, LGG = 100.00, HGG H vs. LGG vs. HGG 95.00 - 91.35 = 70.00 3 H = 99.00, AS = 96.36, OG = H vs. AS vs. OG vs. GB 94.41 - 92.89 92.00, GB = 80.00 3 AS-II = 85.71, AS-III = 90.00, AS-II vs. AS-III vs. 86.08 - 86.85 OG-II = 86.66, OG-III = 80.00, GB OG-II vs. OG-III vs. GB = 85.71 3 H = 100.00, AS-II = 85.71, AS-III H vs. AS-II vs. AS-III vs. 42 Ayadi et al. [86] (2021) 5-fold CV 70:30 - 92.09 - 89.84 = 90.00, OG-II =86.66, OG-III = OG-II vs. OG-III vs. GB 80.00, GB = 82.85 2 G.I = 90.79, G.II = 95.66, G.III = G.I vs. G.II vs. G.III vs. 93.71 - 93.88 G.IV 90.84, G.IV = 98.22 3 100.00 - 100.00 H = 100.00, T = 100.00 H vs. T 3 Custom CNN[w/ DA] H = 100.00, LGG = 98.40, HGG = H vs. LGG vs. HGG 97.22 - 95.45 86.00 3 H = 99.80, AS = 97.09, OG = H vs. AS vs. OG vs. GB 97.02 - 95.75 90.40, GB = 93.71 3 AS-II = 88.50, AS-III = 94.00, AS-II vs. AS-III vs. 88.86 - 87.52 OG-II = 96.00, OG-III = 62.00, GB OG-II vs. OG-III vs. GB = 90.85 3 H = 100.00, AS-II = 93.14, AS-III H vs. AS-II vs. AS-III vs. 95.72 - 91.76 = 88.00, OG-II = 98.66, OG-III = OG-II vs. OG-III vs. GB 76.00, GB = 94.85 Cancers 2024, 16, 300 35 of 55 Table A2. Cont. Data Split No. Reference Classification Task Architecture Acc% AUC% F1% Class Performance % Method Ratio Level MN = 97.90, GL = 99.00, PT = MN vs. GL vs. PT 98.50 - 98.6 Bashir-Gonbadi and 98.60 43 Khotanlou [188] Three-way - - Auto-encoder CNN H = 98.10, LGG = 99.00, HGG = H vs. LGG vs. HGG 99.10 - 99.2 (2021) 97.70 H = 100.00, AS = 100.00, MN = H vs. AS vs. MN vs. PT 99.30 - 99 100.00, PT = 100.00, LGG = 96.60, vs. LGG vs. HGG HGG = 97.80 LGG vs. HGG vs. MT vs. LGG = 81.50, HGG = 87.00, MT Chakrabarty et al. [144] 5-fold CV, 44 PT vs. AN vs. 80:20 Patient Custom CNN 91.95 96.93 93.86 = 98.60, PA = 100.00, AN = (2021) Test H vs. MN 100.00, H = 89.70, MN = 93.30 Decuyper et al. [140] 45 LGG vs. HGG Three-way 73:11:16 Patient Custom CNN 90.00 93.98 - LGG = 89.80, HGG = 90.16 (2021) Díaz-Pernas et al. [151] GL = 99.00, MN = 93.00, PT = 46 MN vs. GL vs. PT 5-fold CV Patient Custom CNN 97.30 - - (2021) 98.00 Gab Allah et al. [94] 1 VGG19 98.54 - - GL = 100, MN = 90.20, 47 MN vs. GL vs. PT Three-way 70:15:15 - PT = 96.92 (2021) * 2 96.59 - - - VGG19 AS-1 vs. AS-II vs. AS-III (Acc) G.I = 99.06, G.II = 94.01, 48 Gilanie et al. [152] (2021) Hold-out 50:25:25 Patient Custom CNN 95.56 - - vs. AS-IV G.III = 95.31, G.IV = 97.85 1 AS = 96.86, OG = 91.27, GB = AS vs. OG vs. GB 97.64 - 94.18 49 Gu et al. [189] (2021) 5-fold CV 70:30 - Custom CNN 93.09 2 MN = 88.75, GL = 94.87, PT = MN vs. GL vs. PT 96.34 - 94.69 98.37 MN = 96.89, GL = 97.82, PT = 50 Guan et al. [153] (2021) MN vs. GL vs. PT 5-fold CV 70:30 Patient EfficientNet 98.04 - 97.79 99.24 1 - 97.27 H = 94.17, T = 97.80 96.49 51 Gull et al. [154] (2021) H vs. T 10-fold CV, Test 70:10:20 Patient GoogLeNet 2 - 97.92 H = 95.83, T = 98.14 97.31 3 - 99.12 H = 97.37, T = 99.42 98.79 G.I = 100.00, G.II = 82.35, G.III = Modality-ensemble CNN 87.00 - - 76.92, G.IV = 92.50 G.I = 0.00, G.II = 23.53, G.III = GrB 64.00 - - 42.31, G.IV = 90.74 52 Gutta et al. [106] (2021) G.I vs. G.II vs. G.III vs. G.IV Three-way 70:15:15 Patient G.I = 0.00, G.II = 35.23, G.III = RF 58.00 - - 7.69, G.IV = 92.50 G.I = 33.00, G.II = 70.00, G.III = SVM 56.00 - - 34.62, G.IV = 72.00 AlexNet - 71.93 - - 53 Hao et al. [79] (2021)* LGG vs. HGG Three-way 60:20:20 Patient TL AlexNet - 79.91 - - Cancers 2024, 16, 300 36 of 55 Table A2. Cont. Data Split No. Reference Classification Task Architecture Acc% AUC% F1% Class Performance % Method Ratio Level H vs. T 99.33 99.95 - H = 100, T = 98.80 H vs. MN vs. GL vs. PT H = 92.10, MN = 94.20, GL = 54 Irmak [190] (2021) 5-fold CV, Test 60:20:20 - Custom CNN 92.66 99.81 - vs. MT 94.40, PT = 88.00, MT = 90.00 G.II = 97.91, G.III = 100, G.IV = G.II vs. G.III vs. G.IV 98.14 99.94 - 97.01 55 Kader et al. [191] (2021) H vs. T - - - DWAE model 99.30 - 96.55 H = 96.90, T = 95.60 Custom CNN 99.25 - 95.23 (R) 95.89, (S) 93.75 GoogLeNet 89.66 - 90.32 (R) 84.85, (S) 96.00 56 Kader et al. [192] (2021) * H vs. T 5-fold CV - AlexNet 87.66 - 88.52 (R) 84.38, (S) 92.31 VGG16 84.48 - 85.25 (R) 81.25, (S) 8.48 5-fold CV, 57 Kakarla et al. [193] (2021) MN vs. GL vs. PT 80:20 - Custom CNN 97.42 - - - Test Ensemble TL CNNs DenseNet169-InceptionV3- 1 58 Kang et al. [133] (2021) * H vs. T Hold-out 80:20 - 92.16 - - - ResNeXt50-AdaBoost H vs. T DenseNet121-ResNeXt-MnasNet 2 - - - 98.83 DenseNet169-ShuffleNet- 3 H vs. MN vs. GL vs. PT 91.58 - - - MnasNet VGG19 (w/o DA) 90.03 - - LGG = 91.05, HGG = 84.03 59 Khan et al. [87] (2021) LGG vs. HGG - - - VGG19 (w/ DA) 94.06 - - LGG = 96.05, HGG = 89.09 TL ResNet50 (w/o DA) 97.48 - 97.20 97.20 60 Kumar et al. [88] (2021) MN vs. GL vs. PT 5-fold CV - TL ResNet50 (w/ DA) 97.08 - 97.20 97.20 MN vs. GL vs. PT 98.34 - - (Acc) MN = 97.81, 61 Masood et al. [194] (2021) Hold-out 70:30 - DenseNet-41-based Mask-RCNN GL = 98.62, PT = 98.60 H vs. T 97.90 - - (Acc) H = 98.06, T = 97.74 MN = 84.00, GL = 95.00, PT = TL InceptionV3 93.31 - 92.67 98.00 62 Noreen et al. [134] (2021) * MN vs. GL vs. PT 10-fold CV - Ensemble 94.34 - - - InceptionV3-KNN-SVM-RF MN = 78.00, GL = 94.00, PT = TL XceptionV3 91.63 - 90.00 100.00 Ensemble 93.79 - - - Xception-KNN-SVM-RF Custom CNN 97.10 98.90 97.00 G.II = 98.00, G.IV = 96.30 AlexNet 92.30 97.00 92.22 G.II = 94.00, G.IV = 90.70 63 Özcan et al. [155] (2021) G.II vs. G.IV 5-fold CV, Test 80:20 Patient GoogLeNet 93.30 98.70 93.30 G.II = 98.00, G.IV = 88.90 SqueezeNet 89.40 97.50 89.30 G.II = 92.00, G.IV = 87.00 [WSI] 2D CNN 77.00 - 88.60 - 64 Pei et al. [97] (2021) AS vs. OG vs. GB Hold-out 85:15 Patient [MRI] 3D CNN 69.80 - 77.10 - [WSI-MRI] Ensemble 2D-3D 80.00 - 88.60 - CNNs 65 Sadad et al. [195] (2021) * MN vs. GL vs. PT Hold-out 80:20 - Custom CNN 99.60 99.00 - - Cancers 2024, 16, 300 37 of 55 Table A2. Cont. Data Split No. Reference Classification Task Architecture Acc% AUC% F1% Class Performance % Method Ratio Level H vs. T 96.51 96.60 - (R) 96.76, (S) 96.43 Ensemble TL AlexNet, AS-II vs. AS-III 97.70 97.04 - (R) 94.63, (S) 99.44 66 Tandel et al. [107] (2021) * 5-fold CV Patient VGG16, ResNet18, OG-2 vs. OG-3 100.00 100.00 - (R) 100.00, (S) 100.00 GoogleNet, ResNet50 LGG vs. HGG 98.43 98.45 - (R) 98.33, (S) 98.57 MN = 94.81, GL = 98.48, PT = 67 Toğaçar et al. [80] (2021) MN vs. GL vs. PT Hold-out 80:20 - Custom CNN - - 96.22 95.38 Yamashiro et al. [145] 68 LGG vs. HGG Hold-out 85:15 Patient Custom CNN 91.30 92.7 - LGG = 69.20, HGG = 100.00 (2021) [WSI] 2D DenseNet 88.90 - 94.30 - 69 Yin et al. [130] (2021) AS vs. OG vs. GB Hold-out 86:14 Patient [MRI] 3D DenseNet 82.00 - 85.70 - [WSI-MRI] Ensemble 2D-3D 94.40 - 97.10 - MN = 97.31, GL = 99.51, PT = 70 Aamir et al. [156] (2022) MN vs. GL vs. PT 5-fold CV Patient Custom CNN 98.95 - 97.98 99.34 MN = 73.94, GL = 76.92, PT = ResNet50 (w/o DA) 72.63 - 71.07 65.05 71 Ahmad et al. [89] (2022) MN vs. GL vs. PT Three-way 60:20:20 - MN = 76.76, GL = 82.87, PT = ResNet50 (w/ CDA) 77.52 - 76.06 69.89 MN = 92.25, GL = 96.15, PT = ResNet50 (w/ GDA) 92.30 - 91.77 86.56 MN = 96.47, GL = 96.50, PT = ResNet50 (w/ CDA+GDA) 96.25 - 96.97 95.70 H vs. T 95.75 - - - 72 Alanazi et al. [160] (2022) Three-way 80:20, 2 Test - TL (on Kaggle-I) Custom CNN MN = 92.00, GL = 98.70, PT = MN vs. GL vs. PT 96.90 - 99.00 98.20 H = 94.70, MN = 97.30, GL = H vs. MN vs. GL vs. PT - 98.00 - - 73 Almalki et al. [121] (2022) * Hold-out 80:20, 2 Test Custom CNN-SVM 98.80, PT = 99.40 MN = 99.20, GL = 94.71, PT = MN vs. GL vs. PT 97.16 - - 99.40 MN = 97.00, GL = 99.00, PT = Custom CNN 98.70 - 98.60 99.00 MN = 97.00, GL = 96.00, PT = VGG16 97.08 - 96.60 74 Amou et al. [81] (2022) MN vs. GL vs. PT Hold-out 90:10 - 99.00 MN = 93.00, GL = 97.00, PT = VGG19 96.43 - 95.56 99.00 MN = 85.00, GL = 97.00, PT = DenseNet201 94.81 - 93.60 100.00 MN = 82.00, GL = 97.00, PT = InceptionV3 92.86 - 92.00 96.00 MN = 57.00, GL = 77.00, PT = ResNet50 89.29 - 89.00 98.00 Cancers 2024, 16, 300 38 of 55 Table A2. Cont. Data Split No. Reference Classification Task Architecture Acc% AUC% F1% Class Performance % Method Ratio Level LOOCV (on 2-stage Ensemble H = 100.00, MN = 99.00, GL = 75 Aurna et al. [82] (2022) H vs. MN vs. GL vs. PT - 98.96 98.90 99.00 dataset) EfficientNetB0-ResNet50- 98.00, PT = 99.00 Custom CNN H = 99.04, LGG = 91.43, HGG = (2+1)D ResNet - - 90.35 82.29 H = 99.88, LGG = 91.08, HGG = TL (2+1)D ResNet - - 92.37 87.05 H = 97.69, LGG = 88.60, HGG = 2D-3D Mixed ResNet - - 86.07 75.05 76 Chatterjee et al. [59] (2022) H vs. LGG vs. HGG 3-fold CV, Test 70:30 Patient H = 99.51, LGG = 93.19, HGG = TL 2D-3D Mixed ResNet 96.98 - 93.45 88.37 H = 99.44, LGG = 92.06, HGG = 3D ResNet18 - - 90.95 82.89 H = 99.97, LGG = 85.52, HGG = TL 3D ResNet18 - - 89.25 83.53 Custom CNN 90.60 95.60 91.48 (R) 91.50, (S) 97.99 DenseNet101 86.80 92.84 87.84 (R) 86.14, (S) 96.07 77 Chitnis et al. [197] (2022) * H vs. MN vs. GL vs. PT Hold-out 88:12 - VGGNet16 88.33 94.31 89.60 (R) 88.15, (S) 98.61 ResNet50 85.79 94.34 86.96 (R) 85.17, (S) 95.77 Modality-ensemble TL CNNs 86.38 - - - 78 Coupet et al. [135] (2022) * H vs. T Three-way 70:15:15 Patient TL 3DUNet 82.96 - - H = 69.81, T = 96.44 79 Dang et al. [98] (2022) LGG vs. HGG Three-way 60:20:20 - VGG 97.44 - - - LGG vs. HGG 67.00 76.00 - (R) 58.00, (S) 78.00 (3D) DenseNet G.I vs. G.II vs. G.III vs. G.I = 100.00, G.II = 63.00, G.III = 80 Danilov et al. [146] (2022) Three-way 80:10:10 - 83.00 95.00 80.25 G.IV 100.00, G.IV = 85.00 LGG vs. HGG 61.00 73.00 - (R) 44.00, (S) 81.00 (2D) TL ResNet200e G.I vs. G.II vs. G.III vs. G.I = 56.00, G.II = 45.00, G.III = 50.00 72.00 35.00 G.IV 32.00, G.IV = 47.00 Radiomics 74.00 82.20 - (R) 80.00, (S) 68.00 81 Ding et al. [57] (2022) * LGG vs. HGG Hold-out Patient VGG16 60.00 71.20 - (R) 68.00, (S) 52.00 Ensemble Radiomics-VGG16-RF 80.00 89.80 - (R) 84.00, (S) 76.00 H = 97.50, MN = 92.50, GL = Bayesian CNN 94.32 - 94.00 85.50, PT = 100.00 MobileNet 93.42 - 94.00 94.00 82 Ekong et al. [198] (2022) H vs. MN vs. GL vs. PT Three-way 80:10:10 - AlexNet 92.75 - 93.00 93.00 VGG16 89.51 - 91.00 91.00 ResNet50 86.58 - 86.00 87.00 83 Gao et al. [112] (2022) * 18 types of tumors * Three-way 72:24:4 Patient DenseNet 81.20 92.00 - (R) 87.60, (S) 84.90 84 Gaur et al. [199] (2022) MN vs. GL vs. PT Three-way 80:10:10 - Custom CNN 85.37 - - - Radiomics 83.70 87.00 83.40 (R) 70.40, (S) 89.90 85 Guo et al. [150] (2022) AS vs. OG vs. GB 3-fold CV - Modality-fusion DenseNet201 84.60 88.30 84.60 (R) 73.10, (S) 93.00 Modality-ensemble 87.80 90.2 87.80 (R) 77.20, (S) 93.00 DenseNet201 Cancers 2024, 16, 300 39 of 55 Table A2. Cont. Data Split No. Reference Classification Task Architecture Acc% AUC% F1% Class Performance % Method Ratio Level H vs. T 96.66 - 97.00 H = 100.00, T = 93.00 86 Gupta et al. [95] (2022) Hold-out 88:12 - InceptionResNetV2-RF MN = 100.00, GL = 100.00, PT = MN vs. GL vs. PT 96.88 - 96.00 85.00 Custom CNN 99.40 - 98.10 (R) 97.20, (S) 98.60 Gurunathan and AlexNet 98.14 - - - 87 LGG vs. HGG Hold-out 75:25 - Krishnan [200] (2022) VGG19 97.97 - - - GoogLeNet 95.69 - - [w/o DA] TL ResNet50 99.10 98.78 99.50 (R) 89.60, (S) 100.00 TL VGG-16 98.78 98.06 97.49 (R) 84.64, (S) 99.80 88 Haq et al. [90] (2022) * MN vs. GL vs. PT Hold-out 70:30 - TL InceptionV3 97.78 97.00 97.39 (R) 92.23, (S) 96.88 [w/ DA] TL ResNet50 99.89 99.56 99.43 (R) 96.13, (S) 99.08 TL VGG-16 98.98 97.98 98.79 (R) 97.87, (S) 100.00 TL InceptionV3 98.50 98.76 98.00 (R) 98.56, (S) 100.00 [WSI] 2D ResNet50 77.70 - 88.60 - 89 Hsu et al. [131] (2022) AS vs. OG vs. GB Three-way 67:11:22 Patient [MRI] 3D ResUNet 69.80 - 77.10 - [WSI-MRI] ResNet50-ResUNet 80.00 - 88.60 - Isunuri and Kakarla [201] 90 MN vs. GL vs. PT 5-fold CV - Custom CNN 97.52 - 97.26 97.19 (2022) 91 Jeong et al. [113] (2022) LGG vs. HGG 5-fold CV - Custom CNN 90.91 96.34 - (R) 92.69, (S) 84.90 SVM-KNN 80.14 80.93 - - 1 AlexNet 91.88 92.67 - - MN vs. GL vs. PT VGGNet 89.96 90.29 - - MN = 98.10, GL = 98.88, PT = 92 Kazemi et al. [108] (2022) * Hold-out 75:25 - AlexNet-VGGNet 98.06 99.14 - 98.50 SVM-KNN 82.44 84.63 - - 2 AlexNet 92.59 92.9 - - G.II vs. G.III vs. G.IV VGGNet 90.05 90.51 - - MN = 98.02, GL = 95.90, PT = AlexNet-VGGNet 98.99 99.23 - 98.95 93 Khazaee et al. [202] (2022) LGG vs. HGG Hold-out 80:20 - TL EfficientNetB0 98.87 - - (R) 98.86, (S) 98.79 Ensemble AlexNet-GoogLeNet- MN = 99.80, GL = 98.96, PT = 94 Kibriya et al. [122] (2022) MN vs. GL vs. PT - - - 99.70 100.00 - ResNet18-SVM 100.00 H vs. T - 90.00 - 90.00 - 95 Koli et al. [203] (2022) Three-way 70:15:15 TL ResNet50 MN = 90.00, GL = 98.00, PT = MN vs. GL vs. PT 96.00 - 95.00 97.00 Lakshmi and Rao [204] 96 H vs. MN vs. GL vs. PT Hold-out 80:20 - InceptionV3 89.00 - - - (2022) Cancers 2024, 16, 300 40 of 55 Table A2. Cont. Data Split No. Reference Classification Task Architecture Acc% AUC% F1% Class Performance % Method Ratio Level MN = 99.03, GL = 98.82, PT = MN vs. GL vs. PT 98.92 98.93 97.87 97 Maqsood et al. [114] (2022) 5-fold CV - TL MobileNetV2-SVM 98.79 LGG vs. HGG 97.47 - 96.71 (R) 97.22, (S) 97.94 98 Murthy et al. [205] (2022) * H vs. T - - - Custom CNN 95.26 - 97.52 (R) 97.12, (S) 50.00 H = 98.00, MN = 100.00, GL = TL EfficientNet 98.78 - 98.75 97.00, PT = 100.00 99 Nayak et al. [206] (2022) MN vs. GL vs. PT Hold-out 80:20 - H = 98.00, MN = 98.00, GL = TL ResNet50 96.33 - 96.50 90.00, PT = 100.00 H = 98.00, MN = 95.00, GL = TL MobileNet 96.94 - 97.00 94.00, PT = 100.00 H = 96.00, MN = 99.00, GL = TL MobileNetV2 94.80 - 95.00 95.00, PT = 90.00 TL VGG16-SoftMax 96.50 - 96.55 (R) 97.03, (S) 95.96 Rajinikanth et al. [124] TL VGG16-DT 96.00 - 96.00 (R) 96.97, (S) 95.05 100 LGG vs. HGG 5-fold CV 90:10 - (2022) TL VGG16-KNN 96.50 - 96.52 (R) 97.00, (S) 96.00 TL VGG16-SVM 97.00 - 97.00 (R) 97.00, (S) 97.00 TL GoogLeNet 93.10 - - H = 95.20, MN = 85.10, 101 Rasool et al. [125] (2022) H vs. MN vs. GL vs. PT Hold-out 80:20 - GL = 97.00, PT = 100.00 H = 98.70, MN = 97.30, GL = GoogLeNet-SVM 98.10 - - 97.80, PT = 98.90 Custom TL GoogLeNet 99.67 - 99.66 (R) 100.00 TL AlexNet 97.80 - 97.66 (R) 97.66 TL GoogLeNet 98.26 - 98.33 (R) 98.66 TL ShuufleNet 98.37 - 98.33 (R) 98.66 TL ResNet50 98.60 - 98.33 (R)98.66 102 Raza et al. [207] (2022) MN vs. GL vs. PT Hold-out 70:30 - TL MobileNetV2 99.00 - 99.00 (R) 99.00 TL SqueezeNet 97.91 - 97.66 (R) 98.00 TL Darknet53 99.13 - 99.00 (R) 99.33 TL ResNet101 98.91 - 98.66 (R) 99.00 TL ExceptionNet 98.69 - 98.00 (R) 98.33 (Acc) MN = 98.92, GL = 96.72, MN vs. GL vs. PT 99.80 - 103 Rizwan et al. [208] (2022) Train, Val+Test 65:35 - Custom CNN - PT = 97.81 (Acc) G.II = 99.00, G.III = 96.00, G.II vs. G.III vs. G.IV 97.14 - - G.IV = 99.00 Cancers 2024, 16, 300 41 of 55 Table A2. Cont. Data Split No. Reference Classification Task Architecture Acc% AUC% F1% Class Performance % Method Ratio Level MN = 99.00, GL = 99.00, PT = TL hybrid GoogLeNet-AlexNet 99.10 99.00 - 99.00 MN = 96.00, GL = 96.00, PT = TL AlexNet 96.00 97.00 - 96.00 MN = 95.00, GL = 95.00, PT = TL VGG16 95.00 95.00 - 95.00 104 Samee et al. [209] (2022) MN vs. GL vs. PT Hold-out 70:30 - MN = 95.00, GL = 95.00, PT = TL MobileNetV2 95.00 95.00 - 95.00 MN = 94.00, GL = 94.00, PT = TL ResNet 94.00 94.00 - 94.00 MN = 92.00, GL = 92.00, PT = TL SqueezeNet 92.00 92.00 - 92.00 10-fold CV, 105 Samee et al. [147] (2022) LGG vs. HGG 70:15:15 Patient Custom CNN 88.60 - - LGG = 80.00, HGG = 88.60 Test Sangeetha et al. [210] 106 H vs. T LOOCV Patient TL (in Rembrandt) CNN 94.00 - - (R) 85.00, (S) 73.00 (2022) SVM-RBF 85.80 - 85.10 (R) 81.90 1 GoogLeNet 94.60 - 90.90 (R) 91.50 LGG vs. HGG vs. PIT CDbLNL 97.21 - 95.72 (R) 95.62 107 Saravanan et al. [109] (2022) 10-fold CV - SVM-RBF 84.80 - 84.10 (R) 80.90 2 GoogLeNet 91.60 - 90.10 (R) 91.50 OLI vs. EP vs. CAM CDbLNL 97.21 - 94.34 (R) 93.86 MN = 96.92, GL = 91.13, PT = TL GoogLeNet-SoftMax 94.90 - 94.30 97.77 108 Sekhar et al. [126] (2022) MN vs. GL vs. PT 5-fold CV Patient MN = 97.96, GL = 94.59, PT = TL GoogLeNet-SVM 97.60 - 97.35 100.00 MN = 94.57, GL = 98.02, PT = TL GoogLeNet-KNN 98.30 - 97.24 99.10 H = 91.10, MN = 89.80, H= GL = 93.30, PT = 97.80 AlexNet-SoftMax 93.30 - - 91.10, MN = 89.80, GL = 93.30, 109 Senan et al. [99] (2022) H vs. MN vs. GL vs. PT Hold-out 80:20 - PT = 97.80 AlexNet-SVM 95.10 - - H = 94.90, MN = 93.60, GL = 93.90, PT = 97.80 ResNet18-SoftMax 93.80 - - H = 87.30, MN = 93.60, GL = 93.30, PT = 97.20 ResNet18-SVM 91.20 - - H = 92.40, MN = 86.10, GL = 91.50, PT = 95.60 TL VGG16 86.05 - - B = 89.47, M = 87.09 110 Srinivas et al. [211] (2022) Benign vs. Malignant Three-way - - TL InceptionV3 64.00 - - B = 5.55, M = 100.00 TL ResNet50 74.00 - - B = 89.47, M = 64.52 Cancers 2024, 16, 300 42 of 55 Table A2. Cont. Data Split No. Reference Classification Task Architecture Acc% AUC% F1% Class Performance % Method Ratio Level TL Ensemble AlexNet, VGGNet, ResNet18, GoogLeNet, ResNet50 111 Tandel et al. [75] (2022) * LGG vs. HGG 5-fold CV - [Whole image] 98.43 98.45 - (R) 98.33, (S) 98.57 [Skull-stripped brain] 98.63 98.63 - (R) 98.63. (S) 98.57 [Tumor ROI] 99.06 99.07 - (R) 99.04, (S) 99.10 Hold-out 70:30 - 95.64 - 92.41 (R) 92.12, (S) 95.97 80:20 - 95.78 - 91.91 (R) 95.12, (S) 95.10 112 Tripathi and Bag [83] (2022) * LGG vs. HGG DST Fusion TL ResNets 90:10 - 96.19 - 94.13 (R) 96.95, (S) 95.77 Average - 95.87 - 92.82 - Tripathi and Bag [141] 113 LGG vs. HGG 10-fold CV Patient Attention-based CNN 95.86 - 94.84 (R) 94.82, (S) 96.81 (2022) Tummala et al. [136] 114 MN vs. GL vs. PT Three-way 70:10:20 - Ensemble ViT 98.70 - - (R) 97.78, (S) 99.42 (2022) - CNN 89.39 - - (R) 98.30 115 Vankdothu et al. [213] (2022) H vs. MN vs. GL vs. PIT Hold-out 88:12 RNN 90.02 - - (R) 98.00 CNN-LSTM 92.00 - - (R) 98.50 [WSI] Ensemble EfficientNet-B2, 82.20 - 88.60 EfficientNet-B3, SE-ResNext10 116 Wang et al. [132] (2022) AS vs. OG vs. GB Three-way 70:10:20 Patient [MRI] 3D CNN 73.30 - 82.90 [WSI-MRI] 2D-3D Ensemble 75.00 - 75.30 - AST = 85.70, OLI = 40.00, GBM = [MRI] TL ResNet34 67.50 - - 117 Xiong et al. [115] (2022) * AS vs. OG vs. GB Three-way 70:15:15 Patient 68.80 AST = 85.70, OLI = 30.00, GBM = [MRI-tabular] TL ResNet34 70.00 - - 81.30 1 83.33 - - (R) 90.8 TL ResNet18 1 88.10 - - (R) 90.1 118 Xu et al. [118] (2022) * LGG vs. HGG Three-way 60:20:20 Patient TL ResNet18+radiomics 2 87.40 - - (R) 93.1 TL ResNet18 2 94.10 - - (R) 97.1 TL ResNet18+radiomics TL AlexNet 87.89 - 88.03 (R) 87.86, (S) 85.42 TL ResNet 91.98 - 91.59 (R) 91.44, (S) 89.79 119 Yazdan et al. [214] (2022) * H vs. MN vs. GL vs. PT k-fold CV - Multi-scale CNN 1 89.27 - 89.41 (R) 89.15, (S) 86.91 Multi-scale CNN 2 94.19 - 94.06 (R) 93.74, (S) 92.62 Multi-scale CNN 3 89.67 - 89.49 (R) 89.24, (S) 88.35 ResNet18-Softmax 97.43 - 97.56 (R) 98.12 1 TL ResNet18-Softmax 98.91 - 98.69 (R) 99.66 H vs. T 60:40 120 Zahoor et al. [103] (2022) * Hold-out - TL ResNet18-SVM 99.16 - 98.94 (R) 97.99 Custom CNN-SVM 99.56 99.90 99.45 (R) 98.99 2 MN = 98.60, GL = 99.30, PT = MN vs. GL vs. PT 80:20 Custom CNN-SVM 99.20 - 99.09 99.50 Cancers 2024, 16, 300 43 of 55 Table A2. Cont. Data Split No. Reference Classification Task Architecture Acc% AUC% F1% Class Performance % Method Ratio Level H = 87.50, MN = 88.00, GL = TL GoogLeNet-SoftMax 88.00 - 88.46 88.50, PT = 88.00 121 AlTahhan et al. [127] (2023) H vs. MN vs. GL vs. PT Three-way 70:30:- - H = 84.00, MN = 84.60, GL = TL AlexNet-SoftMax 85.00 - 86.27 88.00, PT = 83.30 H = 92.60, MN = 92.30, GL = TL AlexNet-SVM 95.00 - 93.62 100.00, PT = 96.00 H = 96.20, MN = 96.00, GL = TL AlexNet-KNN 97.00 - 97.96 100.00, PT = 96.00 Al-Zoghby et al. [137] Ensemble TL VGG-16 & Custom MN = 98.00, GL = 100.00, PT = 122 MN vs. GL vs. PT Hold-out 80:20 - 99.00 99.00 99.00 (2023) CNN 99.00 TL EfficientNetv2 99.85 99.92 98.07 98.05 TL ResNet18 99.62 99.75 96.64 96.71 123 Anagun [215] (2023) MN vs. GL vs. PT Three-way 80:10:10 - TL ResNet200d 99.83 99.84 97.72 97.66 TL InceptionV4 99.69 99.73 97.19 97.37 TL EfficientNetB0 - - 54.50 H = 44.00, T = 30.00 TL InceptionV3 - - 91.50 H = 90.00, T = 94.00 TL ResNet50 - - 85.00 H = 82.00, T = 81.00 124 Anand et al. [91] (2023) H vs. T Three-way 76:14:10 - TL VGG19 95.00 - 96.00 H = 98.00, T = 96.00 Custom CNN w/o DA 96.00 - 96.50 H = 95.00, T = 98.00 Custom CNN w/ DA 97.00 - 97.00 H = 98.00, T = 96.00 Ensemble TL VGG19 & Custom 98.00 - 98.50 H = 98.50, T = 99.00 CNN H = 99.60, MN = 90.62, GL = Attention VGG19 93.53 95.3 90.55 96.76, PT = 91.61 Apostolopoulos et al. [216] VGG19 91.08 - - - 125 H vs. MN vs. GL vs. PT H10-fold CV - (2023) * ResNet152 86.00 - - - MobileNetV2 86.89 - - - InceptionV3 87.13 - - - MN = 82.98, GL = 92.63, PT = TL Xception 91.83 - 90.65 97.31 MN = 84.40, GL = 96.49, PT = TL VGG16 93.54 - 93.01 97.31 MN = 92.91, GL = 98.60, PT = TL DenseNet201 97.22 98.00 96.81 98.39 126 Asif et al. [138] (2023) MN vs. GL vs. PT Hold-out 80:20 - MN = 92.91, GL = 94.74, PT = TL ResNet152V2 95.58 98.00 95.12 98.92 MN = 89.36, GL = 97.54, PT = TL InceptionResNetV2 95.75 96.00 94.96 97.85 Ensemble TL DenseNet201, MN = 96.45, GL = 99.29, PT = ResNet152V2, 98.69 99.00 98.39 99.46 