NeuroImage: Clinical 34 (2022) 102972 Contents lists available at ScienceDirect NeuroImage: Clinical journal homepage: www.elsevier.com/locate/ynicl MAGNIMS recommendations for harmonization of MRI data in MS multicenter studies Nicola De Stefano a, *, Marco Battaglini a, Deborah Pareto b, Rosa Cortese a, Jian Zhang a, Niels Oesingmann c, Ferran Prados d, e, f, Maria A. Rocca g, h, i, Paola Valsasina g, Hugo Vrenken j, Claudia A.M. Gandini Wheeler-Kingshott d, k, l, Massimo Filippi g, h, i, m, n, Frederik Barkhof d, e, j, Àlex Rovira b, on behalf of the MAGNIMS Study Group a Department of Medicine, Surgery and Neuroscience, University of Siena, Siena, Italy b Section of Neuroradiology, Department of Radiology, Hospital Universitari Vall d’Hebron, Universitat Autònoma de Barcelona, Barcelona, Spain c UK Biobank Stockport, Cheshire SK3 0SA, United Kingdom d Queen Square Multiple Sclerosis Centre, Department of Neuroinflammation, UCL Queen Square Institute of Neurology, University College London, London, United Kingdom e Center for Medical Imaging Computing, Medical Physics and Biomedical Engineering, UCL, London, WC1V 6LJ, United Kingdom f e-Health Center, Universitat Oberta de Catalunya, Barcelona, Spain g Neuroimaging Research Unit, Division of Neuroscience, and Neurology Unit, IRCCS San Raffaele Scientific Institute, Milan, Italy h Neurology Unit, IRCCS San Raffaele Scientific Institute, Milan, Italy i Vita-Salute San Raffaele University, Milan, Italy j Amsterdam Neuroscience, MS Center Amsterdam, Department of Radiology and Nuclear Medicine, Amsterdam UMC, Amsterdam, Netherlands k Brain MRI 3T Research Center, C. Mondino National Neurological Institute, Pavia, Italy l Department of Brain and Behavioural Sciences, University of Pavia, Pavia, Italy m Neurorehabilitation Unit, and Neurophysiology Service, IRCCS San Raffaele Scientific Institute, Milan, Italy n Neurophysiology Service, IRCCS San Raffaele Scientific Institute, Milan, Italy A R T I C L E I N F O A B S T R A C T Keywords: There is an increasing need of sharing harmonized data from large, cooperative studies as this is essential to MRI develop new diagnostic and prognostic biomarkers. In the field of multiple sclerosis (MS), the issue has become Harmonization of paramount importance due to the need to translate into the clinical setting some of the most recent MRI Multiple sclerosis achievements. However, differences in MRI acquisition parameters, image analysis and data storage across sites, with their potential bias, represent a substantial constraint. This review focuses on the state of the art, recent technical advances, and desirable future developments of the harmonization of acquisition, analysis and storage of large-scale multicentre MRI data of MS cohorts. Huge efforts are currently being made to achieve all the requirements needed to provide harmonized MRI datasets in the MS field, as proper management of large im­ aging datasets is one of our greatest opportunities and challenges in the coming years. Recommendations based on these achievements will be provided here. Despite the advances that have been made, the complexity of these tasks requires further research by specialized academical centres, with dedicated technical and human resources. Such collective efforts involving different professional figures are of crucial importance to offer to MS patients a personalised management while minimizing consumption of resources. 1. Introduction issue has become prominent due to the need to translate into the clinical setting some of the most important magnetic resonance imaging (MRI) The increasing need of sharing data from large, cooperative studies is achievements. However, the recent increase of large-scale MRI studies of paramount importance as this is essential to develop new diagnostic has highlighted the need to control the MR-related source of variability and prognostic biomarkers. In the field of multiple sclerosis (MS), the introduced by differences in scanners, acquisition protocols and image * Corresponding author at: Department of Medicine, Surgery and Neuroscience, University of Siena, Viale Bracci 2, Siena 53100, Italy. E-mail address: destefano@unisi.it (N. De Stefano). https://doi.org/10.1016/j.nicl.2022.102972 Received 16 July 2021; Received in revised form 22 February 2022; Accepted 23 February 2022 Available online 25 February 2022 2213-1582/© 2022 The Author(s). Published by Elsevier Inc. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/). N. De Stefano et al. NeuroImage: Clinical 34 (2022) 102972 analysis when combining MRI data across multiple sites (Shinohara manuscript was written by the first and last authors and was based on et al., 2017; Biberacher et al., 2016). Such unwanted variability can contributions from each panelist. The draft was then circulated to all have unpredictable effects and potentially influence the detection of members, who iteratively modified the document until a final consensus pathological findings, leading to inconsistent results and spurious re­ was reached by all panel members. lationships with clinical outcomes (Biberacher et al., 2016). Recent studies have tackled the challenges related to the lack of harmonization 2.1. Harmonization of MRI biomarkers in MS of MRI data from multiple sites (Dewey et al., 2019) showing that, with some new procedures, it can be possible to remove unwanted sources of We will tackle here issues related to the lack of harmonization of the scan variability while simultaneously increasing the power and repro­ MRI data from multiple sites and provide tailored solutions to reduce ducibility of statistical analyses. However, the issue of harmonization of sources of variability for MS imaging biomarkers that have been MRI data is far from being solved, with open challenges involving image extensively used in multicenter studies. For each MR-related biomarker, acquisition, analysis and data storage. solutions to harmonize image acquisition and analysis will be consid­ Issues with image acquisition include standardization across scan­ ered. Fig. 1 shows the suitability of each MRI-derived metric towards ners of certain characteristics (e.g., field strength, vendor, sequence harmonization, and Table 1 summarizes the main sources of variability type, coil channels, gradient distortions) and subject-related factors (e. in MRI acquisition and analysis and provides recommendation for their g., positioning, motion, etc.). The key concept is the reproducibility of harmonization. the MR examination, which is difficult to obtain even from the same Using the radar chart, each biomarker was ranked on 5 features that vendor (Shinohara et al., 2017). Practical solutions have been proposed make it suitable to be used in multi-centers studies. The values range by the 2021 MAGNIMS-CMSC-NAIMS guidelines, merging recommen­ from 0 (no suitability) to 5 (excellent suitability). The scores of the dations from the Magnetic Resonance Imaging in Multiple Sclerosis different items were decided via consensus after being proposed by the study group, Consortium of Multiple Sclerosis Centres, and North first author. All co-authors agreed with the proposed scores. This American Imaging in Multiple Sclerosis Cooperative, on the use of MRI multivariate, qualitative representation allows to compare the most in MS for clinical implementation in the diagnostic process and for used MRI biomarkers in MS. The chart shows that while atrophy mea­ establishing prognosis and monitoring patients (Wattjes et al., 2021a). sures (orange) and T2-Lesions (light blue) are more reliable measures in While some serious attempts to standardize image acquisition protocol multi-center studies than functional connectivity related metrics have been performed, this is much less the case for image analysis (green), they do lack in specificity. harmonization, which should include: i) quality assessment (QA) to homogeneously check the initial images, classifying or excluding sub­ 2.2. White matter lesions optimal examinations, and fixing possible errors through a modified pipeline, and ii) choice of the most appropriate analysis method for a The presence of T2-weighted (T2-w) hyperintense brain white matter given task. Recommendations to improve imaging analysis of brain le­ (WM) lesions as detected by MRI is the hallmark of MS and play a crucial sions and volume in longitudinal MS studies have been published by the role in diagnosis, prognosis and treatment response of MS patients MAGNIMS group (Vrenken et al., 2013), but many issues are presently (Wattjes et al., 2021a). The 2017 revision of the MRI criteria for the unsolved and the harmonization of imaging analysis remains a concern. diagnosis of MS recommended a strict standardization of MRI acquisi­ Finally, it is now clear that the availability of efficient data storage and tion to avoid misdiagnosis (Thompson et al., 2018). Recently, the 2021 sharing methods can have a huge impact to facilitate the management of MAGNIMS–CMSC–NAIMS consensus recommendations provided guid­ large imaging datasets. ance on how and when to use MRI for the management of patients with Against this background, this review focuses on the state of the art MS and suggested changes in MRI acquisition protocols to improve and desirable future developments in the harmonization of acquisition, diagnostic accuracy (Wattjes et al., 2021a). analysis, and storage of large-scale multicenter MRI data of MS cohorts. To minimize between-site variability while maintaining maximum We will refer here to strategies aiming at reducing, after merging brain sensitivity in detecting brain WM lesions on clinical scanners, 3D T2w- MRI scans from different datasets, the technical sources of variability fluid attenuated inversion recovery (FLAIR) images with spatial reso­ due to acquisition and analysis but preserving the variability due to the lution of 1 mm3 at a minimum strength of 1.5 T are recommended, as disease-related abnormalities. To do this, we will organize this review as they lead to both improved lesion detection and realignment of follows: first, we will describe the impact on imaging biomarkers used in anatomic orientation to detect new lesions in serial MRI scans (Wattjes MS of the most common sources of variability related to MRI scan et al., 2021b). Magnetic field strength also may play a role, as the visi­ acquisition and analysis and the possible solutions to deal with them; bility of brain WM lesions is higher at 3 T than 1.5 T (Hagens et al., second, we will describe the latest statistical strategies to merge multi­ 2019). centre MRI data and to store and retrieve these large datasets. Third, we For quantitative measurement of brain WM lesion volume (LV) in will highlight recent experiences with large multicenter MRI datasets MS, early work using a semi-automated thresholding technique with a focus on how the issue of data harmonization has been faced. observed that, while LVs differed between fast-FLAIR, RARE (Rapid Finally, we will describe future perspectives and provide conclusive Acquisition with Relaxation Enhancement) and GRASE (GRadient And recommendations. Spin Echo) sequences, repeatability was similar (Rovaris et al., 1999). For automated methods, while the performance on 2D FLAIR is limited, 2. Methods some methods on 3D FLAIR exhibit good performance (Egger et al., 2017). Direct comparisons between scanners are scarce but between- An international workshop was held in Siena, Italy in 2019 under the scanner variability of LV is not as large as the variability found be­ auspices of the MAGNIMS study group, an intellectually independent tween different types of analysis software. However, even with harmo­ European network of academic centers with a common interest in the nized protocols and scanners from a single vendor, small systematic study of MS with MRI (www.magnims.eu). The panel convened to the between-scanner LV differences remain (Shinohara et al., 2017). workshop was composed of experts in the use of MRI for the diagnosis Quantitative measurement of spinal LV is under development and so and management of MS, and included neurologists, neuroradiologists the detection of these lesions represents the main goal. New automatic and physicists from MAGNIMS institutions and one invited non- frameworks for segmenting the spinal cord and intramedullary MS le­ MAGNIMS expert. The panel met to present and discuss data from MS sions have been recently developed (De Leener et al., 2017; Gros et al., multicenter research studies (published in English) leading to advances 2019). The standardization of imaging protocols may potentially in knowledge in the harmonization of MRI data. The first draft of the improve the performance of these tools and help the identification of 2 N. De Stefano et al. NeuroImage: Clinical 34 (2022) 102972 Fig. 1. Consensus Chart on the maturity of the main MRI biomarkers used in MS towards harmonization. very small lesions while reducing the number of false positives (Cohen- dataset was used to evaluate the improvement in manual lesion seg­ Adad et al., 2021b; Cohen-Adad et al., 2021a). Various studies showed mentation when MICA (multisite image harmonization by CDF), a non- higher sensitivity to lesions for axial compared to sagittal images, linear intensity transformation, was first applied to each MR image including lesions located in the lateral columns, lesions of smaller size (Wrobel et al., 2020). This preprocessing step reduced the inter-center and gadolinium (Gd) enhancing lesions (Breckwoldt et al., 2017). variability by one third, leading to a substantial improvement in Studies also showed that proton density (PD) images are sensitive lesion estimation. (Karavasilis et al., 2019), making dual-echo or similar multi-contrast In last few years, an impressive number of new tools to automatically sequences the preferred imaging approach, given the technical chal­ quantify lesions in MS have been based on artificial intelligence (AI) lenges still associated with cerebrospinal fluid suppression in spinal cord algorithms (Zeng et al., 2020). Deep learning (DL) based algorithms FLAIR scans. Ideally, pooling together 3D T2w and T2* weighted axial have shown good performance on independent datasets acquired with images would allow a better automated MS cord lesion segmentation the same MR protocol. Studies showed, however, that their performance (Gros et al., 2019). decreases when applied to images with features (i.e., signal to noise ratio Harmonization of the MRI acquisition [SNR], contrast to noise ratio [CNR]) different from those used in the Lesion number and volume should be assessed on images acquired on training dataset (Mårtensson et al., 2020) or when simple procedures scanners with the same magnetic field strength (Di Perri et al., 2009; such as the facial features removal (FFR) (a step implemented before Stankiewicz et al., 2011). The automated detection of lesions can be data sharing in order to protect the subject’s privacy) that could change improved by correcting for B1 inhomogeneity (Sajja et al., 2006). To our the overall histogram intensity of