InceptionResNetV2 Cancers 2024, 16, 300 44 of 55 Table A2. Cont. Data Split No. Reference Classification Task Architecture Acc% AUC% F1% Class Performance % Method Ratio Level MN = 97.00, GL = 98.00, PT = TL ResNet152 98.85 98.00 - Athisayamani et al. [110] 99.00 127 MN vs. GL vs. PT - - - (2023) CNN 97.00 - - (R) 94.00 SVM 94.00 - - (R) 94.00 SVM 89.53 - - - TL AlexNet 98.67 - - - 128 Bairagi et al. [111] (2023) * H vs. T 10-fold CV, Test 80:20 - TL VGG16 90.67 - - - TL GoogLeNet 91.49 - - - 1 - (R) 93.13, (S) 92.84 92.10 129 Deepa et al. [84] (2023) * H vs. T Hold-out 90:10 - Custom CJHBA Based DRN 2 - 91.84 - (R) 91.55, (S) 91.86 TL VGG16 92.07 - - - TL VGG19 93.05 - - - TL InceptionV3 80.35 - - - TL ResNet50 74.48 - - - 130 El-Wahab et al. [219] (2023) MN vs. GL vs. PT 5-fold CV, Test 80:20 - TL MobileNet 89.16 - - - BTCfCNN 93.08 - 92.21 (R) 92.01, (S) 96.34 TL BTCfCNN 98.63 - 98.46 (R) 98.49, (S) 99.31 (bt folds) TL BTC-fCNN 98.86 - 98.77 (R) 98.83, (S) 99.41 H = 100.00, MN = 98.00, GL = TL InceptionV3 95.72 - 69.00 31.00, PT = 70.00 H = 100.00, MN = 99.00, GL = TL VGG16 95.11 - 69.00 22.00, PT = 80.00 131 Hossain et al. [116] (2023) H vs. MN vs. GL vs. PT three-way 80:10:10 - H = 98.00, MN = 91.00, GL = TL Xception 94.50 - 69.00 39.00, PT = 77.00 H = 100.00, MN = 97.00, GL = TL ResNet50 93.88 - 72.00 28.00, PT = 72.00 H = 100.00, MN = 97.00, GL = TL VGG19 94.19 - 64.00 22.00, PT = 64.00 H = 98.00, MN = 99.00, GL = TL InceptionResNetV2 93.58 - 70.00 33.00, PT = 68.00 Ensemble TL VGG16, H = 100, MN = 93.00, GL = 49.00, 96.94 - 76.00 InceptionV3, Xception PT = 73.00 3D CNN - T1 94.00 - 95.77 - - T1c 94.00 - 95.77 - 132 Hussain et al. [148] (2023) LGG vs. HGG Hold-out - Patient - T2 94.38 - 95.65 - - Flair 93.23 - 95.77 - - Segmentation 94.38 - 95.77 - Ensemble 94.20 - 95.75 - Cancers 2024, 16, 300 45 of 55 Table A2. Cont. Data Split No. Reference Classification Task Architecture Acc% AUC% F1% Class Performance % Method Ratio Level 1 72.00 - - - Radiomics-SVM 1 84.00 - - - Radiomics-KNN 1 92.10 - - - VGG16-SVM 1 88.10 - - - VGG16-KNN 1 93.30 99.00 93.50 93.00 Radiomics+VGG16-SVM 1 96.00 99.00 94.50 95.50 133 Kibriya et al. [119] (2023) H vs. T Hold-out 70:30 - Radiomics+VGG16-KNN 2 96.10 - - - Radiomics-SVM 2 96.00 - - - Radiomics-KNN 2 98.00 - - - VGG16-SVM 2 97.80 - - - VGG16-KNN 2 99.00 100.00 99.00 99.00 Radiomics+VGG16-SVM 2 98.70 100.00 99.00 99.00 Radiomics+VGG16-KNN [w/o DA] TL VGG16 90.50 - - - TL VGG19 90.70 - - - TL ResNet 50 88.02 - - - Krishnapriya and Karuna [92] TL InceptionV3 66.26 - - - 134 H vs. T Hold-out 70:30 - (2023) [w/ DA] TL VGG 16 99.00 - 99.08 98.18 TL VGG19 99.48 - 99.17 98.76 TL ResNet50 97.92 - 82.24 87.27 TL InceptionV3 81.25 - 58.16 63.25 ResNet50-Softmax 86.57 - - - 135 Kumar et al. [128] (2023) * Benign vs. Malignant Hold-out 90:10 - ResNet50-SVM 91.24 - - - TL ResNet50 96.80 - 97.34 Benign = 95.21, Malignant = 97.56 Custom CNN 93.30 98.43 - 91.13 ResNet50 81.10 94.2 - 81.04 136 Mahmud et al. [220] (2023) H vs. M vs GL vs. PT Three-way 80:10:10 - VGG16 71.60 89.6 - 70.03 InceptionV3 80.00 89.14 - 79.81 Muezzinoglu et al. [221] H = 98.40, MN = 98.51, GL = 137 H vs. MN vs. GL vs. PT 10-fold CV - PatchResNet 98.10 98.01 (2023) 95.68, PT = 100.00 H = 98.00, MN = 84.00, GL = CNN - - 92.00 90.00, PT = 97.00 Özkaraca et al. [222] 138 H vs. MN vs. GL vs. PT 10-fold CV, Test 80:20 - H = 96.00, MN = 67.00, GL = (2023) VGG16 - - 85.75 89.00, PT = 94.00 H = 99.00, MN = 83.00, GL = DenseNet - - 84.75 99.00, PT = 58.00 H = 98.00, MN = 91.00, GL = Custom CNN - - 96.5 97.00, R PT = 99.00 TL MobileNetV2 99.85 99.92 99.85 - TL DenseNet201 99.66 99.77 99.67 - Özkaya and Şağıroğlu [224] 139 LGG vs. HGG 10-fold CV - TL Xception 99.63 99.70 99.64 - (2023) TL InceptionV3 99.63 99.74 99.64 - TL EfficientNetV2S 99.24 99.41 99.25 - Cancers 2024, 16, 300 46 of 55 Table A2. Cont. Data Split No. Reference Classification Task Architecture Acc% AUC% F1% Class Performance % Method Ratio Level MN = 95.00, GL = 99.00, PT = Custom CNN 98.04 98.00 98.00 100.00 MN = 79.00, GL = 92.00, PT = VGG16 90.70 93.00 90.00 99.00 MN = 85.00, GL = 94.00, PT = 140 Rasheed et al. [225] (2023) MN vs. GL vs. PT Hold-out 80:20 - VGG19 92.82 94.00 93.00 98.00 MN = 89.00, GL = 95.00, PT = ResNet50 94.77 96.00 95.00 99.00 MN = 90.00, GL = 92.00, PT = MobileNet 93.47 95.00 93.00 99.00 MN = 66.00, GL = 89.00, PT = InceptionV3 85.97 88.00 85.00 98.00 Inception CNN [Flair] 69.00 - 60.00 (R) 75.00, (S) 60.00 141 Rui et al. [149] (2023) * LGG vs. HGG 5-fold CV, Test - Patient [T1c] 74.00 - 70.00 (R) 75.00, (S) 73.00 Modality-ensemble 80.00 - 78.00 (R) 76.00, (S) 87.00 TL VGG16-Softmax 96.93 - 96.64 (R) 99.29 142 Shirehjini et al. [123] (2023) * G.I vs. G.II vs. G.III vs. G.IV Three-way 70:15:15 - TL VGG16-LR 98.15 - 98.12 (R) 97.94 G.I: 96.00, G.II = 100.00, G.III = TL-SVM 99.38 99.93 99.09 100.00, G.IV = 100.00 Custom CNN 98.17 - - (R) 98.79, (S)91.34 143 Srinivasan et al. [226] (2023) H vs. MN vs. GL vs. PT Hold-out 80:20 - UNet 92.61 - - (R) 97.56 (S) 81.51 ResNet 96.23 - - (R) 97.90, (S) 90.23 Ensemble TL AlexNet, [T1] 94.75 94.92 - (R) 94.29, (S) 95.56 144 Tandel et al. [139] (2023) * LGG vs. HGG 5-fold CV - VGG16, ResNet18, GoogLeNet, [T2] 97.98 97.99 - (R) 97.60, (S) 98.37 ResNet50 [Flair] 98.88 98.88 - (R) 98.95, (S) 98.80 G.II = 75.00, G.III = 17.00, G.IV = van der Voort et al. [117] G.II vs. G.III vs. G.IV 71.00 81.00 - 145 Three-way 75:15:15 Patient UNet 95.00 (2023) LGG vs. HGG 84.00 91.00 - (R) 72.00, (S) 93.00 Attention-based custom CNN 95.19 98.40 93.34 (R) 94.01, (S) 99.53 VGG19 - 95.80 - - 146 Wu et al. [157] (2023) LGG vs. HGG Three-way 54:13:33 Patient ResNet50 - 94.10 - - DenseNet201 - 95.70 - - InceptionV4 - 97.00 - - AS: Astrocytoma, Acc: Accuracy, AUC: Area Under the Receiver Operating Characteristic Curve, CDA: Classic Data Augmentation, CDbLNL: Convolutional Neural Network Database Learning with Neighboring Network Limitation, CJHBA: Chronological Jaya Honey Badger Algorithm, CNN: Convolutional Neural Network, CV: Cross-Validation, DA: Data Augmentation, DRLBP: Dominant Rotated Local Binary Patterns, DRN: Deep Residual Network, DT: Decision Tree, DCGAN: Deep Convolutional Generative Adversarial Network, DWAE: Deep Wavelet Auto-Encoder, DWT: Discrete Wavelet Transform, ELM: Extreme Learning, Machine, EP: Ependymoma, GA: Genetic Algorithm, GAN: Generative Adversarial Network, GB: Glioblastoma, GDA: Generative Data Augmentation, GL: Glioma, GL: Glioma, HGG: High-grade Glioma, KNN: K-Nearest Neighbors, LGG: Low-grade Glioma, LOOCV: Leave-One-Out Cross-Validation, LR: Logistic Regression, LSTM: Long Short-Term Memory, MB: Medulloblastoma, MN: Meningioma, MT: Metastasis, OG: Oligodendroglioma, PT: Pituitary, (P): Precision, (R): Recall, RF: Random Forest, RFE: Recursive Feature Elimination, ROI: Region of Interest, (S): Specificity, Std: Standardization, SVM: Support Vector Machine, TL: Transfer Learning, WSI: Whole Slide Image. Papers highlighted with an asterisk (*) indicate that not all outcomes are reported. For comprehensive details, readers are referred to the original paper. Numerical superscripts link models with datasets in Table A1 when different data sources yield individual results. Cancers 2024, 16, 300 47 of 55 References 1. Sohn, E. The reproducibility issues that haunt health-care AI. Nature 2023, 613, 402–403. 2. McDermott, M.; Wang, S.; Marinsek, N.; Ranganath, R.; Foschini, L.; Ghassemi, M. Reproducibility in machine learning for health research: Still a ways to go. Sci. Transl. Med. 2021, 13, eabb1655. 3. Muehlematter, U.; Daniore, P.; Vokinger, K. Approval of artificial intelligence and machine learning-based medical devices in the USA and Europe (2015–20): A comparative analysis. Lancet Digit. Health 2021, 3, e195–e203. 4. Nakagawa, K.; Moukheiber, L.; Celi, L.; Patel, M.; Mahmood, F.; Gondim, D.; Hogarth, M.; Levenson, R. AI in Pathology: What could possibly go wrong? Semin. Diagn. Pathol. 2023, 40, 100–108. 5. Di Nunno, V.; Fordellone, M.; Minniti, G.; Asioli, S.; Conti, A.; Mazzatenta, D.; Balestrini, D.; Chiodini, P.; Agati, R.; Tonon, C.; et al. Machine learning in neuro-oncology: Toward novel development fields. J. Neuro-Oncol. 2022, 159, 333–346. 6. Bacciu, D.; Lisboa, P.; Vellido, A. Deep Learning in Biology and Medicine; World Scientific: London, UK, 2022. 7. Bernal, J.; Kushibar, K.; Clèrigues, A.; Oliver, A.; Lladó, X. Deep learning for medical imaging. In Deep Learning in Biology and Medicine; World Scientific: London, UK, 2022; pp. 11–54. 8. Xue, H.; Hu, G.; Hong, N.; Dunnick, N.; Jin, Z. How to keep artificial intelligence evolving in the medical imaging world? Challenges and opportunities. Sci. Bull. 2023, 68, 648–652. 9. Pati, S.; Baid, U.; Edwards, B.; Sheller, M.; Wang, S.-H.; Reina, G.A.; Foley, P.; Gruzdev, A.; Karkada, D.; Davatzikos, C.; et al. Federated learning enables big data for rare cancer boundary detection. Nat. Commun. 2022, 13, 7346. 10. Thrall, J.; Li, X.; Quanzheng, L.; Cruz, C.; Do, S.; Dreyer, K.; Brink, J. Artificial Intelligence and Machine Learning in Radiology: opportunities, challenges, pitfalls, and criteria for success. J. Am. Coll. Radiol. 2018, 15, 504–508. 11. Liu, Y.; Leong, A.; Zhao, Y.; Xiao, L.; Mak, H.; Tsang, A.; Lau, G.; Leung, G.; Wu, E. A low-cost and shielding-free ultra-low-field brain MRI scanner. Nat. Commun. 2021, 12, 7238. 12. Julià-Sapé, M.; Acosta, D.; Majós, C.; Moreno-Torres, A.; Wesseling, P.; Acebes, J.; Griffiths, J.R.; Arús, C. Comparison between neuroimaging classifications and histopathological diagnoses using an international multicenter brain tumor magnetic resonance imaging database. J. Neurosurg. 2006, 105, 6–14. 13. Arita, K.; Miwa, M.; Bohara, M.; Moinuddin, F.; Kamimura, K.; Yoshimoto, K. Precision of preoperative diagnosis in patients with brain tumor—A prospective study based on “top three list” of differential diagnosis for 1061 patients. Surg. Neurol. Int. 2020, 11, 55. 14. Osborn, A.; Louis, D.; Poussaint, T.; Linscott, L.; Salzman, K.L. The 2021 World Health Organization classification of tumors of the central nervous system: What neuroradiologists need to know. Am. J. Neuroradiol. 2022, 43, 928–937. 15. Wen, P.Y.; Macdonald, D.R.; Reardon, D.A.; Cloughesy, T.F.; Sorensen, A.G.; Galanis, E.; DeGroot, J.; Wick, W.; Gilbert, M.R.; Lassman, A.B.; et al. Updated response assessment criteria for high-grade gliomas: Response assessment in neuro-oncology working group. J. Clin. Oncol. 