the image are applied (de Sitter et al., knowledge, no studies have been performed to assess, using the same 2020). Nonetheless, using transfer learning, specific features of data MR scanner and sequence, the impact of a change in TR or TE on the from a new scanner can be learned to refine an existing trained network, evaluation of brain and spinal MS lesions. By contrast, for voxel di­ which can yield good performance even when re-training based only on mensions, such a comparison was performed and suggested that lesional a single subject (Weeda et al., 2019a; Valverde et al., 2019). measures should be obtained from images with the same voxel di­ Recommendations mensions (Molyneux et al., 1998). Against this background, 3D FLAIR images with same/similar Harmonization of MRI analysis acquisition parameters should be used to outline lesions, after in­ Manual lesion segmentation has always been considered the gold homogeneity correction and intensity normalization. Isotropic 3D FLAIR standard of LV measurement. However, this approach is time- images should be preferred to anisotropic 2D as they are more sensitive consuming and not feasible in large MR datasets. In addition, the to lesion detection and less sensitive to artifacts. (Filippi et al., 2019). growing use of 3D FLAIR images makes this task nearly unfeasible in a The criteria used to perform the QA should be provided. DL methods lab setting (Filippi et al., 2019). The first step to obtain comparable should be applied to images similar to those of the training dataset. results in LV number and estimation is to perform a careful QA to These automated methods take advantage of and improve their perfor­ exclude images with low quality and artefacts. LV can be biased also by mance, when trained on more than one MR modality. Thus, for example, the intrinsic variability in MRI intensity. In a recent paper (Shinohara isotropic 3D T1-W, 3D T2-W and 3D FLAIR images could be used et al., 2017), where a single MS patient was scanned-rescanned (with together to increase the precision of the trained algorithms. The mag­ replacement) seven times at seven different centers all equipped with a netic field strength should be taken into account when merging lesion 3 T scanner and performing the same MRI acquisition protocol, the outputs from scanners with different magnetic fields. center explained 80% of variability in manual lesion outlining. The same 3 N. De Stefano et al. NeuroImage: Clinical 34 (2022) 102972 Table 1 Sources of variability related to MRI scan acquisition and analysis and recommendation to reduce their impact for the most common imaging biomarkers used in MS. Biomarkers Sources of variability in MRI acquisition Sources of variability in MRI analysis Recommendations for harmonization of MRI acquisition and analysis WM lesions • Different acquisition protocols • Inter rater variability in manual lesion • Standardization of imaging acquisition protocols (i.e., use of • Different magnetic field strengths segmentation isotropic 3D FLAIR with spatial resolution of 1 mm3 acquired • Different scanners • Variability in voxel intensity at minimum 1.5 Tesla) • B0 or B1 inhomogeneity • Lack of deep learning generalizability • Inhomogeneity and intensity normalization • Registration pipelines (when used) • Careful QC before the analysis • Use of machine learning-based algorithms on images similar to those of the training dataset • Inclusion of magnetic field strength and image characteristics when merging lesions outputs from different scanners Atrophy • Different acquisition protocols • Presence of black holes • Standardization of imaging acquisition protocols and minimal • Different magnetic field strengths • Variability in voxel intensity hardware or software changes • Different scanners • Defacing • Bias-field correction and intensity normalization • B0 or B1 inhomogeneity • Software variability • Careful QC before the analysis • Lack of automated segmentation • Use of lesion-filled isotropic 3D T1-weighted images acquired generalizability at magnet iso-center • Registration pipelines (when used) MTR-derived • Different acquisition protocols • Registration pipeline (when used) • Careful QC before the analysis metrics • Different magnetic field strengths • Use of constant parameters for acquisition, same transmission • Different scanners coil and correction for B1 errors • Strong dependency on radiofrequency • Careful check of ROI identification on MTR images registered pulse from T1-weighted images • Within scanner coil variability DTI-derived • Differences in vendors and magnetic • Software variability • MRI acquisition with same magnetic field strength, same metrics field strengths number of diffusion gradients, using parallel imaging and • Different protocols (with B0 opposite phase-encoding directions susceptibility distortions, number of • Same voxel size, B0 volumes, TE and TR diffusion gradients) • Careful QC before the analysis • Eddy current distortions, Gibbs ringing • Use of software tools to reduce the effects of table vibration artefacts, table vibration • Denoising • Identical software setting for pre and post processing (correction for eddy currents, motion, and B0 and B1- inhomogeneity) Functional MRI- • Different acquisition protocols • Different preprocessing pipelines (head • MRI acquisition with same magnetic field strength, using derived metrics • Different magnetic field strengths motion and physiological noise correction) opposite phase-encoding directions and same MRI protocol • Different scanners and software • Same temporal signal-to-fluctuation-noise-ratio across • B0 susceptibility distortions and B1 • Method to quantify fMRI activation or scanners inhomogeneity functional connectivity • Careful QC before the analysis • Eye movements artefacts, physiological • EPI alignment and ICA analysis to correct for head motion noise artefacts, head motion • Scan to be performed with closed eyes to avoid/reduce eye movement artefacts. • Identical software setting for pre and post processing (motion, physiological noise correction and B0/B1-inhomogeneity) Abbreviations: MR: Magnetic resonance; DL: Deep learning; QC: Quality Control; MTR: Magnetization Transfer Ratio; GM: Grey Matter; WM: White Matter; ROI: Region of Interest; TE: Echo Time; TR: Repetition Time; ICA: Independent Component Analysis; EPI: Echo planar imaging. 2.3. Atrophy robust and reproducible results (Goodkin et al., 2021). Specific strate­ gies can be adopted to reduce the impact of B0- and B1-inhomogeneities. Atrophy measurements have been extensively used in group studies The former can be reduced by repositioning the patient at the magnet to chart the MS disease course and to test treatment effects (Sastre- isocenter (Caramanos et al., 2010) and using distortion correction Garriga et al., 2020). However, brain volume measurements, both cross- techniques commonly implemented by scanner vendors (Jovicich et al., sectional and longitudinal, can differ systematically between images 2006), including pre-computed displacement tables, phase mapping, acquired i) from scanners of different vendors (with rather poor agree­ and forward/reverse frequency-encoding gradients (Doran et al., 2005). ment) (Biberacher et al., 2016), ii) before and after scanner upgrade The latter can be corrected after image acquisition by using several (Steenwijk et al., 2016), and iii) even from two different scanners of the open-source online software tools (Song et al., 2017), often included in same model in a single site (Takao et al., 2011). Moreover, changes of the analysis algorithms used for biomarker evaluation, e.g. the N3 and T1w sequence (i.e., from magnetization prepared-rapid gradient echo N4 methods (Sled et al., 1998; Tustison et al., 2010). In addition, the [MP-RAGE] to inversion recovery fast spoiled gradient recalled echo variability in MRI-based global and regional volume measurements can [IR-SPGR]) may have greater effects on brain volume measurements be reduced at the statistical analysis taking into account magnet vendor than intra-vendor scanner upgrades (Lee et al., 2019). and magnetic field as confounding factors (Potvin et al., 2016; Battaglini Finally, the effects are not the same for all types of measurements (i. et al., 2019). A standardized protocol should be always used in multi­ e., whole brain versus smaller brain structures) or analysis software center studies, as it has been shown that the variance of brain volume used, as some seem to be more robust than others (Steenwijk et al., measures increases with greater deviation from the standardized pro­ 2017). Measurement of spinal cord volume in MS has mostly focused on tocol in terms of repetition time, echo time and spatial resolution (Potvin the cervical segment, with issues similar to those of the measurement of et al., 2019). brain volumes. Atrophy of the upper cervical cord can be assessed through head Harmonization of the MRI acquisition scans rather than dedicated spinal cord scans (intra-class correlation Unenhanced 3D T1-weighted MRI is the reference sequence for brain coefficient [ICC] = 0.987) making reliable standardized acquisition volume quantification, although T2-FLAIR data have shown to provide more accessible (Liu et al., 2016), although correction for distortion due 4 N. De Stefano et al. NeuroImage: Clinical 34 (2022) 102972 to gradient nonlinearity is needed (Papinutto et al., 2018). Systematic obtained using the same software. between-scanner differences in mean upper cervical cord area (MUCCA) Spinal cord sequences should be acquired according to the selected have been reported for patients scanned on multiple scanners as well as software for the quantification of volume. As for brain atrophy, strictly for multi-site travelling volunteers (Weeda et al., 2019b; Lukas et al., harmonized protocols should be used. 2021). Within-subject identical scanning protocol is mandatory to have reliable quantification of longitudinal spinal cord atrophy (Weeda et al., 2.4. Microstructural biomarkers 2019b; Papinutto and Henry, 2019). Harmonization of MRI analysis Using computational models of biophysical properties of tissues, it is The first preliminary step to obtain comparable results in brain possible to derive metrics sensitive to microstructural integrity and volume assessment is to perform a careful QC to exclude images with a metabolic information of the underlying tissue (Granziera et al., 2021). suboptimal quality. Strategies of voxel intensity normalization can Ultimately, the aim would be to provide patient-specific metrics able to reduce atrophy measurement error (Wrobel et al., 2020; Battaglini et al., predict outcomes and contribute to decision making on disease man­ 2018). DeepHarmony, a method of DL architecture aiming at creating agement (Granziera et al., 2021). intensity transformation maps to equalize images from two different MR To improve the reliability of microstructural biomarkers (e.g., using scanners (Dewey et al., 2019); showed a significant decrease in the brain magnetization transfer ratio [MTR], myelin water imaging [MWI]), volume measurement error on a dataset of 12 subjects scanned at two inter- and intra- intensity image harmonization is required to increase different MR centers. To face the global intensity changes due to the the sensitivity to disease-related effects and to better assess longitudinal pathological features such as the presence of T1-weighted hypointense changes. For specificity to biophysical properties, which allows to un­ lesions, a number of methods include the “filling” of WM lesions before derstand mechanisms of disease progression and treatment response, image segmentation (Nakamura and Fisher, 2009; Sdika and Pelletier, harmonization is needed to evaluate the tissue integrity (e.g. neurite 2009; Battaglini et al., 2012), with a great reduction of the measurement density index, bound pool fraction). error. The defacing step showed a significant systematic increase in the Despite their potential value, non-conventional MRI metrics are still variability of brain volume measurement with over- or underestimation dependent on acquisition parameters, which are heavily influenced by depending on the software used (de Sitter et al., 2020). coil geometry, gradient systems and magnetic field strengths. This, Over the last decade, various studies assessing the variability in brain together with the restricted availability of some non-conventional MRI volume measurements (Sastre-Garriga et al., 2020) showed that sequences, has limited the use of some promising metrics (e.g., MWI) in different software tools have a weak ICC (Guizard et al., 2015; Popescu multicenter studies and is important to consider in other, more widely et al., 2016; Battaglini et al., 2009). Most of these tools showed used measures (e.g., MTR, diffusion metrics), which will be discussed increased variability when applied to MR images of the same subject here. that were acquired with different scans and/or vendors, in comparison Microstructural damage assessment using MTR with those acquired with the same MRI equipment (Storelli et al., 2018). Magnetization transfer (MT) images are obtained by the normalized Several studies have tried to optimize the software setting options in difference between two images with and without MT pulse. They order to minimize the technical variability (Popescu et al., 2012; Lut­ showed sensitivity to the inherent relaxation properties of heteroge­ kenhoff et al., 2014). More recently, several studies investigated the neous tissues (Ropele et al., 2005). MTR is a semi-quantitative metric performance of DL architectures trained on binarized segmented GM obtained by the two-pool model (restricted protons of macromolecules regions obtained by specific software packages (Gabr et al., 2020; and free protons of bulk water), predominantly sensitive to myelin McKinley et al., 2021) and therefore referred to here as “silver stan­ content, but also influenced by axonal loss and inflammation. MTR can dard”, as opposed to the “gold standard” that is represented by the be measured on multi-vendor MRI images using a standardized acqui­ binarised GM map manually segmented. Results showed comparable sition protocol (Barker et al., 2005) and has been used as outcome accuracy between the images used for training and those used for the measure in clinical trials in relapsing-remitting (RR) and progressive MS evaluation, also when a small sample size (e.g., 10–20 subjects) was used (Romme Christensen et al., 2014; Miller et al., 2015). Advances of for training the DL architecture (Narayana et al., 2020). Although modelling of the MT effect led to the development of multi-modal promising, these results need confirmation when the training dataset is acquisition protocols such as the multi-parametric mapping, which al­ acquired with a different protocol compared to the evaluation dataset. lows to quantify several parameters sensitive to tissue microstructural Semiautomated software packages have been developed to assess changes (MT saturation, apparent proton density, relaxation times). spinal cord atrophy both cross-sectionally (De Leener et al., 2017; These parameters have been shown to be comparable, at least between Horsfield et al., 2010), measuring the spinal cord cross-sectional area scanners of the same vendor (Weiskopf et al., 2013). (CSA), and longitudinally, measuring the generalized boundary shift Harmonization of the MRI acquisition integral (GBSI) (Prados et al., 2020). Due to dependencies on the MT weighting radiofrequency pulse, Recommendations harmonizing this protocol is challenging, although recent developments In general, the method of choice for quantification of brain and in acquisition and scanner parameters (i.e. use of different resolutions spinal cord volume remains the consistent use of the most appropriate and selective pulses for saturation) attempted to reduce confounding MR sequence for the type of software package that is going to be used (e. factors, thus improving MT feasibility (Zhang et al., 2019). Indeed, i) MT g., unenhanced 3D gradient-echo T1-weighted), with the identical images from scanners with the same strength of magnet could be merged scanning protocol and as little hardware or software change as possible. if they are equipped with the same transmission coil (Tofts et al., 2006); When dealing with 2D acquisitions, almost all methods reformat input ii) impact of B1-inhomogeneity could be corrected by computing the scans to 1 mm isotropic resolution to improve measurement precision linear regression between the B1 errors and MTR values (Ropele et al., (Amann et al., 2015). Hence, a direct acquisition of isotropic 3D T1- 2005); iii) acquisition parameters should be kept constant, as variability weighted images is preferable as it is less distorted by blurring due to in the MTR measurement can be up to 15% when subjects are acquired interpolation. Thus, we recommend that brain volume measurement on the same scanner but with different MR protocols (Sormani et al., should be assessed on lesion-filled 3D T1-weighted images of patients 2000). placed at magnet iso-center, bias-field corrected, and intensity normal­ Harmonization of MRI analysis ized. The analysis pipeline should be clearly detailed as well as the Features of MTR intensity histograms (Tortorella et al., 2000; Fer­ criteria used to perform the QC. nando et al., 2005) and averaged MTR values (De Stefano et al., 2006) of DL methods are promising but need to be applied to images similar to the whole brain, GM, WM and lesions are the most widely used measures those used for the training, where the output “to be learnt” has been obtained from the MT images. There are currently no open-source 5 N. De Stefano et al. NeuroImage: Clinical 34 (2022) 102972 software tools able to automatically assess MTR values from specific Recently, various studies in the DTI field have applied DL techniques ROIs. However, this should not represent a major limitation for MTR in order to reduce the scan time (Tian et al., 2020), reliably recon­ measurement, once registration and segmentation steps are fine-tuned structing WM tracts (Li et al., 2020) and characterizing pathological on MR images of each center and visually checked. conditions (Marzban et al., 2020). Although promising, such techniques Recommendations need a further extensive validation. A few studies applied AI methods to MT images need to be acquired with the same transmission coil and DTI data, with the aim to characterize different stages of MS (Marzullo corrected for B1 errors. The same acquisition protocol needs to be used. et al., 2019; Oladosu et al., 2021; Kontopodis et al., 2021). To obtain a precise assessment of average MTR values within the ROIs, Recommendations registration needs to be carefully checked to avoid the inclusion of Analysis from multicentre DTI can be more precise if images are inappropriate voxels. acquired on MR scanners with the same magnetic field strength, number Microstructural damage assessment using Diffusion Tensor Imaging (DTI) of gradients and opposite phase-encoding directions. Then, both pre- Information on microstructural integrity can also be assessed using processing (i.e., correction for eddy currents, motion, B0- and B1- diffusion weighted imaging (DWI) with a minimum of 6 gradient di­ inhomogeneity) and post-processing software tools should be used rections, through either model-free approaches or computational with identical and well documented analysis algorithms. modelling of DWI signal (Cercignani and Gandini Wheeler-Kingshott, 2019). The most simplistic model is diffusion tensor imaging (DTI), 2.5. Functional biomarkers which reduces the complex motion of water molecules within a tissue to three eigenvectors and corresponding eigenvalues. DTI provides metrics An important remark should be made on functional MRI (fMRI) and such as fractional anisotropy (FA), mean diffusivity (MD) and axial and functional connectivity (FC). From a set of T2*-weighted EPI volumes it radial diffusivity that are rotationally invariant, within the subject’s is possible to model the local tissue changes in the concentration of frame of reference and heavily influenced by the underlying tissue oxyhaemoglobin, which is diamagnetic, compared to deoxy­ microstructure that poses hindrance or restriction to free motion of haemoglobin, which is paramagnetic. Such changes are also linked to water. the baseline neuronal activity at rest, thus leading to a remarkable in­ Recent studies reported that DTI can be highly reproducible and crease in research studies of resting state fMRI (rs-fMRI) in MS. It is well standardized at brain (Palacios et al., 2017) and spinal cord (Samson known that low frequency signal fluctuations of regions belonging to et al., 2016) levels. However, DTI metrics are non-specific since they are networks supporting specific functions are synchronized, with altered highly sensitive to microstructural alterations and higher order models FC strength occurring in pathologic conditions (Castellazzi et al., 2018). are required to disentangle the source of signal anisotropy. Moreover, Harmonization of rs-fMRI acquisition should aim to strengthen the sta­ the time-dependence of the diffusion signal, that is how the signal tistical inference of FC metrics and potentially propose an approach changes by increasing the observational diffusion time, depends on the based on the single-subject comparison to healthy population. It is size of the underlying axons. This has been extensively studied in the important that rs-fMRI acquisition protocols across scanners have the brain (De Santis et al., 2016) and is particularly important for imaging same sensitivity, which can be measured with parameters such as the the spinal cord, which has larger axons than brain and where, with a temporal signal-to-fluctuation-noise-ratio (SFNR) (Glover et al., 2012). short diffusion time, a high number of large axons may not contribute to Harmonization of the rs-fMRI acquisition the signal (Grussu et al., 2019). To overcome these limitations, advanced 3 T scanners showed significantly higher reproducibility in blood- DWI approaches have been developed (Assaf et al., 2004). However, oxygenation level dependent (BOLD) signal than 1.5 T scanners (Zou although more specific to tissue microstructure, further studies are et al., 2005; Badhwar et al., 2020). B0-susceptibility induced distortion needed to prove the translation of these multiple-scale approaches into can be reduced with the use of parallel imaging and the acquisition of clinical practice. (Cercignani and Gandini Wheeler-Kingshott, 2019). data with opposite phase encoding directions (Andersson et al., 2003). Harmonization of MRI acquisition and data-preprocessing Sophisticated strategies have been recently introduced to reduce the When applying a standardized MRI protocol across different centers, impact of B1-inhomogeneity through the creation of B1 maps by averaged FA and MD measures from large homogeneous ROIs (e.g., the modelling the B1- receive and B1 + transmit fields (Glasser et al., 2013). mid-sagittal section of the corpus callosum) can be merged. On the other Head motion is usually corrected by aligning EPI images (Graedel et al., hand, differences in vendors, magnetic field strength and number of 2017), and the impact of patient movements can be further reduced by diffusion gradients should be statistically taken into account when applying independent component analysis (ICA)-based method (Pruim measures from smaller ROIs are obtained, in order to reduce the intra- et al., 2015); eye movement artefacts can be reduced by asking the scanner and intra-protocol sources of variability (Pagani et al., 2010). subject to keep the eyes closed during the acquisition or to fix on a point B0 susceptibility-induced distortions can be corrected using parallel (Costumero et al., 2020); physiological noise can be reduced by imaging and acquiring MRI data with opposite phase-encoding di­ recording electrocardiogram (ECG) signal and regressing it out in the rections. Susceptibility induced distortions can influence both intra- and statistical analysis (Glover et al., 2000); ICA-based methods can be a inter-subject reproducibility (Ganzetti et al., 2016), but its effect can be good alternative when ECG recording is not available (Behzadi et al., reduced using the DTI-specific software (e.g., FSL topup (Andersson 2007). A short repetition time (e.g., 1,5 s or less) can provide a reliable et al., 2003). Finally, specific DTI related sources of variability such as rs-fMRI sequence to be potentially used in the clinical studies (Jahanian the eddy-current distortions, the Gibbs ringing artefacts (Perrone et al., et al., 2019), by reducing the impact of eye motion and physiological 2015) and those related to table vibration during the acquisition (Gal­ noise compared with longer (i.e., >2 s) EPI sequences (Yang et al., lichan et al., 2009) can be controlled by specific software tools (Max­ 2007). However, this requires the availability of efficient simultaneous imov et al., 2019). Same voxel size, b = 0 (i.e., no -diffusion weighting) multi-slice acquisition techniques on the scanner. The echo time selec­ volumes, echo time and repetition time should be implemented across tion is important for distinguishing between intra- and extra-vascular centers, to reduce the difference in SNR measure and harmonize the BOLD contributions (Ragot and Chen, 2019). Finally, larger voxel overall quality of the images (Laganà et al., 2010). sizes can potentially increase SNR (Bennett and Miller, 2010). Harmonization of MRI analysis Harmonization of rs-fMRI analysis Several software tools are able to derive FA and MD images from DTI In general, rs-fMRI analysis methods can be divided into two groups: data (Soares et al., 2013; Tournier et al., 2019). A comparison among model-dependent, such as seed-based methods (Andrews-Hanna et al., four of them on the reconstruction of the main WM tracts showed poor 2007), and model-free methods, including principal component analysis between-software agreement for FA, axial and radial diffusivity (Chris­ (PCA) (Friston et al., 1993), singular value decomposition (SVD) tidi et al., 2016). (Worsley et al., 2005) and ICA (Beckmann et al., 2005). In a recent study 6 N. De Stefano et al. NeuroImage: Clinical 34 (2022) 102972 (Botvinik-Nezer et al., 2020), 108 datasets with the same task (cogni­ included in the analysis remains mandatory. For example, if datasets of tive)-fMRI were independently analyzed by 70 teams over the world, both 3D and 2D-T1W are merged for atrophy studies a reliability anal­ and no two of them chose identical workflows for data analysis. ysis must be performed (Eshaghi et al., 2021). Different sources of variability were identified, from the smoothing level (the higher the spatial smoothing, the larger the activated region) to the 3. Image and analysis pipeline transfer and storage head-motion correction procedures and the software package used. Overall, this study showed that “analytical choices have a major effect Over the past few years, we moved from small studies with hundreds on reported results”. of subjects acquired cross-sectionally, to large multi-modal studies with Although AI methods have extensively been applied to extract sig­ thousands of subjects and several time points. Collaboration has nificant features for disease classification (Wen et al., 2018), they need increased among centers and consequently the need to merge different yet to be extensively validated on external datasets to prove reliability studies and manage large MRI datasets. This new scenario highlighted and robustness. the key role of efficient data storage, managing and sharing, which can Recommendations positively impact research results, ease longitudinal studies, facilitate The most useful recommendation to date is that a harmonized data transfer, allow multiple users to work on the same dataset, optimize dataset should be acquired on scanners with the same magnetic field computational resources and, ultimately, reduce costs. Recently, Dojat strength, using opposite phase-encoding directions and the same MRI et. al presented a comprehensive work analyzing different software and protocol. Both pre-processing (i.e., correction for head- cardiac- and hardware infrastructures for image transfer and storage (Dojat et al., eyes-related motion artefacts, correction for eddy currents, B0- and B1- 2017). All these features are essential in medical image-based research inhomogeneity) and post-processing analysis software tools should use in the big data era where there is an increasing demand to keep track of an identical and well documented pipeline. the projects to avoid starting from scratch every time that new people start working on the data. In this context, tools such as XNAT (eXtensible 2.6. Mega and Meta-analysis Neuroimaging Archive Toolkit) (Marcus et al., 2007) are platforms that support medical image-based research data handling by efficiently Two distinct strategies can be followed to statistically analyze MRI managing clinical and non-clinical data for multi-center studies or data data from multicenter studies, named mega- and meta-analysis. In the sharing. former, raw/individual data are pooled and analyzed centrally. In the Sometimes, however, it might not be possible to install the infra­ latter, data of each center are locally analyzed and only the summary structure to adopt a data management platform. Despite this, a reliable statistics from all centers are brought together, after weighting the role and robust data sharing between centres is still possible using a common of each center-specific statistic. Both procedures must indicate the folder and data structure. In that sense, BIDS (Brain Imaging Data criteria followed to perform the QA of the images before the analysis, in Structure) (Gorgolewski et al., 2016) has become a standard for anno­ order to avoid bias. This is particularly important when, as in the meta- tating and storing neuroimaging data. The majority of data management analysis procedure, data can be collected retrospectively, with the risk of