2010, 28, 1963–1972. https://doi.org/10.1200/JCO.2009.26.3541. 16. Kumar, A.; Leeds, N.; Fuller, G.; Van Tassel, P.; Maor, M.; Sawaya, R.; Levin, V. Malignant gliomas: MR imaging spectrum of radiation therapy- and chemotherapy-induced necrosis of the brain after treatment. Radiology 2000, 217, 377–384. 17. Segura, P.P.; Quintela, N.V.; García, M.M.; del Barco Berrón, S.; Sarrió, R.G.; Gómez, J.G.; Castaño, A.G.; Martín, L.M.N.; Rubio, O.G.; Losada, E.P. SEOM-GEINO clinical guidelines for high-grade gliomas of adulthood (2022). Clin. Transl. Oncol. 2023, 25, 2634–2646. https://doi.org/10.1007/S12094-023-03245-Y. 18. Da Cruz, L.C.H.; Rodriguez, I.; Domingues, R.; Gasparetto, E.; Sorensen, A. Pseudoprogression and Pseudoresponse: Imaging Challenges in the Assessment of Posttreatment Glioma. AJNR Am. J. Neuroradiol. 2011, 32, 1978–1985. https://doi.org/10.3174/ AJNR.A2397. 19. Wen, P.Y.; van den Bent, M.; Youssef, G.; Cloughesy, T.F.; Ellingson, B.M.; Weller, M.; Galanis, E.; Barboriak, D.P.; de Groot, J.; Gilbert, M.R.; et al. RANO 2.0: Update to the response assessment in neuro-oncology criteria for high- and low-grade gliomas in adults. J. Clin. Oncol. 2023, 41, 5187–5199. https://doi.org/10.1200/JCO.23.01059. 20. Tustison, N.J.; Avants, B.B.; Cook, P.A.; Zheng, Y.; Egan, A.; Yushkevich, P.A.; Gee, J.C. N4ITK: Improved N3 bias correction. IEEE Trans. Med. Imaging 2010, 29, 1310–1320. https://doi.org/10.1109/TMI.2010.2046908. 21. Sled, J.G.; Zijdenbos, A.P.; Evans, A.C. A nonparametric method for automatic correction of intensity nonuniformity in MRI data. IEEE Trans. Med. Imaging 1998, 17, 87–97. 22. Ronneberger, O.; Fischer, P.; Brox, T. U-Net: Convolutional Networks for biomedical image segmentation. In Proceedings of the Medical Image Computing and Computer-Assisted Intervention—MICCAI, Munich, Germany, 5–9 October 2015; Springer International Publishing: Berlin/Heidelberg, Germany, 2015; pp. 234–241. 23. Ren, S.; He, K.; Girshick, R.; Sun, J. Faster R-CNN: Towards real-time object detection with region proposal networks. In Proceedings of the Advances in Neural Information Processing Systems; Cortes, C., Lawrence, N., Lee, D., Sugiyama, M., Garnett, R., Eds.; Curran Associates, Inc.: Red Hook, NY, USA, 2015; Volume 28. 24. He, K.; Gkioxari, G.; Dollár, P.; Girshick, R. Mask R-CNN. In Proceedings of the IEEE International Conference on Computer Vision, Venice, Italy, 22–29 October 2017; pp. 2961–2969. 25. Lisboa, P.; Saralajew, S.; Vellido, A.; Fernández-Domenech, R.; Villmann, T. The Coming of Age of Interpretable and Explainable Machine Learning Models. Neurocomputing 2023, 535, 25–39. 26. Mukherjee, T.; Pournik, O.; Lim Choi Keung, S.; Arvanitis, T. Clinical decision support systems for brain tumour diagnosis and prognosis: A systematic review. Cancers 2023, 15, 3523. https://doi.org/10.3390/cancers15133523. Cancers 2024, 16, 300 48 of 55 27. Bertsimas, D.; Wiberg, H. Machine Learning in Oncology: Methods, applications, and challenges. JCO Clin. Cancer Inform. 2020, 4, 885–894. 28. Jha, A.; Mithun, S.; Sherkhane, U.B.; Jaiswar, V.; Shi, Z.; Kalendralis, P.; Kulkarni, C.; Dinesh, M.S.; Rajamenakshi, R.; Sunder, G.; et al. Implementation of big imaging data pipeline adhering to FAIR principles for Federated Machine Learning in Oncology. IEEE Trans. Radiat. Plasma Med. Sci. 2022, 6, 207–213. 29. Su, X.; Chen, N.; Sun, H.; Liu, Y.; Yang, X.; Wang, W.; Zhang, S.; Tan, Q.; Su, J.; Gong, Q.; et al. Automated Machine Learning based on radiomics features predicts H3 K27M mutation in midline gliomas of the brain. Neuro-Oncol. 2020, 22, 393–401. 30. Mocioiu, V.; Pedrosa de Barros, N.; Ortega-Martorell, S.; Slotboom, J.; Knecht, U.; Arús, C.; Vellido, A.; Julià-Sapé, M. A Machine Learning pipeline for supporting differentiation of glioblastomas from single brain metastases. In Proceedings of the ESANN 2016, European Symposium on Artificial Neural Networks, Computational Intelligence and Machine Learning (ESANN) Bruges (Belgium), Bruges, Belgium, 5–7 October 2016; pp. 247–252. 31. Pitarch, C.; Ribas, V.; Vellido, A. AI-Based Glioma Grading for a Trustworthy Diagnosis: An Analytical Pipeline for Improved Reliability. Cancers 2023, 15, 3369. https://doi.org/10.3390/cancers15133369. 32. Tabassum, M.; Suman, A.; Suero Molina, E.; Pan, E.; Di Ieva, A.; Liu, S. Radiomics and Machine Learning in Brain Tumors and Their Habitat: A Systematic Review. Cancers 2023, 15, 3845. 33. Griethuysen, J.; Fedorov, A.; Parmar, C.; Hosny, A.; Aucoin, N.; Narayan, V.; Beets-Tan, R.; Fillon-Robin, J.; Pieper, S.; Aerts, H. Clinical Decision Support Systems for Brain Tumour Diagnosis and Prognosis: A Systematic Review. Cancer Res. 2017, 77, e104–e107. https://doi.org/10.1158/0008-5472.CAN-17-0339. 34. Hyvärinen, A.; Oja, E. Independent component analysis: Algorithms and applications. Neural Netw. 2000, 13, 411–430. https://doi.org/10.1016/S0893-6080(00)00026-5. 35. Lee, J.; Zhao, Q.; Kent, M.; Platt, S. Tumor Segmentation using temporal Independent Component Analysis for DCE-MRI. BioRxiv 2022, https://doi.org/10.1101/2022.12.16.520830. 36. Chakhoyan, A.; Raymond, C.; Chen, J.; Goldman, J.; Yao, J.; Kaprealian, T.; Pouratian, N.; Ellingson, B. Probabilistic independent component analysis of dynamic susceptibility contrast perfusion MRI in metastatic brain tumors. Cancer Imaging 2019, 19, 14. 37. Lee, D.; Seung, H. Learning the parts of objects by non-negative matrix factorization. Nature 1999, 401, 788–791. 38. Ortega-Martorell, S.; Lisboa, P.; Vellido, A.; Julià-Sapé, M.; Arús, C. Non-negative matrix factorisation methods for the spectral decomposition of MRS data from human brain tumours. BMC Bioinform. 2012, 13, 38. 39. Ungan, G.; Arús, C.; Vellido, A.; Julià-Sapé, M. A Comparison of Non-Negative Matrix Underapproximation Methods for the Decomposition of Magnetic Resonance Spectroscopy Data from Human Brain Tumors. NMR Biomed. 2023, 36, e5020. 40. Sauwen, N.; Acou, M.; Van Cauter, S.; Sima, D.M.; Veraart, J.; Maes, F.; Himmelreich, U.; Achten, E.; Van Huffel, S. Comparison of unsupervised classification methods for brain tumor segmentation using multi-parametric MRI. Neuroimage Clin. 2016, 12, 753–764. https://doi.org/10.1016/j.nicl.2016.09.021. 41. Ashtari, P.; Sima, D.; De Lathauwer, L.; Sappey-Marinier, D.; Maes, F.; Van Huffel, S. Factorizer: A scalable interpretable approach to context modeling for medical image segmentation. Med. Image Anal. 2023, 84, 102706. 42. Dosovitskiy, A.; Beyer, L.; Kolesnikov, A.; Weissenborn, D.; Zhai, X.; Unterthiner, T.; Dehghani, M.; Minderer, M.; Heigold, G.; Gelly, S.; et al. An image is worth 16 × 16 words: Transformers for image recognition at scale. arXiv 2020, arXiv:2010.11929. 43. Lundervold, A.S.; Lundervold, A. An overview of deep learning in medical imaging focusing on MRI. Z. Med. Phys. 2019, 29, 102–127. https://doi.org/10.1016/J.ZEMEDI.2018.11.002. 44. Cai, L.; Gao, J.; Zhao, D. A review of the application of deep learning in medical image classification and segmentation. Ann. Transl. Med. 2020, 8, 713–713. https://doi.org/10.21037/ATM.2020.02.44. 45. Chen, X.; Wang, X.; Zhang, K.; Fung, K.M.; Thai, T.C.; Moore, K.; Mannel, R.S.; Liu, H.; Zheng, B.; Qiu, Y. Recent advances and clinical applications of deep learning in medical image analysis. Med. Image Anal. 2022, 79, 102444. https://doi.org/10.1016/J. MEDIA.2022.102444. 46. Krizhevsky, A.; Sutskever, I.; Hinton, G.E. ImageNet classification with deep convolutional neural networks. In Proceedings of the Advances in Neural Information Processing Systems; Pereira, F., Burges, C., Bottou, L., Weinberger, K., Eds.; Curran Associates, Inc.: Red Hook, NY, USA, 2012; Volume 25. 47. Szegedy, C.; Liu, W.; Jia, Y.; Sermanet, P.; Reed, S.; Anguelov, D.; Erhan, D.; Vanhoucke, V.; Rabinovich, A. Going deeper with convolutions. In Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition, Nashville, TN, USA, 20–25 June 2014; pp. 1–9. https://doi.org/10.1109/CVPR.2015.7298594. 48. Simonyan, K.; Zisserman, A. Very Deep Convolutional Networks for Large-Scale Image Recognition. In Proceedings of the 3rd International Conference on Learning Representations, ICLR 2015—Conference Track Proceedings, San Diego, CA, USA, 7–9 May 2014. 49. He, K.; Zhang, X.; Ren, S.; Sun, J. Deep residual learning for image recognition. In Proceedings of the IEEE Com- puter Society Conference on Computer Vision and Pattern Recognition, Boston, MA, USA, 7–12 June 2015; pp. 770–778. https://doi.org/10.1109/CVPR.2016.90. 50. Huang, G.; Liu, Z.; Van Der Maaten, L.; Weinberger, K.Q. Densely connected convolutional networks. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, Honolulu, HI, USA, 21–26 July 2017; pp. 4700–4708. 51. Tan, M.; Le, Q. EfficientNet: Rethinking model scaling for convolutional neural networks. In Proceedings of the 36th International Conference on Machine Learning, Long Beach, CA, USA 9–15 June 2019; Volume 97, pp. 6105–6114. Cancers 2024, 16, 300 49 of 55 52. Goodfellow, I.J.; Pouget-Abadie, J.; Mirza, M.; Xu, B.; Warde-Farley, D.; Ozair, S.; Courville, A.; Bengio, Y. Generative Adversarial Networks. arXiv 2014, arXiv:1406.2661. 53. Deng, J.; Dong, W.; Socher, R.; Li, L.J.; Li, K.; Fei-Fei, L. Imagenet: A large-scale hierarchical image database. In Proceedings of the 2009 IEEE Conference on Computer Vision and Pattern Recognition, Miami, FL, USA, 20–25 June 2009; pp. 248–255. 54. Lin, T.Y.; Maire, M.; Belongie, S.; Hays, J.; Perona, P.; Ramanan, D.; Dollár, P.; Zitnick, C.L. Microsoft COCO: Common objects in context. In Proceedings of the Computer Vision—ECCV 2014; Fleet, D., Pajdla, T., Schiele, B., Tuytelaars, T., Eds.; Springer: Cham, Switzerland, 2014; pp. 740–755. 55. Yang, J.; Huang, X.; He, Y.; Xu, J.; Yang, C.; Xu, G.; Ni, B. Reinventing 2D Convolutions for 3D Images. IEEE J. Biomed. Health Inform. 2021, 25, 3009–3018. https://doi.org/10.1109/JBHI.2021.3049452. 56. Banerjee, S.; Mitra, S.; Masulli, F.; Rovetta, S. Glioma classification using deep radiomics. SN Comput. Sci. 2020, 1, 209. https://doi.org/10.1007/s42979-020-00214-y. 57. Ding, J.; Zhao, R.; Qiu, Q.; Chen, J.; Duan, J.; Cao, X.; Yin, Y. Developing and validating a deep learning and radiomic model for glioma grading using multiplanar reconstructed magnetic resonance contrast-enhanced T1-weighted imaging: A robust, multi-institutional study. Quant. Imaging Med. Surg. 2022, 12, 1517. https://doi.org/10.21037%2fqims-21-722. 58. Zhuge, Y.; Ning, H.; Mathen, P.; Cheng, J.Y.; Krauze, A.V.; Camphausen, K.; Miller, R.W. Automated glioma grading on conven- tional MRI images using deep convolutional neural networks. Med. Phys. 2020, 47, 3044–3053. https://doi.org/10.1002/mp.14168. 59. Chatterjee, S.; Nizamani, F.A.; Nürnberger, A.; Speck, O. Classification of brain tumours in MR images using deep spatiospatial models. Sci. Rep. 2022, 12, 1505. https://doi.org/10.1038/s41598-022-05572-6. 60. Baheti, B.; Pati, S.; Menze, B.; Bakas, S. Leveraging 2D Deep Learning ImageNet-trained Models for Native 3D Medical Image Analysis. LNCS 2023, 13769, 68–79. https://doi.org/10.1007/978-3-031-33842-7_6. 61. Brain Tumor Segmentation (BraTS) Challenge. Available online: http://www.braintumorsegmentation.org/ (accessed on 10 June 2023). 62. Computational Precision Medicine: Radiology-Pathology Challenge on Brain Tumor Classification 2019 (CPM-RadPath). Available online: https://www.med.upenn.edu/cbica/cpm-rad-path-2019/ (accessed on 30 August 2023). 63. Figshare Brain Tumor Dataset. Available online: https://figshare.com/articles/dataset/brain_tumor_dataset/1512427 (accessed on 1 June 2023). 64. IXI Dataset. Available online: https://brain-development.org/ixi-dataset/ (accessed on 10 June 2023). 65. Hamada, A. Br35H Brain Tumor Detection 2020 Dataset. Available online: https://www.kaggle.com/datasets/ahmedhamada0 /brain-tumor-detection (accessed on 1 June 2023). 66. Bhuvaji, S.; Kadam, A.; Bhumkar, P.; Dedge, S. Brain Tumor Classification (MRI). Available online: https://www.kaggle.com/ datasets/sartajbhuvaji/brain-tumor-classification-mri (accessed on 1 June 2023). 67. Chakrabarty, N. Brain MRI Images Dataset for Brain Tumor Detection, Kaggle, 2019. Available online: https://www.kaggle.com/ datasets/navoneel/brain-mri-images-for-brain-tumor-detection (accessed on 1 June 2023). 68. Radiopaedia. Available online: https://radiopaedia.org/cases/system/central-nervous-system (accessed on 1 June 2023). 69. Scarpace, L.; Flanders, A.E.; Jain, R.; Mikkelsen, T.; Andrews, D.W. Data From REMBRANDT [Data set]. The Cancer Imaging Archive. 2019. Available online: https://doi.org/10.7937/k9/tcia.2015.588ozuzb (accessed on 20 April 2023). 70. Scarpace, L.; Mikkelsen, T.; Cha, S.; Rao, S.; Tekchandani, S.; Gutman, D.; Saltz, J.H.; Erickson, B.J.; Pedano, N.; Flanders, A.E.; et al. The Cancer Genome Atlas Glioblastoma Multiforme Collection (TCGA-GBM) (Version 4) [Data set]. The Cancer Imaging Archive. 2016. Available online: https://doi.org/10.7937/k9/tcia.2016.rnyfuye9 (accessed on 4 March 2023). 71. Pedano, N.; Flanders, A.E.; Scarpace, L.; Mikkelsen, T.; Eschbacher, J.M.; Hermes, B.; Sisneros, V.; Barnholtz-Sloan, J.; Ostrom, Q. The Cancer Genome Atlas Low Grade Glioma Collection (TCGA-LGG) (Version 3) [Data set]. The Cancer Imaging Archive. 2016. Available online: https://doi.org/10.7937/k9/tcia.2016.l4ltd3tk (accessed on 5 March 2023). 72. Upadhyay, N.; Waldman, A.D. Conventional MRI evaluation of gliomas. Br. J. Radiol. 2011, 84, S107. https://doi.org/10.1259/ BJR/65711810. 73. Ge, C.; Qu, Q.; Gu, I.Y.H.; Store Jakola, A. 3D Multi-scale convolutional networks for glioma grading using MR images. In Proceedings of the 2018 25th IEEE International Conference on Image Processing (ICIP), Athens, Greece, 7–10 October 2018; pp. 141–145. https://doi.org/10.1109/ICIP.2018.8451682. 74. Pereira, S.; Meier, R.; Alves, V.; Reyes, M.; Silva, C.A. Automatic brain tumor grading from MRI data using convolutional neural networks and quality assessment. Lect. Notes Comput. Sci. 2018, 11038 LNCS, 106–114. https://doi.org/10.1007/978-3-030-02628-8_12. 75. Tandel, G.S.; Tiwari, A.; Kakde, O. Performance enhancement of MRI-based brain tumor classification using suitable segmentation method and deep learning-based ensemble algorithm. Biomed. Signal Process. Control. 2022, 78, 104018. 76. Deepak, S.; Ameer, P.M. Brain tumor classification using deep CNN features via transfer learning. Comput. Biol. Med. 2019, 111, 103345. https://doi.org/10.1016/j.compbiomed.2019.103345. 77. Swati, Z.N.K.; Zhao, Q.; Kabir, M.; Ali, F.; Ali, Z.; Ahmed, S.; Lu, J. Brain tumor classification for MR images using transfer learning and fine-tuning. Comput. Med. Imaging Graph. 2019, 75, 34–46. https://doi.org/10.1016/j.compmedimag.2019.05.001. 78. Alaraimi, S.; Okedu, K.E.; Tianfield, H.; Holden, R.; Uthmani, O. Transfer learning networks with skip connections for classification of brain tumors. Int. J. Imaging Syst. Technol. 2021, 31, 1564–1582. https://doi.org/10.1002/ima.22546. 79. Hao, R.; Namdar, K.; Liu, L.; Khalvati, F. A Transfer Learning—Based Active Learning Framework for Brain Tumor Classification. Front. Artif. Intell. 2021, 4, 61. https://doi.org/10.3389/frai.2021.635766. Cancers 2024, 16, 300 50 of 55 80. Toğaçar, M.; Ergen, B.; Cömert, Z. Tumor type detection in brain MR images of the deep model developed using hypercolumn technique, attention modules, and residual blocks. Med. Biol. Eng. Comput. 2021, 59, 57–70. https://doi.org/10.1007/S11517-020 -02290-X. 81. Amou, M.A.; Xia, K.; Kamhi, S.; Mouhafid, M. A Novel MRI Diagnosis Method for Brain Tumor Classification Based on CNN and Bayesian Optimization. Healthcare 2022, 10, 494. https://doi.org/10.3390/healthcare10030494. 82. Aurna, N.F.; Yousuf, M.A.; Taher, K.A.; Azad, A.K.M.; Moni, M.A. A classification of MRI brain tumor based on two stage feature level ensemble of deep CNN models. Comput. Biol. Med. 2022, 146, 105539. https://doi.org/10.1016/j.compbiomed.2022.105539. 83. Tripathi, P.C.; Bag, S. A computer-aided grading of glioma tumor using deep residual networks fusion. Comput. Methods Programs Biomed. 2022, 215, 106597. https://doi.org/10.1016/j.cmpb.2021.106597. 84. Deepa, S.; Janet, J.; Sumathi, S.; Ananth, J.P. Hybrid Optimization Algorithm Enabled Deep Learning Approach Brain Tumor Segmentation and Classification Using MRI. J. Digit. Imaging 2023, 36, 1–22. https://doi.org/10.1007/S10278-022-00752-2. 85. Ghassemi, N.; Shoeibi, A.; Rouhani, M. Deep neural network with generative adversarial networks pre-training for brain tumor classification based on MR images. Biomed. Signal Process. Control. 2020, 57, 101678. https://doi.org/10.1016/j.bspc.2019.101678. 86. Ayadi, W.; Elhamzi, W.; Charfi, I.; Atri, M. Deep CNN for brain tumor classification. Neural Process. Lett. 2021, 53, 671–700. https://doi.org/10.1007/s11063-020-10398-2. 87. Khan, A.R.; Khan, S.; Harouni, M.; Abbasi, R.; Iqbal, S.; Mehmood, Z. Brain tumor segmentation using K-means clus- tering and deep learning with synthetic data augmentation for classification. Microsc. Res. Tech. 2021, 84, 1389–1399. https://doi.org/10.1002/jemt.23694. 88. Kumar, R.L.; Kakarla, J.; Isunuri, B.V.; Singh, M. Multi-class brain tumor classification using residual network and global average pooling. Multimed. Tools Appl. 2021, 80, 13429–13438. https://doi.org/10.1007/s11042-020-10335-4. 89. Ahmad, B.; Sun, J.; You, Q.; Palade, V.; Mao, Z. Brain Tumor Classification Using a Combination of Variational Autoencoders and Generative Adversarial Networks. Biomedicines 2022, 10, 223. https://doi.org/10.3390/biomedicines10020223. 90. Haq, A.U.; Li, J.P.; Kumar, R.; Ali, Z.; Khan, I.; Uddin, M.I.; Agbley, B.L.Y. MCNN: A multi-level CNN model for the classification of brain tumors in IoT-healthcare system. J. Ambient. Intell. Humaniz. Comput. 2022, 14, 4695–4706. https://doi.org/10.1007/s126 52-022-04373-z. 91. Anand, V.; Gupta, S.; Gupta, D.; Gulzar, Y.; Xin, Q.; Juneja, S.; Shah, A.; Shaikh, A. Weighted Average Ensemble Deep Learning Model for Stratification of Brain Tumor in MRI Images. Diagnostics 2023, 13, 1320. https://doi.org/10.3390/DIAGNOSTICS13071320. 92. Krishnapriya, S.; Karuna, Y. Pre-trained deep learning models for brain MRI image classification. Front. Hum. Neurosci. 2023, 17, 1150120. https://doi.org/10.3389/FNHUM.2023.1150120. 93. Ge, C.; Gu, I.Y.H.; Jakola, A.S.; Yang, J. Deep semi-supervised learning for brain tumor classification. BMC Med. Imaging 2020, 20, 1–11. https://doi.org/10.1186/s12880-020-00485-0. 94. Gab Allah, A.M.; Sarhan, A.M.; Elshennawy, N.M. Classification of brain MRI tumor images based on deep learning PGGAN augmentation. Diagnostics 2021, 11, 2343. https://doi.org/10.3390/diagnostics11122343. 95. Gupta, R.K.; Bharti, S.; Kunhare, N.; Sahu, Y.; Pathik, N. Brain Tumor Detection and Classification Using Cycle Generative Adversarial Networks. Interdiscip. Sci. Comput. Life Sci. 2022, 14, 485–502. https://doi.org/10.1007/s12539-022-00502-6. 96. Toğaçar, M.; Cömert, Z.; Ergen, B. Classification of brain MRI using hyper column technique with convolutional neural network and feature selection method. Expert Syst. Appl. 2020, 149, 113274. https://doi.org/10.1016/j.eswa.2020.113274. 97. Pei, L.; Hsu, W.W.; Chiang, L.A.; Guo, J.M.; Iftekharuddin, K.M.; Colen, R. A Hybrid Convolutional Neural Network Based- Method for Brain Tumor Classification Using mMRI and WSI. Lect. Notes Comput. Sci. 2021, 12659 LNCS, 487–496. 98. Dang, K.; Vo, T.; Ngo, L.; Ha, H. A deep learning framework integrating MRI image preprocessing methods for brain tumor segmentation and classification. IBRO Neurosci. Rep. 2022, 13, 523–532. https://doi.org/10.1016/J.IBNEUR.2022.10.014. 99. Senan, E.M.; Jadhav, M.E.; Rassem, T.H.; Aljaloud, A.S.; Mohammed, B.A.; Al-Mekhlafi, Z.G. Early Diagnosis of Brain Tumour MRI Images Using Hybrid Techniques between Deep and Machine Learning. Comput. Math. Methods Med. 2022, 2022, 8330833. https://doi.org/10.1155/2022/8330833. 100. Ge, C.; Gu, I.Y.H.; Jakola, A.S.; Yang, J. Deep learning and multi-sensor fusion for glioma classification using multistream 2D convolutional networks. In Proceedings of the Annual International Conference of the IEEE Engineering in Medicine and Biology Society, EMBS, Sarawak, Malaysia, 18-21 July 2018; pp. 5894–5897. https://doi.org/10.1109/EMBC.2018.8513556. 101. Yang, Y.; Yan, L.F.; Zhang, X.; Han, Y.; Nan, H.Y.; Hu, Y.C.; Hu, B.; Yan, S.L.; Zhang, J.; Cheng, D.L.; et al. Glioma grading on conventional MR images: A deep learning study with transfer learning. Front. Neurosci. 2018, 12, 804. https://doi.org/10.338 9/fnins.2018.00804. 102. Lo, C.M.; Chen, Y.C.; Weng, R.C.; Hsieh, K.L.C. Intelligent Glioma Grading Based on Deep Transfer Learning of MRI Radiomic Features. Appl. Sci. 2019, 9, 4926. https://doi.org/10.3390/app9224926. 103. Zahoor, M.M.; Qureshi, S.A.; Bibi, S.; Khan, S.H.; Khan, A.; Ghafoor, U.; Bhutta, M.R. A New Deep Hybrid Boosted and Ensemble Learning-Based Brain Tumor Analysis Using MRI. Sensors 2022, 22, 2726. https://doi.org/10.3390/s22072726. 