platforms support importing or exporting data in BIDS format and BIDS merging MR images with substantial differences in quality. is becoming popular across the neuroimage community. Noteworthy, An example of mega-analysis is the creation of normative data for specific platforms such as XNAT do facilitate sharing of analysis pipe­ biomarkers such as atrophy rate (Battaglini et al., 2019) and volume of lines where the optimized settings for each procedure used are “frozen” deep GM structures (Potvin et al., 2016), where a single regression is and therefore the analysis can be easily run using an identical setting performed on the pooled MRI data. In these cases, statistical analysis from MR images of different centers. acts as a harmonization procedure per se considering, for example, Recommendations different vendors or magnetic field strength as confounding variables. The implementation of a state-of-the-art platform that efficiently However, statistical procedures, such as regressions, can be performed stores, manages and shares a large amount of medical image-based also in the context of data harmonization at voxel-level, for modelling research data that are being acquired nowadays is an investment that and subsequently removing the local unwanted inter-site variability. A will have a major mid- and long-term impact in daily work. The common notable example is ComBat (Fortin et al., 2016; Fortin et al., 2018), a and centralized data access ensures to avoid repetitions (i.e., to compute batch-effect correction tool used in genomics and recently implemented several times brain lesion filling and tissue segmentation), promotes for multi-center MRI data analysis. collaborations across research groups, keeps track of all the changes and In performing meta-analysis, different criteria of gauging the role of data manipulation, avoids data loss and helps to keep a fluent and each center-specific statistics can be employed. METAL (Willer et al., consistent workflow across people and along the projects. Finally, it is 2010), one of the most commonly used software platforms to perform strongly recommended to share the analysis pipeline, as it is known that meta-analysis, has implemented two strategies of weighting, which are the same dataset can provide non-comparable results if analyzed with based on the inverse variance and the sample size. Using METAL with a the same software but with different options (Botvinik-Nezer et al., sample size approach, recently a genome-wide association study 2020). (GWAS) discovered four new loci associated with hippocampal volume (Hibar et al., 2017). These jointed analyses, combining genomic and 4. Current experience on harmonization of big data imaging biomarkers, may have a great potential implication in the MS field. 4.1. Experience of ADNI and European initiatives on Alzheimer’s disease Recommendations Mega– and meta-analyses are able to reduce the impact of the het­ The Alzheimer’s disease neuroimaging initiative (ADNI) is currently erogeneity in acquisition protocols but have both pros and cons the most successful and valuable example of a freely accessible longi­ (Kochunov et al., 2014). On the one hand, mega-analysis, unlike meta- tudinal neuroimaging resource available for the study of Alzheimer’s analysis, allows considering the individual variability and reducing the disease (AD), mild cognitive impairment (MCI), and healthy aging brain inter-center variability, even at voxel level. On the other hand, meta- (Weiner et al., 2017). This public-private partnership, now at its fourth analysis can be used to easily analyze retrospective data and without wave of funding, has the primary goal to test whether serial imaging image data sharing, thus managing large MRI datasets without facing data (MRI and positron emission tomography [PET]) can be combined issues related to data organization and storage. Although both statistical with clinical, neuropsychological and biological markers to measure the approaches may be useful for merging images acquired in different progression from normal aging to AD (Weiner et al., 2017). The most centers with different protocols, a clear indication on the type of images important changes over time have been: (i) inclusion of new subjects 7 N. De Stefano et al. NeuroImage: Clinical 34 (2022) 102972 especially cognitively normal controls and patients with MCI in addition scanning 100,000 participants, the measurements are made in different to previously included subjects that are still “on study” (total expected locations equipped with two scanners each (one 3 T MAGNETOM Skyra number >1000), (ii) regular update of available datasets (last update: 10 and one 1.5 T MAGNETOM Aera). Scanning sites were selected to February 2020), (iii) continuous effort to provide uniform neuroimaging optimize volunteer recruitment from UK Biobank participants across the protocols for reaching an improved consistency in data analysis. As­ UK. The comprehensive brain program runs for about 30 min exclusively sessments are performed 6-monthly with now many years of follow-up on the 3 T systems while an integrated body and cardiac program is available for the original subjects. executed on the 1.5 T scanners for 10 and 20 min, respectively. From the release of ADNI dataset, a large amount of studies has In general, MRI at UK Biobank is subject to the same considerations enhanced our knowledge about AD so far, ranging from machine as other long-term investigations with respect to protocol design, data learning neuroimaging studies for a correct AD/MCI classification and storage and QA. However, the intended number of participants requires prediction of AD conversion (Dimitriadis et al., 2018) to discoveries far specific operational decisions to provide a most consistent dataset: 18 beyond imaging, such as showing the importance of genetic and bio­ participants are invited to each center every day. This implies that a logical factors for AD heterogeneity (Ramanan et al., 2021). delay or scan repeat must remain confined to the corresponding In terms of infrastructure, the free access to raw DICOM images participant and lead to a shortening of the imaging program. through the LONI website (http://adni.loni.usc.edu/) poses some chal­ Along with the examination protocols, a QA procedure was devel­ lenges of use, leading to duplication of efforts and inconsistent data oped with three main aspects to consider over time: status of scanners, usage. However, ADNI is under constant improvement and progressively actual image quality, and data completeness (Alfaro-Almagro et al., focusing on advanced MRI techniques and novel PET tracers. Indeed, in 2018). Specific to UK Biobank in comparison to many other large-scale the last version (ADNI-3), the scan protocol moved to accelerated 1 mm3 imaging studies is the use of the same type of scanner and software 3D T1-weighted sequences, from 2D T2-weighted to 3D T2-weighted version over the complete duration of the project. Identical procedures FLAIR sequences, a wide use of arterial spin labelling (ASL) and and protocol settings apply at all centers. This provides the potential to improved DTI and rs-fMRI protocols (Jack et al., 2015). differentiate the effect of hardware variations with a great precision and Similar European initiatives include PharmaCog (www.alzheimer- to study individual variations in detail. In order to further separate europe.org/Research/PharmaCo), which has a smaller dataset of technical from individual effects, phantom scans are executed daily and healthy controls and MCI patients with high quality MRI, with access the time characteristics of field and noise are analyzed. only through the principal investigators. Using this dataset, it has been shown that fMRI and DTI metrics are quite reproducible across centers 4.3. INNI experience (Jovicich et al., 2016). Recently, MRI structural biomarkers have been identified from this cohort to improve subject selection and to be used as Given the paramount importance of MRI to diagnose and monitor surrogate outcomes of disease progression. Among these, hippocampus MS, recent collaborations promoted the acquisition of relatively stan­ has been confirmed to be the most sensitive region to disease progres­ dard conventional MRI protocols. Good examples are the German sion in MCI with prodromal AD (Marizzoni et al., 2019). Competence Network Multiple Sclerosis (https://www.kompetenznetz- More recently, the European prevention of Alzheimer’s dementia multiplesklerose.de), the North American Registry for Care and (EPAD) program (www.ep-ad.org) aims to deliver a platform, adaptive, Research in Multiple Sclerosis (NARCRMS, https://www.narcrms.org/), phase 2 proof of concept trial for the secondary prevention of AD, the Canadian Prospective Cohort Study to Understand Progression in MS including a longitudinal cohort study (Ritchie et al., 2020). (CanProCo) (Dimitriadis et al., 2018) and the MS PATHS (Mowry et al., EPAD currently involves 21 centers across Europe, and it has 2020). With the exception of CanProCo, these initiatives do not include enrolled >2000 subjects with normal cognition or MCI with yearly advanced MRI techniques, such as DTI and rs-fMRI, which are very follow-up assessment including biological and cognitive markers. The useful to characterize microstructural damage and functional reorgani­ EPAD imaging protocol includes conventional MRI sequences for all zation occurring in this condition. The Italian Neuroimaging Network subjects, advanced techniques such as rsFMRI, DTI, ASL and suscepti­ Initiative (INNI) (www.inni-ms.org/) is one of the latest and promising bility weighted imaging (SWI), in >50% of the dataset with repeated data sharing initiatives in the field of MRI in MS. It started from four measurements (ten Kate et al., 2018). A subset of subjects will also have Italian research centers with recognized expertise in the study of neu­ amyloid-PET for a specific substudy called AMYPAD (www.amypad.eu). roimaging in MS (Filippi et al., 2017), with the support of the Italian MS The EPAD and AMYPAD data are structured in XNAT and will be made Society, to promote the collection of conventional and advanced struc­ publicly available in the future. tural and functional (including DTI and rs-fMRI) MRI scans as well as clinical and neuropsychological data in a centralized repository. MRI 4.2. UK Biobank experience data are added from each centre both retrospectively and prospectively. All imaging biomarkers are analyzed using advanced procedures for UK Biobank (www.ukbiobank.ac.uk) is a large prospective cohort harmonization. study that recruited 500,000 people aged between 40 and 69 years in The main goals of INNI are to identify and validate novel MRI bio­ 2006–2010 from across UK. Participants underwent a range of physical markers, which can be used as predictors and/or outcomes in future MS measures, provided their details, and gave blood, urine, and saliva studies, and to support a standardized use of neuroimaging protocols in samples for future analyses. They also agreed to be followed through MS at national level. According to the first INNI survey in 2017, >1300 linkage to their medical records. The proposal for an imaging subjects were uploaded on the online platform (Filippi et al., 2017). This enhancement in 100,000 participants was made in 2012. Meanwhile, number has progressively increased over time reaching, to date, >2500 three imaging centers have been operating (Stockport, Newcastle, MS patients and healthy controls (with>4200 MRI scans, including Reading) and over 50,000 participants have been scanned. Access to the longitudinal assessments). Although similar MR acquisition protocols imaging datasets is provided (since March 2020) by previous examina­ were implemented by different centers and data have to fulfil minimum tion of the researcher and its research proposal. The simultaneous im­ requirements to be included, a full standardization of the MRI acquisi­ aging of brain, heart, arteries, abdomen, and bone in each individual and tion protocols across sites was not requested, at least in the first project the combination of these data will provide a dataset that is uniquely able phase (Filippi et al., 2017). However, a systematic quantitative QA to address questions regarding the relationships between phenotypes procedure, which is essential to ensure rigorous results, was recently across organs and to assess pathogenic mechanisms operating at a sys­ implemented for all data uploaded to the INNI repository, as an initial temic level. pre-processing step (Storelli et al., 2019). Thus, when centers upload the To meet the temporal, financial, and geographic constraints of MRI scans on the INNI platform, an image QA is performed within 8 N. De Stefano et al. NeuroImage: Clinical 34 (2022) 102972 database content to rate image artefacts, distortions and in­ Examples of how currently available large MRI datasets face the issue of data harmonization from the acquisition to the storage point of view. The tick indicates that the issue is completely (green) or partially (yellow) Availability of mechanisms homogeneities. Only scans with an adequate image quality and a certain security of the stored data degree of harmonization will be used for the INNI research projects. to ensure privacy and Table 2 summarizes the main problems related to harmonization and if they have been addressed or not addressed yet by these large MRI datasets. ^ Software containers, or dockers, are application build and deployment tool. They are based on the idea that you can package your code with dependencies into a deployable unit called container. 5. Future perspectives and conclusions Multicenter demographic, clinical and MRI data, ideally combined Automatic control of the conformity of imported with blood, cerebrospinal fluid and genetic biomarkers are undoubtedly data with a predefined needed to create the required large datasets to generate models based on artificial intelligence, which should be able to predict disability status study protocol and treatment response/failure in MS patients and thus provide a personalized management of patients. However, one of the major challenges to use MRI measures with this approach is the lack of harmonization on acquisition parameters, post-processing analysis tools and storage of the huge amount of imaging data. pipelines in software containers^ Homogeneous acquisition of MR images cross-sectionally and to be shared between centers longitudinally across centers is still an unsolved issue, and this article Incorporation of analysis analyses some solutions that could minimize this problem. These solu­ tions should be, ideally, independent of MR vendor and field strength, and of software and hardware upgrades in order to be widely imple­ mented. There is no doubt that the first requirement is to use a core standardized MRI protocol that should be obtained in a reasonable acquisition time in both clinical and research studies. Different scientific organizations in Europe (MAGNIMS) and in North America (CMSC, NAIMS) have actively worked to provide standardized protocols that can Availability of pipeline of be easily implemented in clinical practice (Wattjes et al., 2021a). In this analyses for testing the context, the use of synthetic images (e.g., Fingerprinting (Hsieh and Svalbe, 2020), Synthetic MR (Gonçalves et al., 2018) could be of reproducibility extremely high value as they can provide harmonized, co-registered and simultaneous measures of different contrast-weighted images (Iglesias et al., 2021). Another essential requirement is to harmonize and validate the different post-processing tools that will facilitate the use of MRI quan­ criteria for inclusion/exclusion titative data for both clinical and research purposes. Ideally, these tools of images depending on the should be available online, open to QA, and harmonized among different Availability of objective MR vendors. Availability of post-processing validated pipeline standards endpoint and software and analysis methods for specific aspects is an unmet need, which clearly requires the cooperation between academic medical centers, biopharmaceutical and bioinformatic companies and MR vendors. This is undoubtedly a win-win strategy between all these partners, whose final goal is to use quantitative and robust data to incorporate in learning health system models intended to provide continual improve­ ment in patient care, particularly regarding personalized selection of Standardization of optimal treatment initiation and sequencing. *Possible loss of privacy for genetic material to identify relatives. quality control procedures for Finally, one recent challenge for researchers using medical imaging is storing, indexing, and sharing their data, which requires professionals with information technology skills, and experts in the complex legal faced; the red cross means that the issue is still open. regulations on data management transfer. The complexity of the above- mentioned tasks overwhelms academic researchers and requires tech­ parameters and implementation nical and human resources that should be provided by the academical Standardization of acquisition Steps towards harmonization centers to support research efforts. Huge efforts are currently being undertaken in order to achieve all the requirements needed to harmo­ nize the acquisition, analysis, storage and secure transfer of imaging in clinical protocols data. Proper management of large imaging datasets is one of our greatest opportunities and challenges in the coming years. This is particularly important for offering MS patients an optimal individualized manage­ ment while minimizing consumption of resources. CRediT authorship contribution statement MRI datasets Uk Biobank Nicola De Stefano: Conceptualization, Methodology, Writing – Table 2 ADNI original draft, Writing – review & editing. Marco Battaglini: Method­ INNI ology, Writing – original draft, Writing – review & editing. Deborah 9 N. De Stefano et al. NeuroImage: Clinical 34 (2022) 102972 Pareto: Methodology, Writing – original draft, Writing – review & enhanced, and gradient spin echo brain MRI scans from multiple sclerosis patients. Magn Reson Imaging 17 (8), 1105–1110. editing. Rosa Cortese: Methodology, Writing – original draft, Writing – Egger, C., Opfer, R., Wang, C., Kepp, T., Sormani, M.P., Spies, L., Barnett, M., review & editing. Jian Zhang: Methodology, Writing – original draft, Schippling, S., 2017. MRI FLAIR lesion segmentation in multiple sclerosis: Does Writing – review & editing. Niels Oesingmann: Methodology, Writing – automated segmentation hold up with manual annotation? Neuroimage Clin 13, original draft, Writing – review & editing. Ferran Prados: Methodology, 264–270. https://doi.org/10.1016/j.nicl.2016.11.020. De Leener, B., Lévy, S., Dupont, S.M., Fonov, V.S., Stikov, N., Louis Collins, D., Callot, V., Writing – original draft, Writing – review & editing. Maria A. Rocca: Cohen-Adad, J., 2017. SCT: Spinal Cord Toolbox, an open-source software for Methodology, Writing – original draft, Writing – review & editing. Paola processing spinal cord MRI data. Neuroimage 145, 24–43. https://doi.org/10.1016/ Valsasina: Methodology, Writing – original draft, Writing – review & j.neuroimage.2016.10.009. Gros, C., De Leener, B., Badji, A., Maranzano, J., Eden, D., Dupont, S.M., Talbott, J., editing. Hugo Vrenken: Methodology, Writing – original draft, Writing Zhuoquiong, R., Liu, Y., Granberg, T., Ouellette, R., Tachibana, Y., Hori, M., – review & editing. Claudia A.M. Gandini Wheeler-Kingshott: Meth­ Kamiya, K., Chougar, L., Stawiarz, L., Hillert, J., Bannier, E., Kerbrat, A., Edan, G., odology, Writing – original draft, Writing – review & editing. Massimo Labauge, P., Callot, V., Pelletier, J., Audoin, B., Rasoanandrianina, H., Brisset, J.-C., Valsasina, P., Rocca, M.A., Filippi, M., Bakshi, R., Tauhid, S., Prados, F., Filippi: Methodology, Writing – original draft, Writing – review & Yiannakas, M., Kearney, H., Ciccarelli, O., Smith, S., Treaba, C.A., Mainero, C., editing. Frederik Barkhof: Methodology, Writing – original draft, Lefeuvre, J., Reich, D.S., Nair, G., Auclair, V., McLaren, D.G., Martin, A.R., Writing – review & editing. Àlex Rovira: Methodology, Writing – Fehlings, M.G., Vahdat, S., Khatibi, A., Doyon, J., Shepherd, T., Charlson, E., Narayanan, S., Cohen-Adad, J., 2019. Automatic segmentation of the spinal cord and original draft, Writing – review & editing. intramedullary multiple sclerosis lesions with convolutional neural networks. Neuroimage 184, 901–915. https://doi.org/10.1016/j.neuroimage.2018.09.081. Cohen-Adad, J., Alonso-Ortiz, E., Abramovic, M., Arneitz, C., Atcheson, N., Barlow, L., Declaration of Competing Interest Barry, R.L., Barth, M., Battiston, M., Büchel, C., Budde, M., Callot, V., Combes, A.J. E., De Leener, B., Descoteaux, M., de Sousa, P.L., Dostál, M., Doyon, J., Dvorak, A., Eippert, F., Epperson, K.R., Epperson, K.S., Freund, P., Finsterbusch, J., Foias, A., The authors declare that they have no known competing financial Fratini, M., Fukunaga, I., Wheeler-Kingshott, C.A.M.G., Germani, G., Gilbert, G., interests or personal relationships that could have appeared to influence Giove, F., Gros, C., Grussu, F., Hagiwara, A., Henry, P.-G., Horák, T., Hori, M., the work reported in this paper. Joers, J., Kamiya, K., Karbasforoushan, H., Keřkovský, M., Khatibi, A., Kim, J.-W., Kinany, N., Kitzler, H., Kolind, S., Kong, Y., Kudlička, P., Kuntke, P., Kurniawan, N. D., Kusmia, S., Labounek, R., Laganà, M.M., Laule, C., Law, C.S., Lenglet, C., References Leutritz, T., Liu, Y., Llufriu, S., Mackey, S., Martinez-Heras, E., Mattera, L., Nestrasil, I., O’Grady, K.P., Papinutto, N., Papp, D., Pareto, D., Parrish, T.B., Pichiecchio, A., Prados, F., Rovira, À., Ruitenberg, M.J., Samson, R.S., Savini, G., Shinohara, R.T., Oh, J., Nair, G., Calabresi, P.A., Davatzikos, C., Doshi, J., Henry, R.G., Seif, M., Seifert, A.C., Smith, A.K., Smith, S.A., Smith, Z.A., Solana, E., Suzuki, Y., Kim, G., Linn, K.A., Papinutto, N., Pelletier, D., Pham, D.L., Reich, D.S., Rooney, W., Tackley, G., Tinnermann, A., Valošek, J., Van De Ville, D., Yiannakas, M.C., Roy, S., Stern, W., Tummala, S., Yousuf, F., Zhu, A., Sicotte, N.L., Bakshi, R., 2017. Weber, K.A., Weiskopf, N., Wise, R.G., Wyss, P.O., Xu, J., 2021a. Generic acquisition Volumetric Analysis from a Harmonized Multisite Brain MRI Study of a Single protocol for quantitative MRI of the spinal cord. Nat Protoc 16 (10), 4611–4632. Subject with Multiple Sclerosis. AJNR Am J Neuroradiol 38 (8), 1501–1509. https:// https://doi.org/10.1038/s41596-021-00588-0. doi.org/10.3174/ajnr.A5254. Cohen-Adad, J., Alonso-Ortiz, E., Abramovic, M., Arneitz, C., Atcheson, N., Barlow, L., Biberacher, V., Schmidt, P., Keshavan, A., Boucard, C.C., Righart, R., Sämann, P., Barry, R.L., Barth, M., Battiston, M., Büchel, C., Budde, M., Callot, V., Combes, A.J. Preibisch, C., Fröbel, D., Aly, L., Hemmer, B., Zimmer, C., Henry, R.G., Mühlau, M., E., De Leener, B., Descoteaux, M., de Sousa, P.L., Dostál, M., Doyon, J., Dvorak, A., 2016. Intra- and interscanner variability of magnetic resonance imaging based Eippert, F., Epperson, K.R., Epperson, K.S., Freund, P., Finsterbusch, J., Foias, A., volumetry in multiple sclerosis. Neuroimage 142, 188–197. https://doi.org/ Fratini, M., Fukunaga, I., Gandini Wheeler-Kingshott, C.A.M., Germani, G., 10.1016/j.neuroimage.2016.07.035. Gilbert, G., Giove, F., Gros, C., Grussu, F., Hagiwara, A., Henry, P.-G., Horák, T., Dewey, B.E., Zhao, C., Reinhold, J.C., Carass, A., Fitzgerald, K.C., Sotirchos, E.S., Hori, M., Joers, J., Kamiya, K., Karbasforoushan, H., Keřkovský, M., Khatibi, A., Saidha, S., Oh, J., Pham, D.L., Calabresi, P.A., van Zijl, P.C.M., Prince, J.L., 2019. Kim, J.-W., Kinany, N., Kitzler, H.H., Kolind, S., Kong, Y., Kudlička, P., Kuntke, P., DeepHarmony: A deep learning approach to contrast harmonization across scanner Kurniawan, N.D., Kusmia, S., Labounek, R., Laganà, M.M., Laule, C., Law, C.S., changes. Magn Reson Imaging 64, 160–170. https://doi.org/10.1016/j. Lenglet, C., Leutritz, T., Liu, Y., Llufriu, S., Mackey, S., Martinez-Heras, E., mri.2019.05.041. Mattera, L., Nestrasil, I., O’Grady, K.P., Papinutto, N., Papp, D., Pareto, D., Wattjes, M.P., Ciccarelli, O., Reich, D.S., Banwell, B., de Stefano, N., Enzinger, C., Parrish, T.B., Pichiecchio, A., Prados, F., Rovira, À., Ruitenberg, M.J., Samson, R.S., Fazekas, F., Filippi, M., Frederiksen, J., Gasperini, C., Hacohen, Y., Kappos, L., Li, D. Savini, G., Seif, M., Seifert, A.C., Smith, A.K., Smith, S.A., Smith, Z.A., Solana, E., K.B., Mankad, K., Montalban, X., Newsome, S.D., Oh, J., Palace, J., Rocca, M.A., Suzuki, Y., Tackley, G., Tinnermann, A., Valošek, J., Van De Ville, D., Yiannakas, M. Sastre-Garriga, J., Tintoré, M., Traboulsee, A., Vrenken, H., Yousry, T., Barkhof, F., C., Weber II, K.A., Weiskopf, N., Wise, R.G., Wyss, P.O., Xu, J., 2021b. Open-access Rovira, À., Wattjes, M.P., Ciccarelli, O., de Stefano, N., Enzinger, C., Fazekas, F., quantitative MRI data of the spinal cord and reproducibility across participants, sites Filippi, M., Frederiksen, J., Gasperini, C., Hacohen, Y., Kappos, L., Mankad, K., and manufacturers. Sci Data 8 (1). https://doi.org/10.1038/s41597-021-00941-8. Montalban, X., Palace, J., Rocca, M.A., Sastre-Garriga, J., Tintore, M., Vrenken, H., Breckwoldt, M.O., Gradl, J., Hähnel, S., Hielscher, T., Wildemann, B., Diem, R., Yousry, T., Barkhof, F., Rovira, A., Li, D.K.B., Traboulsee, A., Newsome, S.D., Platten, M., Wick, W., Heiland, S., Bendszus, M., 2017. Increasing the sensitivity of Banwell, B., Oh, J., Reich, D.S., Reich, D.S., Oh, J., 2021a. 2021 MAGNIMS-CMSC- MRI for the detection of multiple sclerosis lesions by long axial coverage of the spinal NAIMS consensus recommendations on the use of MRI in patients with multiple cord: A prospective study in 119 patients. J Neurol 264 (2), 341–349. https://doi. sclerosis. Lancet Neurol 20 (8), 653–670. https://doi.org/10.1016/S1474-4422(21) org/10.1007/s00415-016-8353-3. 00095-8. Karavasilis, E., Velonakis, G., Argiropoulos, G., Athanasakos, A., Poulou, L.S., Toulas, P., Vrenken, H., Jenkinson, M., Horsfield, M.A., Battaglini, M., van Schijndel, R.A., Kelekis, N.L., Efstathopoulos, E.P., 2019. Proton Density Fat Suppressed MRI in 3T Rostrup, E., Geurts, J.J.G., Fisher, E., Zijdenbos, A., Ashburner, J., Miller, D.H., Increases the Sensitivity of Multiple Sclerosis Lesion Detection in the Cervical Spinal Filippi, M., Fazekas, F., Rovaris, M., Rovira, A., Barkhof, F., de Stefano, N., 2013. Cord. Clin Neuroradiol 29 (1), 45–50. https://doi.org/10.1007/s00062-017-0626-4. Recommendations to improve imaging and analysis of brain lesion load and atrophy Di Perri, C., Dwyer, M.G., Wack, D.S., et al., 2009. Signal abnormalities on 1.5 and 3 in longitudinal studies of multiple sclerosis. J Neurol 260 (10), 2458–2471. https:// Tesla brain MRI in multiple sclerosis patients and healthy controls. A morphological doi.org/10.1007/s00415-012-6762-5. and spatial quantitative comparison study. Neuroimage 47, 1352–1362. https://doi. Thompson, A.J., Banwell, B.L., Barkhof, F., Carroll, W.M., Coetzee, T., Comi, G., org/10.1016/j.neuroimage.2009.04.019. Correale, J., Fazekas, F., Filippi, M., Freedman, M.S., Fujihara, K., Galetta, S.L., Stankiewicz, J.M., Glanz, B.I., Healy, B.C., Arora, A., Neema, M., Benedict, R.H.B., Hartung, H.P., Kappos, L., Lublin, F.D., Marrie, R.A., Miller, A.E., Miller, D.H., Guss, Z.D., Tauhid, S., Buckle, G.J., Houtchens, M.K., Khoury, S.J., Weiner, H.L., Montalban, X., Mowry, E.M., Sorensen, P.S., Tintoré, M., Traboulsee, A.L., Guttmann, C.R.G., Bakshi, R., 2011. Brain MRI lesion load at 1.5T and 3T versus Trojano, M., Uitdehaag, B.M.J., Vukusic, S., Waubant, E., Weinshenker, B.G., clinical status in multiple sclerosis. J Neuroimaging 21 (2), e50–e56. https://doi. Reingold, S.C., Cohen, J.A., 2018. Diagnosis of multiple sclerosis: 2017 revisions of org/10.1111/j.1552-6569.2009.00449.x. the McDonald criteria. Lancet Neurol 17 (2), 162–173. https://doi.org/10.1016/ Sajja, B.R., Datta, S., He, R., Mehta, M., Gupta, R.K., Wolinsky, J.S., Narayana, P.A., S1474-4422(17)30470-2. 2006. Unified approach for multiple sclerosis lesion segmentation on brain MRI. Ann Wattjes MP, Ciccarelli O, Reich DS, et al (2021) International 2020 MAGNIMS-CMSC- Biomed Eng 34 (1), 142–151. https://doi.org/10.1007/s10439-005-9009-0. NAIMS consensus recommendations on the use of MRI in multiple sclerosis. The Molyneux, P.D., Tubridy, N., Parker, G.J., et al., 1998. The effect of section thickness on Lancet Neurology in press. MR lesion detection and quantification in multiple sclerosis. AJNR Am J Neuroradiol Hagens, M.HJ., Burggraaff, J., Kilsdonk, I.D., Ruggieri, S., Collorone, S., Cortese, R., 19, 1715–1720. Cawley, N., Sbardella, E., Andelova, M., Amann, M., Lieb, J.M., Pantano, P., Filippi, M., Preziosa, P., Banwell, B.L., Barkhof, F., Ciccarelli, O., De Stefano, N., Lissenberg-Witte, B.I., Killestein, J., Oreja-Guevara, C., Wuerfel, J., Ciccarelli, O., Geurts, J.J.G., Paul, F., Reich, D.S., Toosy, A.T., Traboulsee, A., Wattjes, M.P., Gasperini, C., Lukas, C., Rovira, A., Barkhof, F., Wattjes, M.P., 2019. Impact of 3 Yousry, T.A., Gass, A., Lubetzki, C., Weinshenker, B.G., Rocca, M.A., 2019. Tesla MRI on interobserver agreement in clinically isolated syndrome: A MAGNIMS Assessment of lesions on magnetic resonance imaging in multiple sclerosis: Practical multicentre study. Mult Scler 25 (3), 352–360. https://doi.org/10.1177/ guidelines. Brain 142 (7), 1858–1875. https://doi.org/10.1093/brain/awz144. 