104. Rehman, A.; Naz, S.; Razzak, M.I.; Akram, F.; Imran, M. A Deep Learning-Based Framework for Automatic Brain Tumors Classification Using Transfer Learning. Circuits Syst. Signal Process. 2020, 39, 757–775. https://doi.org/10.1007/s00034-019-01246-3. 105. Tandel, G.S.; Balestrieri, A.; Jujaray, T.; Khanna, N.N.; Saba, L.; Suri, J.S. Multiclass magnetic resonance imaging brain tumor classification using artificial intelligence paradigm. Comput. Biol. Med. 2020, 122, 103804. https://doi.org/10.1016/j.compbiomed. 2020.103804. Cancers 2024, 16, 300 51 of 55 106. Gutta, S.; Acharya, J.; Shiroishi, M.S.; Hwang, D.; Nayak, K.S. Improved Glioma Grading Using Deep Convolutional Neural Networks. Am. J. Neuroradiol. 2021, 42, 233–239. https://doi.org/10.3174/AJNR.A6882. 107. Tandel, G.S.; Tiwari, A.; Kakde, O.G. Performance optimisation of deep learning models using majority voting algorithm for brain tumour classification. Comput. Biol. Med. 2021, 135, 104564. https://doi.org/10.1016/j.compbiomed.2021.104564. 108. Kazemi, A.; Shiri, M.E.; Sheikhahmadi, A.; Khodamoradi, M. Classifying tumor brain images using parallel deep learning algorithms. Comput. Biol. Med. 2022, 148, 10–4825. https://doi.org/10.1016/j.compbiomed.2022.105775. 109. Saravanan, S.; Kumar, V.V.; Sarveshwaran, V.; Indirajithu, A.; Elangovan, D.; Allayear, S.M. Computational and Mathematical Methods in Medicine Glioma Brain Tumor Detection and Classification Using Convolutional Neural Network. Comput. Math. Methods Med. 2022, 2022, 4380901. https://doi.org/10.1155/2022/4380901. 110. Athisayamani, S.; Antonyswamy, R.S.; Sarveshwaran, V.; Almeshari, M.; Alzamil, Y.; Ravi, V. Feature Extraction Using a Residual Deep Convolutional Neural Network (ResNet-152) and Optimized Feature Dimension Reduction for MRI Brain Tumor Classification. Diagnostics 2023, 13, 668. https://doi.org/10.3390/DIAGNOSTICS13040668. 111. Bairagi, V.K.; Gumaste, P.P.; Rajput, S.H.; Chethan, K.S. Automatic brain tumor detection using CNN transfer learning approach. Med. Biol. Eng. Comput. 2023, 1, 1–16. https://doi.org/10.1007/S11517-023-02820-3. 112. Gao, P.; Shan, W.; Guo, Y.; Wang, Y.; Sun, R.; Cai, J.; Li, H.; Chan, W.S.; Liu, P.; Yi, L.; et al. Development and Validation of a Deep Learning Model for Brain Tumor Diagnosis and Classification Using Magnetic Resonance Imaging. JAMA Netw. Open 2022, 5, e2225608. https://doi.org/10.1001/jamanetworkopen.2022.25608. 113. Jeong, S.W.; Cho, H.H.; Lee, S.; Park, H. Robust multimodal fusion network using adversarial learning for brain tumor grading. Comput. Methods Programs Biomed. 2022, 226, 107165. https://doi.org/10.1016/j.cmpb.2022.107165. 114. Maqsood, S.; Damaševičius, R.; Maskeliūnas, R. Multi-Modal Brain Tumor Detection Using Deep Neural Network and Multiclass SVM. Medicina 2022, 58, 1090. https://doi.org/10.3390/medicina58081090. 115. Xiong, D.; Ren, X.; Huang, W.; Wang, R.; Ma, L.; Gan, T.; Ai, K.; Wen, T.; Li, Y.; Wang, P.; et al. Noninvasive Classification of Glioma Subtypes Using Multiparametric MRI to Improve Deep Learning. Diagnostics 2022, 12, 3063. https://doi.org/10.3390/ diagnostics12123063. 116. Hossain, S.; Chakrabarty, A.; Gadekallu, T.R.; Alazab, M.; Piran, M.J. Vision Transformers, Ensemble Model, and Transfer Learning Leveraging Explainable AI for Brain Tumor Detection and Classification. IEEE J. Biomed. Health Inform. 2023. https://doi.org/10.1109/jbhi.2023.3266614. 117. van der Voort, S.R.; Incekara, F.; Wijnenga, M.M.; Kapsas, G.; Gahrmann, R.; Schouten, J.W.; Nandoe Tewarie, R.; Lycklama, G.J.; De Witt Hamer, P.C.; Eijgelaar, R.S.; et al. Combined molecular subtyping, grading, and segmentation of glioma using multi-task deep learning. Neuro-Oncol. 2023, 25, 279–289. https://doi.org/10.1093/neuonc/noac166. 118. Xu, C.; Peng, Y.; Zhu, W.; Chen, Z.; Li, J.; Tan, W.; Zhang, Z.; Chen, X. An automated approach for predicting glioma grade and survival of LGG patients using CNN and radiomics. Front. Oncol. 2022, 12, 969907. https://doi.org/10.3389/fonc.2022.969907. 119. Kibriya, H.; Amin, R.; Kim, J.; Nawaz, M.; Gantassi, R. A Novel Approach for Brain Tumor Classification Using an Ensemble of Deep and Hand-Crafted Features. Sensors 2023, 23, 4693. https://doi.org/10.3390/S23104693. 120. Kutlu, H.; Avcı, E. A Novel Method for Classifying Liver and Brain Tumors Using Convolutional Neural Networks, Discrete Wavelet Transform and Long Short-Term Memory Networks. Sensors 2019, 19, 1992. https://doi.org/10.3390/S19091992. 121. Almalki, Y.E.; Ali, M.U.; Kallu, K.D.; Masud, M.; Zafar, A.; Alduraibi, S.K.; Irfan, M.; Basha, M.A.A.; Alshamrani, H.A.; Alduraibi, A.K.; et al. Isolated Convolutional-Neural-Network-Based Deep-Feature Extraction for Brain Tumor Classification Using Shallow Classifier. Diagnostics 2022, 12, 1793. https://doi.org/10.3390/DIAGNOSTICS12081793. 122. Kibriya, H.; Amin, R.; Alshehri, A.H.; Masood, M.; Alshamrani, S.S.; Alshehri, A. A Novel and Effective Brain Tumor Classification Model Using Deep Feature Fusion and Famous Machine Learning Classifiers. Comput. Intell. Neurosci. 2022, 2022, 7897669. https://doi.org/10.1155/2022/7897669. 123. Shirehjini, O.F.; Mofrad, F.B.; Shahmohammadi, M.; Karami, F. Grading of gliomas using transfer learning on MRI images. Magn. Reson. Mater. Phys. Biol. Med. 2023, 36, 43–53. https://doi.org/10.1007/S10334-022-01046-Y. 124. Rajinikanth, V.; Kadry, S.; Damaševičius, R.; Sujitha, R.A.; Balaji, G.; Mohammed, M.A. Glioma/glioblastoma detection in brain MRI using pre-trained deep-learning scheme. In Proceedings of the 2022 Third International Conference on Intelligent Computing Instrumentation and Control Technologies (ICICICT), Guangzhou, China, 12–14 August 2022; pp. 987–990. 125. Rasool, M.; Ismail, N.; Boulila, W.; Ammar, A.; Samma, H.; Yafooz, W.S.; Emara, A.H. A Hybrid Deep Learning Model for Brain Tumour Classification. Entropy 2022, 24, 799. https://doi.org/10.3390/E24060799. 126. Sekhar, A.; Biswas, S.; Hazra, R.; Sunaniya, A.K.; Mukherjee, A.; Yang, L. Brain Tumor Classification Using Fine-Tuned GoogLeNet Features and Machine Learning Algorithms: IoMT Enabled CAD System. IEEE J. Biomed. Health Inform. 2022, 26, 983–991. https://doi.org/10.1109/jbhi.2021.3100758. 127. AlTahhan, F.E.; Khouqeer, G.A.; Saadi, S.; Elgarayhi, A.; Sallah, M. Refined Automatic Brain Tumor Classification Using Hybrid Convolutional Neural Networks for MRI Scans. Diagnostics 2023, 13, 864. https://doi.org/10.3390/diagnostics13050864. 128. Kumar, S.; Choudhary, S.; Jain, A.; Singh, K.; Ahmadian, A.; Mohd.; Bajuri, Y. Brain Tumor Classification Using Deep Neural Network and Transfer Learning. Brain Topogr. 2023, 36, 305–318. https://doi.org/10.1007/S10548-023-00953-0. 129. Ma, X.; Jia, F. Brain tumor classification with multimodal MR and pathology images. Lect. Notes Comput. Sci. 2020, 11993 LNCS, 343–352. https://doi.org/10.1007/978-3-030-46643-5_34. Cancers 2024, 16, 300 52 of 55 130. Yin, B.; Cheng, H.; Wang, F.; Wang, Z. Brain tumor classification based on MRI images and noise reduced pathology images. In Proceedings of the Brainlesion: Glioma, Multiple Sclerosis, Stroke and Traumatic Brain Injuries: 6th International Workshop, BrainLes 2020, Lima, Perú, 4–8 October 2020 ; Springer: Berlin/Heidelberg, Germany, 2021; pp. 465–474. 131. Hsu, W.W.; Guo, J.M.; Pei, L.; Chiang, L.A.; Li, Y.F.; Hsiao, J.C.; Colen, R.; Liu, P. A weakly supervised deep learning-based method for glioma subtype classification using WSI and mpMRIs. Sci. Rep. 2022, 12, 1–12. https://doi.org/10.1038/s41598-022-09985-1. 132. Wang, X.; Wang, R.; Yang, S.; Zhang, J.; Wang, M.; Zhong, D.; Zhang, J.; Han, X. Combining Radiology and Pathology for Automatic Glioma Classification. Front. Bioeng. Biotechnol. 2022, 10, 841958. https://doi.org/10.3389/fbioe.2022.841958. 133. Kang, J.; Ullah, Z.; Gwak, J. MRI-Based Brain Tumor Classification Using Ensemble of Deep Features and Machine Learning Classifiers. Sensors 2021, 21, 2222. https://doi.org/10.3390/s21062222. 134. Noreen, N.; Palaniappan, S.; Qayyum, A.; Ahmad, I.; Alassafi, M.O. Brain Tumor Classification Based on Fine-Tuned Models and the Ensemble Method. Comput. Mater. Contin. 2021, 67, 3967–3982. https://doi.org/10.32604/cmc.2021.014158. 135. Coupet, M.; Urruty, T.; Leelanupab, T.; Naudin, M.; Bourdon, P.; Maloigne, C.F.; Guillevin, R. A multi-sequences MRI deep framework study applied to glioma classfication. Multimed. Tools Appl. 2022, 81, 13563–13591. 136. Tummala, S.; Kadry, S.; Bukhari, S.A.C.; Rauf, H.T. Classification of Brain Tumor from Magnetic Resonance Imaging Using Vision Transformers Ensembling. Curr. Oncol. 2022, 29, 7498–7511. https://doi.org/10.3390/curroncol29100590. 137. Al-Zoghby, A.M.; Al-Awadly, E.M.K.; Moawad, A.; Yehia, N.; Ebada, A.I. Dual Deep CNN for Tumor Brain Classification. Diagnostics 2023, 13, 2050. https://doi.org/10.3390/DIAGNOSTICS13122050. 138. Asif, S.; Zhao, M.; Chen, X.; Zhu, Y. BMRI-NET: A Deep Stacked Ensemble Model for Multi-class Brain Tumor Classification from MRI Images. Interdiscip. Sci. Comput. Life Sci. 2023, 1, 1–16. https://doi.org/10.1007/S12539-023-00571-1. 139. Tandel, G.S.; Tiwari, A.; Kakde, O.G.; Gupta, N.; Saba, L.; Suri, J.S. Role of Ensemble Deep Learning for Brain Tumor Classification in Multiple Magnetic Resonance Imaging Sequence Data. Diagnostics 2023, 13, 481. https://doi.org/10.3390/diagnostics13030481. 140. Decuyper, M.; Bonte, S.; Deblaere, K.; Holen, R.V. Automated MRI based pipeline for segmentation and prediction of grade, IDH mutation and 1p19q co-deletion in glioma. Comput. Med. Imaging Graph. 2021, 88, 101831. https://doi.org/10.1016/j. compmedimag.2020.101831. 141. Tripathi, P.C.; Bag, S. An attention-guided CNN framework for segmentation and grading of glioma using 3D MRI scans. IEEE/ACM Trans. Comput. Biol. Bioinform. 2022, 3, 1890–1904. https://doi.org/10.1109/TCBB.2022.3220902. 142. Mzoughi, H.; Njeh, I.; Wali, A.; Slima, M.B.; BenHamida, A.; Mhiri, C.; Mahfoudhe, K.B. Deep Multi-Scale 3D Convolutional Neural Network (CNN) for MRI Gliomas Brain Tumor Classification. J. Digit. Imaging 2020, 33, 903–915. https://doi.org/10.100 7/s10278-020-00347-9. 143. Pei, L.; Vidyaratne, L.; Rahman, M.M.; Iftekharuddin, K.M. Context aware deep learning for brain tumor segmentation, subtype classification, and survival prediction using radiology images. Sci. Rep. 2020, 10, 19726. https://doi.org/10.1038/s41598-020-74419-9. 