1352458517751647. Wrobel, J., Martin, M.L., Bakshi, R., Calabresi, P.A., Elliot, M., Roalf, D., Gur, R.C., Rovaris, M., Rocca, M.A., Yousry, I., Yousry, T.A., Colombo, B., Comi, G., Filippi, M., Gur, R.E., Henry, R.G., Nair, G., Oh, J., Papinutto, N., Pelletier, D., Reich, D.S., 1999. Lesion load quantification on fast-FLAIR, rapid acquisition relaxation- 10 N. De Stefano et al. NeuroImage: Clinical 34 (2022) 102972 Rooney, W.D., Satterthwaite, T.D., Stern, W., Prabhakaran, K., Sicotte, N.L., Harmonization validity for morphometry measurements. Neuroimage Clin 24, Shinohara, R.T., Goldsmith, J., 2020. Intensity warping for multisite MRI 101943. https://doi.org/10.1016/j.nicl.2019.101943. harmonization. Neuroimage 223, 117242. https://doi.org/10.1016/j. Liu, Y., Lukas, C., Steenwijk, M.D., Daams, M., Versteeg, A., Duan, Y., Li, K., Weiler, F., neuroimage.2020.117242. Hahn, H.K., Wattjes, M.P., Barkhof, F., Vrenken, H., 2016. Multicenter Validation of Zeng, C., Gu, L., Liu, Z., Zhao, S., 2020. Review of Deep Learning Approaches for the Mean Upper Cervical Cord Area Measurements from Head 3D T1-Weighted MR Segmentation of Multiple Sclerosis Lesions on Brain MRI. Front Neuroinform 14, Imaging in Patients with Multiple Sclerosis. AJNR Am J Neuroradiol 37 (4), 610967. https://doi.org/10.3389/fninf.2020.610967. 749–754. https://doi.org/10.3174/ajnr.A4635. Mårtensson, G., Ferreira, D., Granberg, T., Cavallin, L., Oppedal, K., Padovani, A., Papinutto, N., Bakshi, R., Bischof, A., Calabresi, P.A., Caverzasi, E., Constable, R.T., Rektorova, I., Bonanni, L., Pardini, M., Kramberger, M.G., Taylor, J.-P., Hort, J., Datta, E., Kirkish, G., Nair, G., Oh, J., Pelletier, D., Pham, D.L., Reich, D.S., Snædal, J., Kulisevsky, J., Blanc, F., Antonini, A., Mecocci, P., Vellas, B., Tsolaki, M., Rooney, W., Roy, S., Schwartz, D., Shinohara, R.T., Sicotte, N.L., Stern, W.A., Kłoszewska, I., Soininen, H., Lovestone, S., Simmons, A., Aarsland, D., Westman, E., Tagge, I., Tauhid, S., Tummala, S., Henry, R.G., 2018. Gradient nonlinearity effects 2020. The reliability of a deep learning model in clinical out-of-distribution MRI on upper cervical spinal cord area measurement from 3D T1 -weighted brain MRI data: A multicohort study. Med Image Anal 66, 101714. https://doi.org/10.1016/j. acquisitions. Magn Reson Med 79 (3), 1595–1601. https://doi.org/10.1002/ media.2020.101714. mrm.26776. de Sitter, A., Visser, M., Brouwer, I., Cover, K.S., van Schijndel, R.A., Eijgelaar, R.S., Weeda, M.M., Middelkoop, S.M., Steenwijk, M.D., Daams, M., Amiri, H., Brouwer, I., Müller, D.M.J., Ropele, S., Kappos, L., Rovira, Á., Filippi, M., Enzinger, C., Killestein, J., Uitdehaag, B.M.J., Dekker, I., Lukas, C., Bellenberg, B., Barkhof, F., Frederiksen, J., Ciccarelli, O., Guttmann, C.R.G., Wattjes, M.P., Witte, M.G., de Witt Pouwels, P.J.W., Vrenken, H., 2019b. Validation of mean upper cervical cord area Hamer, P.C., Barkhof, F., Vrenken, H., 2020. Facing privacy in neuroimaging: (MUCCA) measurement techniques in multiple sclerosis (MS): High reproducibility Removing facial features degrades performance of image analysis methods. Eur and robustness to lesions, but large software and scanner effects. Neuroimage Clin Radiol 30 (2), 1062–1074. https://doi.org/10.1007/s00330-019-06459-3. 24, 101962. https://doi.org/10.1016/j.nicl.2019.101962. Weeda, M.M., Brouwer, I., de Vos, M.L., de Vries, M.S., Barkhof, F., Pouwels, P.J.W., Lukas, C., Bellenberg, B., Prados, F., et al., 2021. Quantification of cervical cord cross- Vrenken, H., 2019a. Comparing lesion segmentation methods in multiple sclerosis: sectional area: Which acquisition, vertebra level and analysis software? A multi- Input from one manually delineated subject is sufficient for accurate lesion center repeatability study on a travelling healthy volunteer. Front Neurol 12. segmentation. NeuroImage: Clinical 24, 102074. https://doi.org/10.1016/j. https://doi.org/10.3389/fneur.2021.693333. nicl.2019.102074. Papinutto, N., Henry, R.G., 2019. Evaluation of Intra- and Interscanner Reliability of MRI Valverde, S., Salem, M., Cabezas, M., Pareto, D., Vilanova, J.C., Ramió-Torrentà, L., Protocols for Spinal Cord Gray Matter and Total Cross-Sectional Area Measurements. Rovira, À., Salvi, J., Oliver, A., Lladó, X., 2019. One-shot domain adaptation in J Magn Reson Imaging 49 (4), 1078–1090. https://doi.org/10.1002/jmri.26269. multiple sclerosis lesion segmentation using convolutional neural networks. Battaglini, M., Jenkinson, M., De Stefano, N., 2018. SIENA-XL for improving the Neuroimage Clin 21, 101638. https://doi.org/10.1016/j.nicl.2018.101638. assessment of gray and white matter volume changes on brain MRI. Human Brain Sastre-Garriga, J., Pareto, D., Battaglini, M., Rocca, M.A., Ciccarelli, O., Enzinger, C., Mapping 39 (3), 1063–1077. https://doi.org/10.1002/hbm.23828. Wuerfel, J., Sormani, M.P., Barkhof, F., Yousry, T.A., De Stefano, N., Tintoré, M., Nakamura, K., Fisher, E., 2009. Segmentation of brain magnetic resonance images for Filippi, M., Gasperini, C., Kappos, L., Río, J., Frederiksen, J., Palace, J., Vrenken, H., measurement of gray matter atrophy in multiple sclerosis patients. Neuroimage 44 Montalban, X., Rovira, À., 2020. MAGNIMS consensus recommendations on the use (3), 769–776. https://doi.org/10.1016/j.neuroimage.2008.09.059. of brain and spinal cord atrophy measures in clinical practice. Nature Reviews Sdika, M., Pelletier, D., 2009. Nonrigid registration of multiple sclerosis brain images Neurology 16 (3), 171–182. https://doi.org/10.1038/s41582-020-0314-x. using lesion inpainting for morphometry or lesion mapping. Hum Brain Mapp 30 (4), Steenwijk, M.D., Vrenken, H., Jonkman, L.E., Daams, M., Geurts, J.JG., Barkhof, F., 1060–1067. https://doi.org/10.1002/hbm.20566. Pouwels, P.JW., 2016. High-resolution T1-relaxation time mapping displays subtle, Battaglini, M., Jenkinson, M., De Stefano, N., 2012. Evaluating and reducing the impact clinically relevant, gray matter damage in long-standing multiple sclerosis. Mult of white matter lesions on brain volume measurements. Hum Brain Mapp 33 (9), Scler 22 (10), 1279–1288. https://doi.org/10.1177/1352458515615953. 2062–2071. https://doi.org/10.1002/hbm.21344. Takao, H., Hayashi, N., Ohtomo, K., 2011. Effect of scanner in longitudinal studies of Guizard, N., Fonov, V.S., García-Lorenzo, D., Nakamura, K., Aubert-Broche, B., brain volume changes. J Magn Reson Imaging 34 (2), 438–444. https://doi.org/ Collins, D.L., Muñoz-Barrutia, A., 2015. Spatio-Temporal Regularization for 10.1002/jmri.22636. Longitudinal Registration to Subject-Specific 3d Template. PLoS One 10 (8), Lee, H., Nakamura, K., Narayanan, S., Brown, R.A., Arnold, D.L., 2019. Estimating and e0133352. https://doi.org/10.1371/journal.pone.0133352. accounting for the effect of MRI scanner changes on longitudinal whole-brain Popescu, V., Schoonheim, M.M., Versteeg, A., Chaturvedi, N., Jonker, M., Xavier de volume change measurements. NeuroImage 184, 555–565. https://doi.org/ Menezes, R., Gallindo Garre, F., Uitdehaag, B.M.J., Barkhof, F., Vrenken, H., 10.1016/j.neuroimage.2018.09.062. Derfuss, T., 2016. Grey Matter Atrophy in Multiple Sclerosis: Clinical Interpretation Steenwijk, M.D., Amiri, H., Schoonheim, M.M., de Sitter, A., Barkhof, F., Pouwels, P.J. Depends on Choice of Analysis Method. PLoS ONE 11 (1), e0143942. https://doi. W., Vrenken, H., 2017. Agreement of MSmetrix with established methods for org/10.1371/journal.pone.0143942. measuring cross-sectional and longitudinal brain atrophy. NeuroImage: Clinical 15, Battaglini, M., Giorgio, A., Stromillo, M.L., Bartolozzi, M.L., Guidi, L., Federico, A., De 843–853. https://doi.org/10.1016/j.nicl.2017.06.034. Stefano, N., 2009. Voxel-wise assessment of progression of regional brain atrophy in Goodkin, O., Prados, F., Vos, S.B., Pemberton, H., Collorone, S., Hagens, M.H.J., relapsing-remitting multiple sclerosis. J Neurol Sci 282 (1-2), 55–60. https://doi. Cardoso, M.J., Yousry, T.A., Thornton, J.S., Sudre, C.H., Barkhof, F., 2021. FLAIR- org/10.1016/j.jns.2009.02.322. only joint volumetric analysis of brain lesions and atrophy in clinically isolated Storelli, L., Rocca, M.A., Pagani, E., Van Hecke, W., Horsfield, M.A., De Stefano, N., syndrome (CIS) suggestive of multiple sclerosis. Neuroimage Clin 29, 102542. Rovira, A., Sastre-Garriga, J., Palace, J., Sima, D., Smeets, D., Filippi, M., 2018. https://doi.org/10.1016/j.nicl.2020.102542. Measurement of Whole-Brain and Gray Matter Atrophy in Multiple Sclerosis: Caramanos, Z., Fonov, V.S., Francis, S.J., Narayanan, S., Pike, G.B., Collins, D.L., Assessment with MR Imaging. Radiology 288 (2), 554–564. https://doi.org/ Arnold, D.L., 2010. Gradient distortions in MRI: Characterizing and correcting for 10.1148/radiol.2018172468. their effects on SIENA-generated measures of brain volume change. Neuroimage 49 Popescu, V., Battaglini, M., Hoogstrate, W.S., Verfaillie, S.C.J., Sluimer, I.C., van (2), 1601–1611. https://doi.org/10.1016/j.neuroimage.2009.08.008. Schijndel, R.A., van Dijk, B.W., Cover, K.S., Knol, D.L., Jenkinson, M., Barkhof, F., de Jovicich, J., Czanner, S., Greve, D., Haley, E., van der Kouwe, A., Gollub, R., Kennedy, D., Stefano, N., Vrenken, H., 2012. Optimizing parameter choice for FSL-Brain Schmitt, F., Brown, G., MacFall, J., Fischl, B., Dale, A., 2006. Reliability in multi-site Extraction Tool (BET) on 3D T1 images in multiple sclerosis. Neuroimage 61 (4), structural MRI studies: Effects of gradient non-linearity correction on phantom and 1484–1494. https://doi.org/10.1016/j.neuroimage.2012.03.074. human data. Neuroimage 30 (2), 436–443. https://doi.org/10.1016/j. Lutkenhoff, E.S., Rosenberg, M., Chiang, J., Zhang, K., Pickard, J.D., Owen, A.M., neuroimage.2005.09.046. Monti, M.M., Wu, X., 2014. Optimized brain extraction for pathological brains Doran, S.J., Charles-Edwards, L., Reinsberg, S.A., Leach, M.O., 2005. A complete (optiBET). PLoS One 9 (12), e115551. https://doi.org/10.1371/journal. distortion correction for MR images: I. Gradient warp correction. Phys Med Biol 50 pone.0115551. (7), 1343–1361. https://doi.org/10.1088/0031-9155/50/7/001. Gabr, R.E., Coronado, I., Robinson, M., Sujit, S.J., Datta, S., Sun, X., Allen, W.J., Song, S., Zheng, Y., He, Y., 2017. A review of Methods for Bias Correction in Medical Lublin, F.D., Wolinsky, J.S., Narayana, P.A., 2020. Brain and lesion segmentation in Images. BMER 3 (1). https://doi.org/10.18103/bme10.18103/bme.V3. multiple sclerosis using fully convolutional neural networks: A large-scale study. N110.18103/bme.v3i1.1550. Mult Scler 26 (10), 1217–1226. https://doi.org/10.1177/1352458519856843. Sled, J.G., Zijdenbos, A.P., Evans, A.C., 1998. A nonparametric method for automatic McKinley, R., Wepfer, R., Aschwanden, F., Grunder, L., Muri, R., Rummel, C., Verma, R., correction of intensity nonuniformity in MRI data. IEEE Trans Med Imaging 17, Weisstanner, C., Reyes, M., Salmen, A., Chan, A., Wagner, F., Wiest, R., 2021. 87–97. https://doi.org/10.1109/42.668698. Simultaneous lesion and brain segmentation in multiple sclerosis using deep neural Tustison, N.J., Avants, B.B., Cook, P.A., Yuanjie Zheng, Egan, A., Yushkevich, P.A., networks. Sci Rep 11 (1). https://doi.org/10.1038/s41598-020-79925-4. Gee, J.C., 2010. N4ITK: Improved N3 Bias Correction. IEEE Transactions on Medical Narayana, P.A., Coronado, I., Sujit, S.J., Wolinsky, J.S., Lublin, F.D., Gabr, R.E., 2020. Imaging 29 (6), 1310–1320. https://doi.org/10.1109/TMI.2010.2046908. Deep-Learning-Based Neural Tissue Segmentation of MRI in Multiple Sclerosis: Effect Potvin, O., Mouiha, A., Dieumegarde, L., Duchesne, S., 2016. Normative data for of Training Set Size. J Magn Reson Imaging 51 (5), 1487–1496. https://doi.org/ subcortical regional volumes over the lifetime of the adult human brain. NeuroImage 10.1002/jmri.26959. 137, 9–20. https://doi.org/10.1016/j.neuroimage.2016.05.016. Horsfield, M.A., Sala, S., Neema, M., Absinta, M., Bakshi, A., Sormani, M.P., Rocca, M.A., Battaglini, M., Gentile, G., Luchetti, L., et al., 2019. Lifespan normative data on rates of Bakshi, R., Filippi, M., 2010. Rapid semi-automatic segmentation of the spinal cord brain volume changes. Neurobiology of Aging 81, 30–37. https://doi.org/10.1016/j. from magnetic resonance images: Application in multiple sclerosis. Neuroimage 50 neurobiolaging.2019.05.0100. (2), 446–455. https://doi.org/10.1016/j.neuroimage.2009.12.121. Potvin, O., Chouinard, I., Dieumegarde, L., Bartha, R., Bellec, P., Collins, D.L., Prados, F., Moccia, M., Johnson, A., Yiannakas, M., Grussu, F., Cardoso, M.J., Descoteaux, M., Hoge, R., Ramirez, J., Scott, C.J.M., Smith, E.E., Strother, S.C., Ciccarelli, O., Ourselin, S., Barkhof, F., Wheeler-Kingshott, C., 2020. Generalised Black, S.E., Duchesne, S., 2019. The Canadian Dementia Imaging Protocol: boundary shift integral for longitudinal assessment of spinal cord atrophy. Neuroimage 209, 116489. https://doi.org/10.1016/j.neuroimage.2019.116489. 11 N. De Stefano et al. NeuroImage: Clinical 34 (2022) 102972 Amann, M., Andělová, M., Pfister, A., Mueller-Lenke, N., Traud, S., Reinhardt, J., Ganzetti, M., Wenderoth, N., Mantini, D., 2016. Intensity Inhomogeneity Correction of Magon, S., Bendfeldt, K., Kappos, L., Radue, E.-W., Stippich, C., Sprenger, T., 2015. Structural MR Images: A Data-Driven Approach to Define Input Algorithm Subcortical brain segmentation of two dimensional T1-weighted data sets with Parameters. Front Neuroinform 10, 10. https://doi.org/10.3389/fninf.2016.00010. FMRIB’s Integrated Registration and Segmentation Tool (FIRST). Neuroimage Clin 7, Andersson, J.L.R., Skare, S., Ashburner, J., 2003. How to correct susceptibility 43–52. https://doi.org/10.1016/j.nicl.2014.11.010. distortions in spin-echo echo-planar images: Application to diffusion tensor imaging. Granziera, C., Wuerfel, J., Barkhof, F., Calabrese, M., De Stefano, N., Enzinger, C., Neuroimage 20 (2), 870–888. https://doi.org/10.1016/S1053-8119(03)00336-7. Evangelou, N., Filippi, M., Geurts, J.J.G., Reich, D.S., Rocca, M.A., Ropele, S., Perrone, D., Aelterman, J., Pižurica, A., Jeurissen, B., Philips, W., Leemans, A., 2015. The Rovira, À., Sati, P., Toosy, A.T., Vrenken, H., Gandini Wheeler-Kingshott, C.A.M., effect of Gibbs ringing artifacts on measures derived from diffusion MRI. Kappos, L., Barkhof, F., de Stefano, N., Sastre-Garriga, J., Ciccarelli, O., Enzinger, C., NeuroImage 120, 441–455. https://doi.org/10.1016/j.neuroimage.2015.06.068. Filippi, M., Gasperini, C., Kappos, L., Palace, J., Vrenken, H., Rovira, À., Rocca, M.A., Gallichan, D., Scholz, J., Bartsch, A., Behrens, T.E., Robson, M.D., Miller, K.L., 2009. Yousry, T., 2021. Quantitative magnetic resonance imaging towards clinical Addressing a systematic vibration artifact in diffusion-weighted MRI. Hum Brain application in multiple sclerosis. Brain 144 (5), 1296–1311. https://doi.org/ Mapp NA–NA. https://doi.org/10.1002/hbm.20856. 10.1093/brain/awab029. Maximov, I.I., Alnæs, D., Westlye, L.T., 2019. Towards an optimised processing pipeline Ropele, S., Filippi, M., Valsasina, P., Korteweg, T., Barkhof, F., Tofts, P.S., Samson, R., for diffusion magnetic resonance imaging data: Effects of artefact corrections on Miller, D.H., Fazekas, F., 2005. Assessment and correction of B1-induced errors in diffusion metrics and their age associations in UK Biobank. Hum Brain Mapp 40 (14), magnetization transfer ratio measurements. Magn Reson Med 53 (1), 134–140. 4146–4162. https://doi.org/10.1002/hbm.24691. https://doi.org/10.1002/mrm.20310. Laganà, M., Rovaris, M., Ceccarelli, A., Venturelli, C., Marini, S., Baselli, G., 2010. DTI Barker, G.J., Schreiber, W.G., Gass, A., Ranjeva, J.P., Campi, A., Waesberghe, J.H.T.M.V., Parameter Optimisation for Acquisition at 1.5T: SNR Analysis and Clinical Franconi, J.-M., Watt, H.C., Tofts, P.S., 2005. A standardised method for measuring Application. Computational Intelligence and Neuroscience 2010, 1–8. magnetisation transfer ratio on MR imagers from different manufacturers–the Soares, J.M., Marques, P., Alves, V., Sousa, N., 2013. A hitchhiker’s guide to diffusion EuroMT sequence. MAGMA 18 (2), 76–80. https://doi.org/10.1007/s10334-004- tensor imaging. Front Neurosci 7, 31. https://doi.org/10.3389/fnins.2013.00031. 0095-z. Tournier, J.-D., Smith, R., Raffelt, D., Tabbara, R., Dhollander, T., Pietsch, M., Romme Christensen, J., Ratzer, R., Bornsen, L., Lyksborg, M., Garde, E., Dyrby, T.B., Christiaens, D., Jeurissen, B., Yeh, C.-H., Connelly, A., 2019. MRtrix3: A fast, flexible Siebner, H.R., Sorensen, P.S., Sellebjerg, F., 2014. Natalizumab in progressive MS: and open software framework for medical image processing and visualisation. Results of an open-label, phase 2A, proof-of-concept trial. Neurology 82 (17), Neuroimage 202, 116137. https://doi.org/10.1016/j.neuroimage.2019.116137. 1499–1507. https://doi.org/10.1212/WNL.0000000000000361. Christidi, F., Karavasilis, E., Samiotis, K., Bisdas, S., Papanikolaou, N., 2016. Fiber Miller, D.H., Fox, R.J., Phillips, J.T., Hutchinson, M., Havrdova, E., Kita, M., Wheeler- tracking: A qualitative and quantitative comparison between four different software Kingshott, C.A.M., Tozer, D.J., MacManus, D.G., Yousry, T.A., Goodsell, M., tools on the reconstruction of major white matter tracts. Eur J Radiol Open 3, Yang, M., Zhang, R., Viglietta, V., Dawson, K.T., 2015. Effects of delayed-release 153–161. https://doi.org/10.1016/j.ejro.2016.06.002. dimethyl fumarate on MRI measures in the phase 3 CONFIRM study. Neurology 84 Tian, Q., Bilgic, B., Fan, Q., Liao, C., Ngamsombat, C., Hu, Y., Witzel, T., Setsompop, K., (11), 1145–1152. https://doi.org/10.1212/WNL.0000000000001360. Polimeni, J.R., Huang, S.Y., 2020. DeepDTI: High-fidelity six-direction diffusion Weiskopf, N., Suckling, J., Williams, G., Correia, M.M., Inkster, B., Tait, R., Ooi, C., tensor imaging using deep learning. Neuroimage 219, 117017. https://doi.org/ Bullmore, E.T., Lutti, A., 2013. Quantitative multi-parameter mapping of R1, PD(*), 10.1016/j.neuroimage.2020.117017. MT, and R2(*) at 3T: A multi-center validation. Front Neurosci 7. https://doi.org/ Li, B.o., de Groot, M., Steketee, R.M.E., Meijboom, R., Smits, M., Vernooij, M.W., 10.3389/fnins.2013.00095. Ikram, M.A., Liu, J., Niessen, W.J., Bron, E.E., 2020. Neuro4Neuro: A neural network Zhang, L., Chen, T., Tian, H., Xue, H., Ren, H., Li, L.i., Fan, Q., Wen, B., Ren, Z., 2019. approach for neural tract segmentation using large-scale population-based diffusion Reproducibility of inhomogeneous magnetization transfer (ihMT): A test-retest, imaging. Neuroimage 218, 116993. https://doi.org/10.1016/j. multi-site study. Magn Reson Imaging 57, 243–249. https://doi.org/10.1016/j. neuroimage.2020.116993. mri.2018.11.010. Marzban, E.N., Eldeib, A.M., Yassine, I.A., Kadah, Y.M., Ginsberg, S.D., 2020. Tofts, P.S., Steens, S.C.A., Cercignani, M., Admiraal-Behloul, F., Hofman, P.A.M., van Alzheimer’s disease diagnosis from diffusion tensor images using convolutional Osch, M.J.P., Teeuwisse, W.M., Tozer, D.J., van Waesberghe, J.H.T.M., Yeung, R., neural networks. PLoS One 15 (3), e0230409. https://doi.org/10.1371/journal. Barker, G.J., van Buchem, M.A., 2006. Sources of variation in multi-centre brain pone.0230409. MTR histogram studies: Body-coil transmission eliminates inter-centre differences. Marzullo, A., Kocevar, G., Stamile, C., Durand-Dubief, F., Terracina, G., Calimeri, F., MAGMA 19 (4), 209–222. https://doi.org/10.1007/s10334-006-0049-8. Sappey-Marinier, D., 2019. Classification of Multiple Sclerosis Clinical Profiles via Sormani, M.P., Iannucci, G., Rocca, M.A., et al., 2000. Reproducibility of magnetization Graph Convolutional Neural Networks. Front Neurosci 13. https://doi.org/10.3389/ transfer ratio histogram-derived measures of the brain in healthy volunteers. AJNR fnins.2019.00594. Am J Neuroradiol 21, 133–136. Oladosu, O., Liu, W.-Q., Pike, B.G., Koch, M., Metz, L.M., Zhang, Y., 2021. Advanced Tortorella, C., Viti, B., Bozzali, M., Sormani, M.P., Rizzo, G., Gilardi, M.F., Comi, G., Analysis of Diffusion Tensor Imaging Along With Machine Learning Provides New Filippi, M., 2000. A magnetization transfer histogram study of normal-appearing Sensitive Measures of Tissue Pathology and Intra-Lesion Activity in Multiple brain tissue in MS. Neurology 54 (1), 186. Sclerosis. Front Neurosci 15. https://doi.org/10.3389/fnins.2021.634063. Fernando, K.T.M., Tozer, D.J., Miszkiel, K.A., Gordon, R.M., Swanton, J.K., Dalton, C.M., Kontopodis, E., Papadaki, E., Trivzakis, E., Maris, T., Simos, P., Papadakis, G., Barker, G.J., Plant, G.T., Thompson, A.J., Miller, D.H., 2005. Magnetization transfer Tsatsakis, A., Spandidos, D., Karantanas, A., Marias, K., 2021. Emerging deep histograms in clinically isolated syndromes suggestive of multiple sclerosis. Brain learning techniques using magnetic resonance imaging data applied in multiple 128 (12), 2911–2925. https://doi.org/10.1093/brain/awh654. sclerosis and clinical isolated syndrome patients (Review). Exp Ther Med 22 (4). De Stefano, N., Battaglini, M., Stromillo, M.L., et al., 2006. Brain damage as detected by https://doi.org/10.3892/etm.2021.10583. magnetization transfer imaging is less pronounced in benign than in early relapsing Castellazzi, G., Debernard, L., Melzer, T.R., Dalrymple-Alford, J.C., D’Angelo, E., multiple sclerosis. Brain 129, 2008–2016. https://doi.org/10.1093/brain/awl152. Miller, D.H., Gandini Wheeler-Kingshott, C.A.M., Mason, D.F., 2018. Functional Cercignani, M., Gandini Wheeler-Kingshott, C., 2019. From micro- to macro-structures in Connectivity Alterations Reveal Complex Mechanisms Based on Clinical and multiple sclerosis: what is the added value of diffusion imaging. NMR in biomedicine Radiological Status in Mild Relapsing Remitting Multiple Sclerosis. Front Neurol 9. 32 (4). https://doi.org/10.1002/nbm.3888. https://doi.org/10.3389/fneur.2018.00690. Palacios, E.M., Martin, A.J., Boss, M.A., Ezekiel, F., Chang, Y.S., Yuh, E.L., Vassar, M.J., Glover, G.H., Mueller, B.A., Turner, J.A., van Erp, T.G.M., Liu, T.T., Greve, D.N., Schnyer, D.M., MacDonald, C.L., Crawford, K.L., Irimia, A., Toga, A.W., Voyvodic, J.T., Rasmussen, J., Brown, G.G., Keator, D.B., Calhoun, V.D., Lee, H.J., Mukherjee, P., 2017. Toward Precision and Reproducibility of Diffusion Tensor Ford, J.M., Mathalon, D.H., Diaz, M., O’Leary, D.S., Gadde, S., Preda, A., Lim, K.O., Imaging: A Multicenter Diffusion Phantom and Traveling Volunteer Study. AJNR Am Wible, C.G., Stern, H.S., Belger, A., McCarthy, G., Ozyurt, B., Potkin, S.G., 2012. J Neuroradiol 38 (3), 537–545. https://doi.org/10.3174/ajnr.A5025. Function biomedical informatics research network recommendations for prospective Samson, R.S., Lévy, S., Schneider, T., Smith, A.K., Smith, S.A., Cohen-Adad, J., Gandini multicenter functional MRI studies. J Magn Reson Imaging 36 (1), 39–54. https:// Wheeler-Kingshott, C.A.M., Leemans, A., 2016. ZOOM or Non-ZOOM? Assessing doi.org/10.1002/jmri.23572. Spinal Cord Diffusion Tensor Imaging Protocols for Multi-Centre Studies. PLoS One Zou, K.H., Greve, D.N., Wang, M., Pieper, S.D., Warfield, S.K., White, N.S., 11 (5), e0155557. https://doi.org/10.1371/journal.pone.0155557. Manandhar, S., Brown, G.G., Vangel, M.G., Kikinis, R., Wells, W.M., 2005. De Santis, S., Jones, D.K., Roebroeck, A., 2016. Including diffusion time dependence in Reproducibility of functional MR imaging: Preliminary results of prospective multi- the extra-axonal space improves in vivo estimates of axonal diameter and density in institutional study performed by Biomedical Informatics Research Network. human white matter. Neuroimage 130, 91–103. https://doi.org/10.1016/j. Radiology 237 (3), 781–789. https://doi.org/10.1148/radiol.2373041630. neuroimage.2016.01.047. Badhwar, AmanPreet, Collin-Verreault, Y., Orban, P., Urchs, S., Chouinard, I., Vogel, J., Grussu, F., Ianuş, A., Tur, C., Prados, F., Schneider, T., Kaden, E., Ourselin, S., Potvin, O., Duchesne, S., Bellec, P., 2020. Multivariate consistency of resting-state Drobnjak, I., Zhang, H., Alexander, D.C., Gandini Wheeler-Kingshott, C.A.M., 2019. fMRI connectivity maps acquired on a single individual over 2.5 years, 13 sites and 3 Relevance of time-dependence for clinically viable diffusion imaging of the spinal vendors. NeuroImage 205, 116210. https://doi.org/10.1016/j. cord. Magn Reson Med 81 (2), 1247–1264. https://doi.org/10.1002/mrm.27463. neuroimage.2019.116210. Assaf, Y., Freidlin, R.Z., Rohde, G.K., Basser, P.J., 2004. New modeling and experimental Glasser, M.F., Sotiropoulos, S.N., Wilson, J.A., Coalson, T.S., Fischl, B., Andersson, J.L., framework to characterize hindered and restricted water diffusion in brain white Xu, J., Jbabdi, S., Webster, M., Polimeni, J.R., Van Essen, D.C., Jenkinson, M., 2013. matter. Magn Reson Med 52 (5), 965–978. https://doi.org/10.1002/mrm.20274. The minimal preprocessing pipelines for the Human Connectome Project. Pagani, E., Hirsch, J.G., Pouwels, P.J.W., Horsfield, M.A., Perego, E., Gass, A., Neuroimage 80, 105–124. https://doi.org/10.1016/j.neuroimage.2013.04.127. Roosendaal, S.D., Barkhof, F., Agosta, F., Rovaris, M., Caputo, D., Giorgio, A., Graedel, N.N., McNab, J.A., Chiew, M., Miller, K.L., 2017. Motion correction for Palace, J., Marino, S., De Stefano, N., Ropele, S., Fazekas, F., Filippi, M., 2010. functional MRI with three-dimensional hybrid radial-Cartesian EPI. Magn Reson Intercenter differences in diffusion tensor MRI acquisition. J Magn Reson Imaging 31 Med 78 (2), 527–540. https://doi.org/10.1002/mrm.26390. (6), 1458–1468. https://doi.org/10.1002/jmri.22186. Pruim, R.H.R., Mennes, M., van Rooij, D., Llera, A., Buitelaar, J.K., Beckmann, C.F., 2015. ICA-AROMA: A robust ICA-based strategy for removing motion artifacts from 12 N. De Stefano et al. NeuroImage: Clinical 34 (2022) 102972 fMRI data. Neuroimage 112, 267–277. https://doi.org/10.1016/j. Willer, C.J., Li, Y., Abecasis, G.R., 2010. METAL: Fast and efficient meta-analysis of neuroimage.2015.02.064. genomewide association scans. Bioinformatics 26 (17), 2190–2191. https://doi.org/ Costumero, V., Bueichekú, E., Adrián-Ventura, J., Ávila, C., 2020. Opening or closing 10.1093/bioinformatics/btq340. eyes at rest modulates the functional connectivity of V1 with default and salience Hibar, D.P., Adams, H.H.H., Jahanshad, N., Chauhan, G., Stein, J.L., Hofer, E., networks. Sci Rep 10, 9137. https://doi.org/10.1038/s41598-020-66100-y. Renteria, M.E., Bis, J.C., Arias-Vasquez, A., Ikram, M.K., Desrivières, S., Vernooij, M. Glover, G.H., Li, T.Q., Ress, D., 2000. Image-based method for retrospective correction of W., Abramovic, L., Alhusaini, S., Amin, N., Andersson, M., Arfanakis, K., Aribisala, B. physiological motion effects in fMRI: RETROICOR. Magn Reson Med 44, 162–167. S., Armstrong, N.J., Athanasiu, L., Axelsson, T., Beecham, A.H., Beiser, A., https://doi.org/10.1002/1522-2594(200007)44:1<162::aid-mrm23>3.0.co;2-e. Bernard, M., Blanton, S.H., Bohlken, M.M., Boks, M.P., Bralten, J., Brickman, A.M., Behzadi, Y., Restom, K., Liau, J., Liu, T.T., 2007. A component based noise correction Carmichael, O., Chakravarty, M.M., Chen, Q., Ching, C.R.K., Chouraki, V., Cuellar- method (CompCor) for BOLD and perfusion based fMRI. NeuroImage 37 (1), 90–101. Partida, G., Crivello, F., Den Braber, A., Doan, N.T., Ehrlich, S., Giddaluru, S., https://doi.org/10.1016/j.neuroimage.2007.04.042. Goldman, A.L., Gottesman, R.F., Grimm, O., Griswold, M.E., Guadalupe, T., Jahanian, H., Holdsworth, S., Christen, T., Wu, H., Zhu, K., Kerr, A.B., Middione, M.J., Gutman, B.A., Hass, J., Haukvik, U.K., Hoehn, D., Holmes, A.J., Hoogman, M., Dougherty, R.F., Moseley, M., Zaharchuk, G., 2019. Advantages of short repetition Janowitz, D., Jia, T., Jørgensen, K.N., Karbalai, N., Kasperaviciute, D., Kim, S., time resting-state functional MRI enabled by simultaneous multi-slice imaging. Klein, M., Kraemer, B., Lee, P.H., Liewald, D.C.M., Lopez, L.M., Luciano, M., J Neurosci Methods 311, 122–132. https://doi.org/10.1016/j. Macare, C., Marquand, A.F., Matarin, M., Mather, K.A., Mattheisen, M., McKay, D.R., jneumeth.2018.09.033. Milaneschi, Y., Muñoz Maniega, S., Nho, K., Nugent, A.C., Nyquist, P., Loohuis, L.M. Yang, H., Long, X.-Y., Yang, Y., Yan, H., Zhu, C.-Z., Zhou, X.-P., Zang, Y.-F., Gong, Q.-Y., O., Oosterlaan, J., Papmeyer, M., Pirpamer, L., Pütz, B., Ramasamy, A., Richards, J. 2007. Amplitude of low frequency fluctuation within visual areas revealed by S., Risacher, S.L., Roiz-Santiañez, R., Rommelse, N., Ropele, S., Rose, E.J., Royle, N. resting-state functional MRI. Neuroimage 36 (1), 144–152. https://doi.org/ A., Rundek, T., Sämann, P.G., Saremi, A., Satizabal, C.L., Schmaal, L., Schork, A.J., 10.1016/j.neuroimage.2007.01.054. Shen, L.i., Shin, J., Shumskaya, E., Smith, A.V., Sprooten, E., Strike, L.T., Teumer, A., Ragot, D.M., Chen, J.J., 2019. Characterizing contrast origins and noise contribution in Tordesillas-Gutierrez, D., Toro, R., Trabzuni, D., Trompet, S., Vaidya, D., Van der spin-echo EPI BOLD at 3 T. Magn Reson Imaging 57, 328–336. https://doi.org/ Grond, J., Van der Lee, S.J., Van der Meer, D., Van Donkelaar, M.M.J., Van Eijk, K.R., 10.1016/j.mri.2018.11.005. Van Erp, T.G.M., Van Rooij, D., Walton, E., Westlye, L.T., Whelan, C.D., Windham, B. Bennett, C.M., Miller, M.B., 2010. How reliable are the results from functional magnetic G., Winkler, A.M., Wittfeld, K., Woldehawariat, G., Wolf, C., Wolfers, T., Yanek, L.R., resonance imaging? Ann N Y Acad Sci 1191, 133–155. https://doi.org/10.1111/ Yang, J., Zijdenbos, A., Zwiers, M.P., Agartz, I., Almasy, L., Ames, D., Amouyel, P., j.1749-6632.2010.05446.x. Andreassen, O.A., Arepalli, S., Assareh, A.A., Barral, S., Bastin, M.E., Becker, D.M., Andrews-Hanna, J.R., Snyder, A.Z., Vincent, J.L., Lustig, C., Head, D., Raichle, M., Becker, J.T., Bennett, D.A., Blangero, J., van Bokhoven, H., Boomsma, D.I., Buckner, R.L., 2007. Disruption of large-scale brain systems in advanced aging. Brodaty, H., Brouwer, R.M., Brunner, H.G., Buckner, R.L., Buitelaar, J.K., Neuron 56 (5), 924–935. https://doi.org/10.1016/j.neuron.2007.10.038. Bulayeva, K.B., Cahn, W., Calhoun, V.D., Cannon, D.M., Cavalleri, G.L., Cheng, C.-Y., Friston, K.J., Frith, C.D., Liddle, P.F., Frackowiak, R.S.J., 1993. Functional connectivity: Cichon, S., Cookson, M.R., Corvin, A., Crespo-Facorro, B., Curran, J.E., Czisch, M., The principal-component analysis of large (PET) data sets. J Cereb Blood Flow Metab Dale, A.M., Davies, G.E., De Craen, A.J.M., De Geus, E.J.C., De Jager, P.L., De 13 (1), 5–14. https://doi.org/10.1038/jcbfm.1993.4. Zubicaray, G.I., Deary, I.J., Debette, S., DeCarli, C., Delanty, N., Depondt, C., Worsley, K.J., Chen, J.