144. Chakrabarty, S.; Sotiras, A.; Milchenko, M.; Lamontagne, P.; Hileman, M.; Marcus, D. MRI-based identification and classification of major intracranial tumor types by using a 3D convolutional neural network: A retrospective multi-institutional analysis. Radiol. Artif. Intell. 2021, 3, e200301. https://doi.org/10.1148/ryai.2021200301. 145. Yamashiro, H.; Teramoto, A.; Saito, K.; Fujita, H. Development of a Fully Automated Glioma-Grading Pipeline Using Post- Contrast T1-Weighted Images Combined with Cloud-Based 3D Convolutional Neural Network. Appl. Sci. 2021, 11, 5118. https://doi.org/10.3390/APP11115118. 146. Danilov, G.; Korolev, V.; Shifrin, M.; Ilyushin, E.; Maloyan, N.; Saada, D.; Ishankulov, T.; Afandiev, R.; Shevchenko, A.; Konakova, T.; et al. Noninvasive Glioma Grading with Deep Learning: A Pilot Study. Stud. Health Technol. Inform. 2022, 290, 675–678. https://doi.org/10.3233/SHTI220163. 147. Samee, N.A.; Ahmad, T.; Mahmoud, N.F.; Atteia, G.; Abdallah, H.A.; Rizwan, A. Clinical Decision Support Framework for Segmentation and Classification of Brain Tumor MRIs Using a U-Net and DCNN Cascaded Learning Algorithm. Healthcare 2022, 10, 2340. https://doi.org/10.3390/HEALTHCARE10122340. 148. Hussain, S.; Haider, S.; Maqsood, S.; Damaševičius, R.; Maskeliūnas, R.; Khan, M. ETISTP: An Enhanced Model for Brain Tumor Identification and Survival Time Prediction. Diagnostics 2023, 13, 1456. https://doi.org/10.3390/DIAGNOSTICS13081456. 149. Rui, W.; Zhang, S.; Shi, H.; Sheng, Y.; Zhu, F.; Yao, Y.; Chen, X.; Cheng, H.; Zhang, Y.; Aili, A.; et al. Deep Learning-Assisted Quantitative Susceptibility Mapping as a Tool for Grading and Molecular Subtyping of Gliomas. Phenomics 2023, 3, 243–254. https://doi.org/10.1007/S43657-022-00087-6. 150. Guo, S.; Wang, L.; Chen, Q.; Wang, L.; Zhang, J.; Zhu, Y. Multimodal MRI Image Decision Fusion-Based Network for Glioma Classification. Front. Oncol. 2022, 12, 819673. https://doi.org/10.3389/fonc.2022.819673. 151. Díaz-Pernas, F.J.; Martínez-Zarzuela, M.; Antón-Rodríguez, M.; González-Ortega, D. A deep learning approach for brain tumor classification and segmentation using a multiscale convolutional neural network. Healthcare 2021, 9, 153. https: //doi.org/10.3390/healthcare9020153. 152. Gilanie, G.; Bajwa, U.I.; Waraich, M.M.; Anwar, M.W. Risk-free WHO grading of astrocytoma using convolutional neural networks from MRI images. Multimed. Tools Appl. 2021, 80, 4295–4306. https://doi.org/10.1007/s11042-020-09970-8. 153. Guan, Y.; Aamir, M.; Rahman, Z.; Ali, A.; Abro, W.A.; Dayo, Z.A.; Bhutta, M.S.; Hu, Z.; Guan, Y.; Aamir, M.; et al. A framework for efficient brain tumor classification using MRI images. Math. Biosci. Eng. 2021, 18, 5790–5815. https://doi.org/10.3934/mbe. 2021292. Cancers 2024, 16, 300 53 of 55 154. Gull, S.; Akbar, S.; Khan, H.U. Automated Detection of Brain Tumor through Magnetic Resonance Images Using Convolutional Neural Network. BioMed Res. Int. 2021, 2021, 3365043. https://doi.org/10.1155/2021/3365043. 155. Özcan, H.; Emiroğlu, B.G.; Sabuncuoğlu, H.; Özdoğan, S.; Soyer, A.; Saygı, T. A comparative study for glioma classification using deep convolutional neural networks. Math. Biosci. Eng. 2021, 18, 1550–1572. https://doi.org/10.3934/mbe.2021080. 156. Aamir, M.; Rahman, Z.; Dayo, Z.A.; Abro, W.A.; Uddin, M.I.; Khan, I.; Imran, A.S.; Ali, Z.; Ishfaq, M.; Guan, Y.; et al. A deep learning approach for brain tumor classification using MRI images. Comput. Electr. Eng. 2022, 101, 108105. https: //doi.org/10.1016/j.compeleceng.2022.108105. 157. Wu, P.; Wang, Z.; Zheng, B.; Li, H.; Alsaadi, F.E.; Zeng, N. AGGN: Attention-based glioma grading network with multi-scale feature extraction and multi-modal information fusion. Comput. Biol. Med. 2023, 152, 106457. https://doi.org/10.1016/j. compbiomed.2022.106457. 158. Badža, M.M.; Barjaktarović, M.Č. Classification of brain tumors from MRI images using a convolutional neural network. Appl. Sci. 2020, 10, 1999. https://doi.org/10.3390/app10061999. 159. Ismael, S.A.A.; Mohammed, A.; Hefny, H. An enhanced deep learning approach for brain cancer MRI images classification using residual networks. Artif. Intell. Med. 2020, 102, 101779. https://doi.org/10.1016/j.artmed.2019.101779. 160. Alanazi, M.F.; Ali, M.U.; Hussain, S.J.; Zafar, A.; Mohatram, M.; Irfan, M.; AlRuwaili, R.; Alruwaili, M.; Ali, N.H.; Albarrak, A.M. Brain tumor/mass classification framework using magnetic-resonance-imaging-based isolated and developed transfer deep-learning model. Sensors 2022, 22, 372. https://doi.org/10.3390/s22010372. 161. O’Reilly, T.; Teeuwisse, W.M.; de Gans, D.; Koolstra, K.; Webb, A.G. In vivo 3D brain and extremity MRI at 50 mT using a permanent magnet Halbach array. Magn. Reson. Med. 2021, 85, 495–505. https://doi.org/https://doi.org/10.1002/mrm.28396. 162. Cooley, C.Z.; McDaniel, P.C.; Stockmann, J.P.; Srinivas, S.A.; Cauley, S.F.; Śliwiak, M.; Sappo, C.R.; Vaughn, C.F.; Guerin, B.; Rosen, M.S.; et al. A portable scanner for magnetic resonance imaging of the brain. Nat. Biomed. Eng. 2020, 5, 229–239. https://doi.org/10.1038/s41551-020-00641-5. 163. Man, C.; Lau, V.; Su, S.; Zhao, Y.; Xiao, L.; Ding, Y.; Leung, G.K.; Leong, A.T.; Wu, E.X. Deep learning enabled fast 3D brain MRI at 0.055 tesla. Sci. Adv. 2023, 9, eadi9327. https://doi.org/10.1126/SCIADV.ADI9327. 164. Swoop Portable MR System. Availabe online: https://hyperfine.io.assets/pdfs/Swoop (accessed on 6 November 2023). 165. Altaf, A.; Baqai, M.W.S.; Urooj, F.; Alam, M.S.; Aziz, H.F.; Mubarak, F.; Knopp, E.A.; Siddiqui, K.M.; Enam, S.A. Utilization of an ultra-low-field, portable magnetic resonance imaging for brain tumor assessment in lower middle-income countries. Surg. Neurol. Int. 2023, 14, 260. https://doi.org/10.25259/SNI_123_2023. 166. Altaf, A.; Baqai, M.W.S.; Urooj, F.; Alam, M.S.; Aziz, H.F.; Mubarak, F.; Knopp, E.; Siddiqui, K.; Enam, S.A. Intraoperative use of ultra-low-field, portable magnetic resonance imaging—First report. Surg. Neurol. Int. 2023, 14, 212. https://doi.org/10.25259 /SNI_124_2023. 167. Abd-Ellah, M.K.; Awad, A.I.; Hamed, H.F.; Khalaf, A.A. Parallel deep CNN structure for glioma detection and classification via brain MRI Images. In Proceedings of the 2019 31st International Conference on Microelectronics (ICM), Cairo, Egypt, 15–18 December 2019; pp. 304–307. 168. Anaraki, A.K.; Ayati, M.; Kazemi, F. Magnetic resonance imaging-based brain tumor grades classification and grading via convolutional neural networks and genetic algorithms. Biocybern. Biomed. Eng. 2019, 39, 63–74. https://doi.org/10.1016/j.bbe.20 18.10.004. 169. Hemanth, D.J.; Anitha, J.; Naaji, A.; Geman, O.; Popescu, D.E.; Son, L.H. A Modified Deep Convolutional Neural Network for Abnormal Brain Image Classification. IEEE Access 2019, 7, 4275–4283. https://doi.org/10.1109/ACCESS.2018.2885639. 170. Cubuk, E.D.; Zoph, B.; Mane, D.; Vasudevan, V.; Le, Q.V. AutoAugment: Learning Augmentation Policies from Data. arXiv 2018, https://doi.org/10.48550/arXiv.1805.09501 171. Muneer, K.V.A.; Rajendran, V.R.; Joseph, K.P. Glioma Tumor Grade Identification Using Artificial Intelligent Techniques. J. Med. Syst. 2019, 43, 1–12. https://doi.org/10.1007/s10916-019-1228-2. 172. Rajini, N.H. Brain Tumor Image Classification and Grading Using Convolutional Neural Network and Particle Swarm Optimiza- tion Algorithm. Int. J. Eng. Adv. Technol. (IJEAT) 2019, 8, 2249–8958. 173. Rahmathunneesa, A.P.; Muneer, K.V.A. Performance analysis of pre-trained deep learning networks for brain tumor categorization. In Proceedings of the 2019 9th International Conference on Advances in Computing and Communication (ICACC), Changsha, China, 18–20 October 2019; pp. 253–257. https://doi.org/10.1109/ICACC48162.2019.8986151. 174. Sajjad, M.; Khan, S.; Muhammad, K.; Wu, W.; Ullah, A.; Baik, S.W. Multi-grade brain tumor classification using deep CNN with extensive data augmentation. J. Comput. Sci. 2019, 30, 174–182. https://doi.org/10.1016/j.jocs.2018.12.003. 175. Sultan, H.H.; Salem, N.M.; Al-Atabany, W. Multi-Classification of Brain Tumor Images Using Deep Neural Network. IEEE Access 2019, 7, 69215–69225. https://doi.org/10.1109/access.2019.2919122. 176. Toğaçar, M.; Ergen, B.; Cömert, Z. BrainMRNet: Brain tumor detection using magnetic resonance images with a novel convolu- tional neural network model. Med. Hypotheses 2020, 134, 109531. https://doi.org/10.1016/j.mehy.2019.109531. 177. Amin, J.; Sharif, M.; Gul, N.; Yasmin, M.; Shad, S.A. Brain tumor classification based on DWT fusion of MRI sequences using convolutional neural network. Pattern Recognit. Lett. 2020, 129, 115–122. https://doi.org/10.1016/j.patrec.2019.11.016. 178. Afshar, P.; Plataniotis, K.N.; Mohammadi, A. BoostCaps: A boosted capsule network for brain tumor classification. In Proceedings of the Annual International Conference of the IEEE Engineering in Medicine and Biology Society, Virtual, 20–24 July 2020; pp. 1075–1079. https://doi.org/10.1109/EMBC44109.2020.9175922. Cancers 2024, 16, 300 54 of 55 179. Bhanothu, Y.; Kamalakannan, A.; Rajamanickam, G. Detection and classification of brain tumor in MRI images using deep convolutional network. In Proceedings of the 2020 6th International Conference on Advanced Computing and Communication Systems (ICACCS), Coimbatore, India, 6–7 March 2020; pp. 248–252. https://doi.org/10.1109/ICACCS48705.2020.9074375. 180. Çinar, A.; Yildirim, M. Detection of tumors on brain MRI images using the hybrid convolutional neural network architecture. Med. Hypotheses 2020, 139, 109684. https://doi.org/10.1016/j.mehy.2020.109684. 181. Khan, H.A.; Jue, W.; Mushtaq, M.; Mushtaq, M.U.; Khan, H.A.; Jue, W.; Mushtaq, M.; Mushtaq, M.U. Brain tumor classification in MRI image using convolutional neural network. Math. Biosci. Eng. 2020, 17, 6203–6216. https://doi.org/10.3934/mbe.2020328. 182. Mohammed, B.A.; Al-Ani, S. An efficient approach to diagnose brain tumors through deep CNN. Math. Biosci. Eng. 2020, 18, 851–867. https://doi.org/10.3934/mbe.2021045. 183. Naser, M.A.; Deen, M.J. Brain tumor segmentation and grading of lower-grade glioma using deep learning in MRI images. Comput. Biol. Med. 2020, 121, 103758. https://doi.org/10.1016/j.compbiomed.2020.103758. 184. Noreen, N.; Palaniappan, S.; Qayyum, A.; Ahmad, I.; Imran, M.; Shoaib, M. A deep learning model based on concatenation approach for the diagnosis of brain tumor. IEEE Access 2020, 8, 55135–55144. https://doi.org/10.1109/ACCESS.2020.2978629. 