-I., Lerch, J., Evans, A.C., 2005. Comparing functional connectivity DeStefano, A., Dillman, A., Djurovic, S., Donohoe, G., Drevets, W.C., Duggirala, R., via thresholding correlations and singular value decomposition. Philos Trans R Soc Dyer, T.D., Enzinger, C., Erk, S., Espeseth, T., Fedko, I.O., Fernández, G., Ferrucci, L., Lond B Biol Sci 360 (1457), 913–920. https://doi.org/10.1098/rstb.2005.1637. Fisher, S.E., Fleischman, D.A., Ford, I., Fornage, M., Foroud, T.M., Fox, P.T., Beckmann, C.F., DeLuca, M., Devlin, J.T., Smith, S.M., 2005. Investigations into resting- Francks, C., Fukunaga, M., Gibbs, J.R., Glahn, D.C., Gollub, R.L., Göring, H.H.H., state connectivity using independent component analysis. Philos Trans R Soc Lond B Green, R.C., Gruber, O., Gudnason, V., Guelfi, S., Håberg, A.K., Hansell, N.K., Biol Sci 360 (1457), 1001–1013. https://doi.org/10.1098/rstb.2005.1634. Hardy, J., Hartman, C.A., Hashimoto, R., Hegenscheid, K., Heinz, A., Le Hellard, S., Botvinik-Nezer, R., Holzmeister, F., Camerer, C.F., Dreber, A., Huber, J., Hernandez, D.G., Heslenfeld, D.J., Ho, B.-C., Hoekstra, P.J., Hoffmann, W., Johannesson, M., Kirchler, M., Iwanir, R., Mumford, J.A., Adcock, R.A., Avesani, P., Hofman, A., Holsboer, F., Homuth, G., Hosten, N., Hottenga, J.-J., Huentelman, M., Baczkowski, B.M., Bajracharya, A., Bakst, L., Ball, S., Barilari, M., Bault, N., Hulshoff Pol, H.E., Ikeda, M., Jack Jr, C.R., Jenkinson, M., Johnson, R., Jönsson, E. Beaton, D., Beitner, J., Benoit, R.G., Berkers, R.M.W.J., Bhanji, J.P., Biswal, B.B., G., Jukema, J.W., Kahn, R.S., Kanai, R., Kloszewska, I., Knopman, D.S., Bobadilla-Suarez, S., Bortolini, T., Bottenhorn, K.L., Bowring, A., Braem, S., Kochunov, P., Kwok, J.B., Lawrie, S.M., Lemaître, H., Liu, X., Longo, D.L., Lopez, O. Brooks, H.R., Brudner, E.G., Calderon, C.B., Camilleri, J.A., Castrellon, J.J., L., Lovestone, S., Martinez, O., Martinot, J.-L., Mattay, V.S., McDonald, C., Cecchetti, L., Cieslik, E.C., Cole, Z.J., Collignon, O., Cox, R.W., Cunningham, W.A., McIntosh, A.M., McMahon, F.J., McMahon, K.L., Mecocci, P., Melle, I., Meyer- Czoschke, S., Dadi, K., Davis, C.P., Luca, A.D., Delgado, M.R., Demetriou, L., Lindenberg, A., Mohnke, S., Montgomery, G.W., Morris, D.W., Mosley, T.H., Dennison, J.B., Di, X., Dickie, E.W., Dobryakova, E., Donnat, C.L., Dukart, J., Mühleisen, T.W., Müller-Myhsok, B., Nalls, M.A., Nauck, M., Nichols, T.E., Duncan, N.W., Durnez, J., Eed, A., Eickhoff, S.B., Erhart, A., Fontanesi, L., Fricke, G. Niessen, W.J., Nöthen, M.M., Nyberg, L., Ohi, K., Olvera, R.L., Ophoff, R.A., M., Fu, S., Galván, A., Gau, R., Genon, S., Glatard, T., Glerean, E., Goeman, J.J., Pandolfo, M., Paus, T., Pausova, Z., Penninx, B.W.J.H., Pike, G.B., Potkin, S.G., Golowin, S.A.E., González-García, C., Gorgolewski, K.J., Grady, C.L., Green, M.A., Psaty, B.M., Reppermund, S., Rietschel, M., Roffman, J.L., Romanczuk-Seiferth, N., Guassi Moreira, J.F., Guest, O., Hakimi, S., Hamilton, J.P., Hancock, R., Rotter, J.I., Ryten, M., Sacco, R.L., Sachdev, P.S., Saykin, A.J., Schmidt, R., Handjaras, G., Harry, B.B., Hawco, C., Herholz, P., Herman, G., Heunis, S., Schmidt, H., Schofield, P.R., Sigursson, S., Simmons, A., Singleton, A., Sisodiya, S.M., Hoffstaedter, F., Hogeveen, J., Holmes, S., Hu, C.-P., Huettel, S.A., Hughes, M.E., Smith, C., Smoller, J.W., Soininen, H., Steen, V.M., Stott, D.J., Sussmann, J.E., Iacovella, V., Iordan, A.D., Isager, P.M., Isik, A.I., Jahn, A., Johnson, M.R., Thalamuthu, A., Toga, A.W., Traynor, B.J., Troncoso, J., Tsolaki, M., Tzourio, C., Johnstone, T., Joseph, M.J.E., Juliano, A.C., Kable, J.W., Kassinopoulos, M., Uitterlinden, A.G., Hernández, M.C.V., Van der Brug, M., van der Lugt, A., van der Koba, C., Kong, X.-Z., Koscik, T.R., Kucukboyaci, N.E., Kuhl, B.A., Kupek, S., Laird, A. Wee, N.J.A., Van Haren, N.E.M., van ’t Ent, D., Van Tol, M.-J., Vardarajan, B.N., R., Lamm, C., Langner, R., Lauharatanahirun, N., Lee, H., Lee, S., Leemans, A., Vellas, B., Veltman, D.J., Völzke, H., Walter, H., Wardlaw, J.M., Wassink, T.H., Leo, A., Lesage, E., Li, F., Li, M.Y.C., Lim, P.C., Lintz, E.N., Liphardt, S.W., Losecaat Weale, M.E., Weinberger, D.R., Weiner, M.W., Wen, W., Westman, E., White, T., Vermeer, A.B., Love, B.C., Mack, M.L., Malpica, N., Marins, T., Maumet, C., Wong, T.Y., Wright, C.B., Zielke, R.H., Zonderman, A.B., Martin, N.G., Van Duijn, C. McDonald, K., McGuire, J.T., Melero, H., Méndez Leal, A.S., Meyer, B., Meyer, K.N., M., Wright, M.J., Longstreth, W.T., Schumann, G., Grabe, H.J., Franke, B., Launer, L. Mihai, G., Mitsis, G.D., Moll, J., Nielson, D.M., Nilsonne, G., Notter, M.P., Olivetti, E., J., Medland, S.E., Seshadri, S., Thompson, P.M., Ikram, M.A., 2017. Novel genetic Onicas, A.I., Papale, P., Patil, K.R., Peelle, J.E., Pérez, A., Pischedda, D., Poline, J.-B., loci associated with hippocampal volume. Nat Commun 8 (1). https://doi.org/ Prystauka, Y., Ray, S., Reuter-Lorenz, P.A., Reynolds, R.C., Ricciardi, E., Rieck, J.R., 10.1038/ncomms13624. Rodriguez-Thompson, A.M., Romyn, A., Salo, T., Samanez-Larkin, G.R., Sanz- Kochunov, P., Jahanshad, N., Sprooten, E., Nichols, T.E., Mandl, R.C., Almasy, L., Morales, E., Schlichting, M.L., Schultz, D.H., Shen, Q., Sheridan, M.A., Silvers, J.A., Booth, T., Brouwer, R.M., Curran, J.E., de Zubicaray, G.I., Dimitrova, R., Skagerlund, K., Smith, A., Smith, D.V., Sokol-Hessner, P., Steinkamp, S.R., Duggirala, R., Fox, P.T., Elliot Hong, L., Landman, B.A., Lemaitre, H., Lopez, L.M., Tashjian, S.M., Thirion, B., Thorp, J.N., Tinghög, G., Tisdall, L., Tompson, S.H., Toro- Martin, N.G., McMahon, K.L., Mitchell, B.D., Olvera, R.L., Peterson, C.P., Starr, J.M., Serey, C., Torre Tresols, J.J., Tozzi, L., Truong, V., Turella, L., van ‘t Veer, A.E., Sussmann, J.E., Toga, A.W., Wardlaw, J.M., Wright, M.J., Wright, S.N., Bastin, M.E., Verguts, T., Vettel, J.M., Vijayarajah, S., Vo, K., Wall, M.B., Weeda, W.D., Weis, S., McIntosh, A.M., Boomsma, D.I., Kahn, R.S., den Braber, A., de Geus, E.J.C., Deary, I. White, D.J., Wisniewski, D., Xifra-Porxas, A., Yearling, E.A., Yoon, S., Yuan, R., J., Hulshoff Pol, H.E., Williamson, D.E., Blangero, J., van ’t Ent, D., Thompson, P.M., Yuen, K.S.L., Zhang, L., Zhang, X.u., Zosky, J.E., Nichols, T.E., Poldrack, R.A., Glahn, D.C., 2014. Multi-site study of additive genetic effects on fractional Schonberg, T., 2020. Variability in the analysis of a single neuroimaging dataset by anisotropy of cerebral white matter: Comparing meta and megaanalytical many teams. Nature 582 (7810), 84–88. https://doi.org/10.1038/s41586-020-2314- approaches for data pooling. Neuroimage 95, 136–150. https://doi.org/10.1016/j. 9. neuroimage.2014.03.033. Wen, D., Wei, Z., Zhou, Y., Li, G., Zhang, X.u., Han, W., 2018. Deep Learning Methods to Eshaghi, A., Young, A.L., Wijeratne, P.A., Prados, F., Arnold, D.L., Narayanan, S., Process fMRI Data and Their Application in the Diagnosis of Cognitive Impairment: A Guttmann, C.R.G., Barkhof, F., Alexander, D.C., Thompson, A.J., Chard, D., Brief Overview and Our Opinion. Front Neuroinform 12. https://doi.org/10.3389/ Ciccarelli, O., 2021. Identifying multiple sclerosis subtypes using unsupervised fninf.2018.00023. machine learning and MRI data. Nature Communications 12 (1). https://doi.org/ Fortin, J.-P., Sweeney, E.M., Muschelli, J., Crainiceanu, C.M., Shinohara, R.T., 2016. 10.1038/s41467-021-22265-2. Removing inter-subject technical variability in magnetic resonance imaging studies. Dojat, M., Kennedy, D.N., Niessen, W., 2017. Editorial: MAPPING: MAnagement and Neuroimage 132, 198–212. https://doi.org/10.1016/j.neuroimage.2016.02.036. Processing of Images for Population ImagiNG. Front ICT 4. https://doi.org/10.3389/ Fortin, J.-P., Cullen, N., Sheline, Y.I., Taylor, W.D., Aselcioglu, I., Cook, P.A., Adams, P., fict.2017.00018. Cooper, C., Fava, M., McGrath, P.J., McInnis, M., Phillips, M.L., Trivedi, M.H., Marcus, D.S., Olsen, T.R., Ramaratnam, M., Buckner, R.L., 2007. The Extensible Weissman, M.M., Shinohara, R.T., 2018. Harmonization of cortical thickness Neuroimaging Archive Toolkit: An informatics platform for managing, exploring, measurements across scanners and sites. NeuroImage 167, 104–120. https://doi. and sharing neuroimaging data. Neuroinformatics 5 (1), 11–33. org/10.1016/j.neuroimage.2017.11.024. 13 N. De Stefano et al. NeuroImage: Clinical 34 (2022) 102972 Gorgolewski, K.J., Auer, T., Calhoun, V.D., Craddock, R.C., Das, S., Duff, E.P., Progression in Amnestic Mild Cognitive Impairment Patients with Prodromal Flandin, G., Ghosh, S.S., Glatard, T., Halchenko, Y.O., Handwerker, D.A., Hanke, M., Alzheimer’s Disease: Structural Brain Biomarkers. J Alzheimers Dis 69 (1), 3–14. Keator, D., Li, X., Michael, Z., Maumet, C., Nichols, B.N., Nichols, T.E., Pellman, J., https://doi.org/10.3233/JAD-180152. Poline, J.-B., Rokem, A., Schaefer, G., Sochat, V., Triplett, W., Turner, J.A., Ritchie, C.W., Muniz-Terrera, G., Kivipelto, M., Solomon, A., Tom, B., Molinuevo, J.L., Varoquaux, G., Poldrack, R.A., 2016. The brain imaging data structure, a format for 2020. The European Prevention of Alzheimer’s Dementia (EPAD) Longitudinal organizing and describing outputs of neuroimaging experiments. Sci Data 3 (1). Cohort Study: Baseline Data Release V500.0. J Prev Alzheimers Dis 1–7. https://doi. https://doi.org/10.1038/sdata.2016.44. org/10.14283/jpad.2019.46. Weiner, M.W., Veitch, D.P., Aisen, P.S., Beckett, L.A., Cairns, N.J., Green, R.C., ten Kate, M., Ingala, S., Schwarz, A.J., Fox, N.C., Chételat, G., van Berckel, B.N.M., Harvey, D., Jack, C.R., Jagust, W., Morris, J.C., Petersen, R.C., Salazar, J., Saykin, A. Ewers, M., Foley, C., Gispert, J.D., Hill, D., Irizarry, M.C., Lammertsma, A.A., J., Shaw, L.M., Toga, A.W., Trojanowski, J.Q., 2017. The Alzheimer’s Disease Molinuevo, J.L., Ritchie, C., Scheltens, P., Schmidt, M.E., Visser, P.J., Waldman, A., Neuroimaging Initiative 3: Continued innovation for clinical trial improvement. Wardlaw, J., Haller, S., Barkhof, F., 2018. Secondary prevention of Alzheimer’s Alzheimers Dement 13 (5), 561–571. https://doi.org/10.1016/j.jalz.2016.10.006. dementia: Neuroimaging contributions. Alzheimers Res Ther 10 (1). https://doi.org/ Dimitriadis, S.I., Liparas, D., Alzheimer’s Disease Neuroimaging Initiative, 2018. How 10.1186/s13195-018-0438-z. random is the random forest? Random forest algorithm on the service of structural Alfaro-Almagro, F., Jenkinson, M., Bangerter, N.K., Andersson, J.L.R., Griffanti, L., imaging biomarkers for Alzheimer’s disease: from Alzheimer’s disease neuroimaging Douaud, G., Sotiropoulos, S.N., Jbabdi, S., Hernandez-Fernandez, M., Vallee, E., initiative (ADNI) database. Neural Regen Res 13, 962–970. https://doi.org/ Vidaurre, D., Webster, M., McCarthy, P., Rorden, C., Daducci, A., Alexander, D.C., 10.4103/1673-5374.233433. Zhang, H., Dragonu, I., Matthews, P.M., Miller, K.L., Smith, S.M., 2018. Image Ramanan, V.K., Lesnick, T.G., Przybelski, S.A., Heckman, M.G., Knopman, D.S., Graff- processing and Quality Control for the first 10,000 brain imaging datasets from UK Radford, J., Lowe, V.J., Machulda, M.M., Mielke, M.M., Jack, C.R., Petersen, R.C., Biobank. NeuroImage 166, 400–424. https://doi.org/10.1016/j. Ross, O.A., Vemuri, P., 2021. Coping with brain amyloid: Genetic heterogeneity and neuroimage.2017.10.034. cognitive resilience to Alzheimer’s pathophysiology. Acta Neuropathol Commun 9 Mowry, E.M., Bermel, R.A., Williams, J.R., Benzinger, T.L.S., de Moor, C., Fisher, E., (1). https://doi.org/10.1186/s40478-021-01154-1. Hersh, C.M., Hyland, M.H., Izbudak, I., Jones, S.E., Kieseier, B.C., Kitzler, H.H., Jack, C.R., Barnes, J., Bernstein, M.A., Borowski, B.J., Brewer, J., Clegg, S., Dale, A.M., Krupp, L., Lui, Y.W., Montalban, X., Naismith, R.T., Nicholas, J.A., Pellegrini, F., Carmichael, O., Ching, C., DeCarli, C., Desikan, R.S., Fennema-Notestine, C., Fjell, A. Rovira, A., Schulze, M., Tackenberg, B., Tintore, M., Tivarus, M.E., Ziemssen, T., M., Fletcher, E., Fox, N.C., Gunter, J., Gutman, B.A., Holland, D., Hua, X., Insel, P., Rudick, R.A., 2020. Harnessing Real-World Data to Inform Decision-Making: Kantarci, K., Killiany, R.J., Krueger, G., Leung, K.K., Mackin, S., Maillard, P., Multiple Sclerosis Partners Advancing Technology and Health Solutions (MS Malone, I.B., Mattsson, N., McEvoy, L., Modat, M., Mueller, S., Nosheny, R., PATHS). Front Neurol 11. https://doi.org/10.3389/fneur.2020.00632. Ourselin, S., Schuff, N., Senjem, M.L., Simonson, A., Thompson, P.M., Rettmann, D., Filippi, M., Tedeschi, G., Pantano, P., De Stefano, N., Zaratin, P., Rocca, M.A., 2017. The Vemuri, P., Walhovd, K., Zhao, Y., Zuk, S., Weiner, M., 2015. Magnetic resonance Italian Neuroimaging Network Initiative (INNI): Enabling the use of advanced MRI imaging in Alzheimer’s Disease Neuroimaging Initiative 2. Alzheimers Dement 11 techniques in patients with MS. Neurol Sci 38 (6), 1029–1038. https://doi.org/ (7), 740–756. https://doi.org/10.1016/j.jalz.2015.05.002. 10.1007/s10072-017-2903-z. Jovicich, J., Minati, L., Marizzoni, M., Marchitelli, R., Sala-Llonch, R., Bartrés-Faz, D., Storelli, L., Rocca, M.A., Pantano, P., Pagani, E., De Stefano, N., Tedeschi, G., Zaratin, P., Arnold, J., Benninghoff, J., Fiedler, U., Roccatagliata, L., Picco, A., Nobili, F., Filippi, M., 2019. MRI quality control for the Italian Neuroimaging Network Blin, O., Bombois, S., Lopes, R., Bordet, R., Sein, J., Ranjeva, J.-P., Didic, M., Gros- Initiative: Moving towards big data in multiple sclerosis. J Neurol 266 (11), Dagnac, H., Payoux, P., Zoccatelli, G., Alessandrini, F., Beltramello, A., Bargalló, N., 2848–2858. https://doi.org/10.1007/s00415-019-09509-4. Ferretti, A., Caulo, M., Aiello, M., Cavaliere, C., Soricelli, A., Parnetti, L., Hsieh, J.J.L., Svalbe, I., 2020. Magnetic resonance fingerprinting: From evolution to Tarducci, R., Floridi, P., Tsolaki, M., Constantinidis, M., Drevelegas, A., Rossini, P. clinical applications. J Med Radiat Sci 67 (4), 333–344. https://doi.org/10.1002/ M., Marra, C., Schönknecht, P., Hensch, T., Hoffmann, K.-T., Kuijer, J.P., Visser, P.J., jmrs.413. Barkhof, F., Frisoni, G.B., 2016. Longitudinal reproducibility of default-mode Gonçalves, F.G., Serai, S.D., Zuccoli, G., 2018. Synthetic Brain MRI: Review of Current network connectivity in healthy elderly participants: A multicentric resting-state Concepts and Future Directions. Top Magn Reson Imaging 27, 387–393. https://doi. fMRI study. Neuroimage 124, 442–454. https://doi.org/10.1016/j. org/10.1097/RMR.0000000000000189. neuroimage.2015.07.010. Iglesias, J.E., Billot, B., Balbastre, Y., Tabari, A., Conklin, J., Gilberto González, R., Marizzoni, M., Ferrari, C., Jovicich, J., Albani, D., Babiloni, C., Cavaliere, L., Didic, M., Alexander, D.C., Golland, P., Edlow, B.L., Fischl, B., 2021. Joint super-resolution and Forloni, G., Galluzzi, S., Hoffmann, K.-T., Molinuevo, J.L., Nobili, F., Parnetti, L., synthesis of 1 mm isotropic MP-RAGE volumes from clinical MRI exams with scans Payoux, P., Ribaldi, F., Rossini, P.M., Schönknecht, P., Salvatore, M., Soricelli, A., of different orientation, resolution and contrast. Neuroimage 237, 118206. https:// Hensch, T., Tsolaki, M., Visser, P.J., Wiltfang, J., Richardson, J.C., Bordet, R., doi.org/10.1016/j.neuroimage.2021.118206. Blin, O., Frisoni, G.B., Perry, G., 2019. Predicting and Tracking Short Term Disease 14