185. Saxena, P.; Maheshwari, A.; Maheshwari, S. Predictive Modeling of Brain Tumor: A Deep Learning Approach. Adv. Intell. Syst. Comput. 2020, 1189, 275–285. https://doi.org/10.1007/978-981-15-6067-5_30. 186. Sharif, M.I.; Li, J.P.; Khan, M.A.; Saleem, M.A. Active deep neural network features selection for segmentation and recognition of brain tumors using MRI images. Pattern Recognit. Lett. 2020, 129, 181–189. https://doi.org/10.1016/j.patrec.2019.11.019. 187. Vimal Kurup, R.; Sowmya, V.; Soman, K. Effect of data pre-processing on brain tumor classification using capsulenet. In Proceedings of the ICICCT 2019—System Reliability, Quality Control, Safety, Maintenance and Management: Applica- tions to Electrical, Electronics and Computer Science and Engineering, Hyderabad, India, 9–11 January 2019; Springer: Berlin/Heidelberg, Germany, 2020; pp. 110–119. 188. Bashir-Gonbadi, F.; Khotanlou, H. Brain tumor classification using deep convolutional autoencoder-based neural network: Multi- task approach. Multimed. Tools Appl. 2021, 80, 19909–19929. 189. Gu, X.; Shen, Z.; Xue, J.; Fan, Y.; Ni, T. Brain Tumor MR Image Classification Using Convolutional Dictionary Learning with Local Constraint. Front. Neurosci. 2021, 15, 679847. https://doi.org/10.3389/fnins.2021.679847. 190. Irmak, E. Multi-Classification of Brain Tumor MRI Images Using Deep Convolutional Neural Network with Fully Optimized Framework. Iran. J. Sci. Technol. Trans. Electr. Eng. 2021, 45, 1015–1036. https://doi.org/10.1007/S40998-021-00426-9. 191. Kader, I.A.E.; Xu, G.; Shuai, Z.; Saminu, S.; Javaid, I.; Ahmad, I.S.; Kamhi, S. Brain Tumor Detection and Classification on MR Images by a Deep Wavelet Auto-Encoder Model. Diagnostics 2021, 11, 1589. https://doi.org/10.3390/diagnostics11091589. 192. Kader, I.A.E.; Xu, G.; Shuai, Z.; Saminu, S.; Javaid, I.; Ahmad, I.S. Differential Deep Convolutional Neural Network Model for Brain Tumor Classification. Brain Sci. 2021, 11, 352. https://doi.org/10.3390/brainsci11030352. 193. Kakarla, J.; Isunuri, B.V.; Doppalapudi, K.S.; Bylapudi, K.S.R. Three-class classification of brain magnetic resonance images using average-pooling convolutional neural network. Int. J. Imaging Syst. Technol. 2021, 31, 1731–1740. https://doi.org/10.1002/ima.22554. 194. Masood, M.; Nazir, T.; Nawaz, M.; Mehmood, A.; Rashid, J.; Kwon, H.Y.; Mahmood, T.; Hussain, A. A novel deep learning method for recognition and classification of brain tumors from MRI images. Diagnostics 2021, 11, 744. https://doi.org/10.3390/ diagnostics11050744. 195. Sadad, T.; Rehman, A.; Munir, A.; Saba, T.; Tariq, U.; Ayesha, N.; Abbasi, R. Brain tumor detection and multi-classification using advanced deep learning techniques. Microsc. Res. Tech. 2021, 84, 1296–1308. https://doi.org/10.1002/JEMT.23688. 196. MohamedMetwalySherif. Brain Tumor Dataset. 2020. Available online: https://www.kaggle.com/datasets/mohamedmetwalysherif/ braintumordataset (accessed on 10 June 2023). 197. Chitnis, S.; Hosseini, R.; Xie, P. Brain tumor classification based on neural architecture search. Sci. Rep. 2022, 12, 1–12. https://doi.org/10.1038/s41598-022-22172-6. 198. Ekong, F.; Yu, Y.; Patamia, R.A.; Feng, X.; Tang, Q.; Mazumder, P.; Cai, J. Bayesian Depth-Wise Convolutional Neural Network Design for Brain Tumor MRI Classification. Diagnostics 2022, 12, 1657. https://doi.org/10.3390/DIAGNOSTICS12071657. 199. Gaur, L.; Bhandari, M.; Razdan, T.; Mallik, S.; Zhao, Z. Explanation-Driven Deep Learning Model for Prediction of Brain Tumour Status Using MRI Image Data. Front. Genet. 2022, 13, 448. https://doi.org/10.3389/fgene.2022.822666. 200. Gurunathan, A.; Krishnan, B. A Hybrid CNN-GLCM Classifier For Detection And Grade Classification Of Brain Tumor. Brain Imaging Behav. 2022, 16, 1410–1427. https://doi.org/10.1007/s11682-021-00598-2. 201. Isunuri, B.V.; Kakarla, J. Three-class brain tumor classification from magnetic resonance images using separable convolution based neural network. Concurr. Comput. Pract. Exp. 2022, 34, e6541. https://doi.org/10.1002/cpe.6541. 202. Khazaee, Z.; Langarizadeh, M.; Ahmadabadi, M.E.S. Developing an Artificial Intelligence Model for Tumor Grading and Classification, Based on MRI Sequences of Human Brain Gliomas. Int. J. Cancer Manag. 2022, 15, 120638. https://doi.org/10.581 2/ijcm.120638. 203. Koli, R.; Lotya, S.; Govekar, P.; Sachdev, K.; Bhatia, G. Detection and classification of brain tumor using MRI images. In Proceedings of the ICT Analysis and Applications, Goa, India, 29–30 July 2022; Springer: Berlin/Heidelberg, Germany, 2022; pp. 393–402. 204. Lakshmi, M.J.; Rao, S.N. Brain tumor magnetic resonance image classification: A deep learning approach. Soft Comput. 2022, 26, 6245–6253. https://doi.org/10.1007/S00500-022-07163-Z. Cancers 2024, 16, 300 55 of 55 205. Murthy, M.Y.B.; Koteswararao, A.; Babu, M.S. Adaptive fuzzy deformable fusion and optimized CNN with ensemble classification for automated brain tumor diagnosis. Biomed. Eng. Lett. 2022, 12, 37–58. https://doi.org/10.1007/s13534-021-00209-5. 206. Nayak, D.R.; Padhy, N.; Mallick, P.K.; Zymbler, M.; Kumar, S. Brain Tumor Classification Using Dense Efficient-Net. Axioms 2022, 11, 34. https://doi.org/10.3390/axioms11010034. 207. Raza, A.; Ayub, H.; Khan, J.A.; Ahmad, I.; S. Salama, A.; Daradkeh, Y.I.; Javeed, D.; Ur Rehman, A.; Hamam, H. A hybrid deep learning-based approach for brain tumor classification. Electronics 2022, 11, 1146. https://doi.org/10.3390/electronics11071146. 208. Rizwan, M.; Shabbir, A.; Javed, A.R.; Shabbir, M.; Baker, T.; Obe, D.A.J. Brain Tumor and Glioma Grade Classification Using Gaussian Convolutional Neural Network. IEEE Access 2022, 10, 29731–29740. https://doi.org/10.1109/ACCESS.2022.3153108. 209. Samee, N.A.; Mahmoud, N.F.; Atteia, G.; Abdallah, H.A.; Alabdulhafith, M.; Al-Gaashani, M.S.; Ahmad, S.; Muthanna, M.S.A. Classification Framework for Medical Diagnosis of Brain Tumor with an Effective Hybrid Transfer Learning Model. Diagnostics 2022, 12, 2541. https://doi.org/10.3390/DIAGNOSTICS12102541. 210. Sangeetha, S.K.; Muthukumaran, V.; Deeba, K.; Rajadurai, H.; Maheshwari, V.; Dalu, G.T. Multiconvolutional Transfer Learning for 3D Brain Tumor Magnetic Resonance Images. Comput. Intell. Neurosci. 2022, 2022, 8722476. https://doi.org/10.1155/2022/8 722476. 211. Srinivas, C.; Nandini, N.P.; Zakariah, M.; Alothaibi, Y.A.; Shaukat, K.; Partibane, B.; Awal, H. Deep Transfer Learning Approaches in Performance Analysis of Brain Tumor Classification Using MRI Images. J. Healthc. Eng. 2022, 2022, 3264367. https://doi.org/10.1155/2022/3264367. 212. Erickson, B.; Akkus, Z.; Sedlar, J.; Korfiatis, P. Data from LGG-1p19qDeletion (Version 2) [Data set]. The Cancer Imaging Archive. 2017. Available online: https://doi.org/10.7937/K9/TCIA.2017.DWEHTZ9V (accessed on 14 July 2023). 213. Vankdothu, R.; Hameed, M.A.; Fatima, H. A brain tumor identification and classification using deep learning based on CNN- LSTM method. Comput. Electr. Eng. 2022, 101, 107960. https://doi.org/10.1016/j.compeleceng.2022.107960. 214. Yazdan, S.A.; Ahmad, R.; Iqbal, N.; Rizwan, A.; Khan, A.N.; Kim, D.H. An Efficient Multi-Scale Convolutional Neural Network Based Multi-Class Brain MRI Classification for SaMD. Tomography 2022, 8, 1905–1927. https://doi.org/10.3390/TOMOGRAPHY8040161. 215. Anagun, Y. Smart brain tumor diagnosis system utilizing deep convolutional neural networks. Multimed. Tools Appl. 2023, 82, 44527–44553. https://doi.org/10.1007/S11042-023-15422-W. 216. Apostolopoulos, I.D.; Aznaouridis, S.; Tzani, M. An Attention-Based Deep Convolutional Neural Network for Brain Tumor and Disorder Classification and Grading in Magnetic Resonance Imaging. Information 2023, 14, 174. https://doi.org/10.3390/info140 30174. 217. Komaravolu, A. Brain Tumor MRI Images. Available online: https://www.kaggle.com/datasets/adityakomaravolu/brain- tumor-mri-images (accessed on 10 June 2023). 218. Yaseen, R. Brain Tumor Data MRI. Available online: https://www.kaggle.com/datasets/roroyaseen/brain-tumor-data-mri (accessed on 10 June 2023). 219. El-Wahab, B.S.A.; Nasr, M.E.; Khamis, S.; Ashour, A.S. BTC-fCNN: Fast Convolution Neural Network for Multi-class Brain Tumor Classification. Health Inf. Sci. Syst. 2023, 11, 1–22. https://doi.org/10.1007/S13755-022-00203-W. 220. Mahmud, M.I.; Mamun, M.; Abdelgawad, A. A Deep Analysis of Brain Tumor Detection from MR Images Using Deep Learning Networks. Algorithms 2023, 16, 176. https://doi.org/10.3390/A16040176. 221. Muezzinoglu, T.; Baygin, N.; Tuncer, I.; Barua, P.D.; Baygin, M.; Dogan, S.; Tuncer, T.; Palmer, E.E.; Cheong, K.H.; Acharya, U.R. PatchResNet: Multiple Patch Division–Based Deep Feature Fusion Framework for Brain Tumor Classification Using MRI Images. J. Digit. Imaging 2023, 12, 1–15. https://doi.org/10.1007/S10278-023-00789-X. 222. Özkaraca, O.; İhsan Bağrıaçık, O.; Gürüler, H.; Khan, F.; Hussain, J.; Khan, J.; e Laila, U. Multiple Brain Tumor Classification with Dense CNN Architecture Using Brain MRI Images. Life 2023, 13, 349. https://doi.org/10.3390/LIFE13020349. 223. Nickparvar, M. Brain Tumor MRI Dataset. 2021. Available online: https://www.kaggle.com/datasets/masoudnickparvar/brain- tumor-mri-dataset?select=Training (accessed on 4 June 2023). 224. Özkaya, C.; Şağıroğlu, C. Glioma Grade Classification Using CNNs and Segmentation with an Adaptive Approach Using Histogram Features in Brain MRIs. IEEE Access 2023, 11, 52275–52287. https://doi.org/10.1109/ACCESS.2023.3273532. 225. Rasheed, Z.; Ma, Y.K.; Ullah, I.; Shloul, T.A.; Tufail, A.B.; Ghadi, Y.Y.; Khan, M.Z.; Mohamed, H.G. Automated Classification of Brain Tumors from Magnetic Resonance Imaging Using Deep Learning. Brain Sci. 2023, 13, 602. https://doi.org/10.3390/ BRAINSCI13040602. 226. Srinivasan, S.; Bai, P.S.M.; Mathivanan, S.K.; Muthukumaran, V.; Babu, J.C.; Vilcekova, L. Grade Classification of Tumors from Brain Magnetic Resonance Images Using a Deep Learning Technique. Diagnostics 2023, 13, 1153. https://doi.org/10.3390/ DIAGNOSTICS13061153. 227. van der Voort, S.R.; Incekara, F.; Wijnenga, M.M.; Kapsas, G.; Gahrmann, R.; Schouten, J.W.; Dubbink, H.J.; Vincent, A.J.; van den Bent, M.J.; French, P.J.; et al. The Erasmus Glioma Database (EGD): Structural MRI scans, WHO 2016 subtypes, and segmentations of 774 patients with glioma. Data Brief 2021, 37, 107191. https://doi.org/10.1016/j.dib.2021.107191. Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.