ORIGINAL RESEARCH published: 12 July 2021 doi: 10.3389/fnins.2021.691244 Edited by: Danny J. J. Wang, MRI Radiomic Signature of White University of Southern California, Los Angeles, United States Matter Hyperintensities Is Reviewed by: Dafna Ben Bashat, The Wohl institute for Advanced Associated With Clinical Phenotypes Imaging, Israel Martin Bretzner 1,2* , Anna K. Bonkhoff 1 , Markus D. Schirmer 1 , Sungmin Hong 1 , Yong Fan, Adrian V. Dalca 1,3,4 , Kathleen L. Donahue 1 , Anne-Katrin Giese 1 , Mark R. Etherton 1 , University of Pennsylvania, Pamela M. Rist 1,5 , Marco Nardin 1 , Razvan Marinescu 1,4 , Clinton Wang 1,4 , United States Robert W. Regenhardt 1 , Xavier Leclerc 2 , Renaud Lopes 2,6 , Oscar R. Benavente 7 , *Correspondence: John W. Cole 8 , Amanda Donatti 9 , Christoph J. Griessenauer 10,11 , Laura Heitsch 12,13 , Martin Bretzner Lukas Holmegaard 14 , Katarina Jood 14 , Jordi Jimenez-Conde 15 , Steven J. Kittner 8 , mbretzner@mgh.harvard.edu; Robin Lemmens 16,17 , Christopher R. Levi 18,19 , Patrick F. McArdle 20 , martin.bretzner@gmail.com † The Caitrin W. McDonough 21 , James F. Meschia 22 , Chia-Ling Phuah 13 , Arndt Rolfs 23 , members of the MRI-GENIE Stefan Ropele 24 , Jonathan Rosand 25 , Jaume Roquer 26 , Tatjana Rundek 26 , GISCOME Investigators and the Ralph L. Sacco 26 , Reinhold Schmidt 24 , Pankaj Sharma 27,28 , Agnieszka Slowik 29 , International and Stroke Genetics Consortium are given in Alessandro Sousa 8 , Tara M. Stanne 14 , Daniel Strbian 30 , Turgut Tatlisumak 31,32 , Supplementary Table 1. Vincent Thijs 33 , Achala Vagal 34 , Johan Wasselius 35,36 , Daniel Woo 37 , Ona Wu 3 , Ramin Zand 38 , Bradford B. Worrall 39 , Jane M. Maguire 40 , Arne Lindgren 41,42 , Specialty section: Christina Jern 14 , Polina Golland 4 , Grégory Kuchcinski 2 , This article was submitted to Natalia S. Rost1 and on behalf of the MRI-GENIE GISCOME Investigators the International Brain Imaging Methods, and Stroke Genetics Consortium † a section of the journal 1 Frontiers in Neuroscience J. Philip Kistler Stroke Research Center, Massachusetts General Hospital, Boston, MA, United States, 2 Inserm, CHU Lille, U1172 - LilNCog (JPARC) - Lille Neurosciences and Cognition, University of Lille, Lille, France, 3 A. A. Martinos Center for Received: 05 April 2021 Biomedical Imaging, Massachusetts General Hospital, Harvard Medical School, Boston, MA, United States, 4 Computer Accepted: 15 June 2021 Science and Artificial Intelligence Laboratory, Massachusetts Institute of Technology, Cambridge, MA, United States, Published: 12 July 2021 5 Division of Preventive Medicine, Department of Medicine, Brigham and Women’s Hospital and Harvard Medical School, Citation: Boston, MA, United States, 6 CNRS, Institut Pasteur de Lille, US 41 - UMS 2014 - PLBS, Lille, France, 7 Department of Bretzner M, Bonkhoff AK, Medicine, Division of Neurology, University of British Columbia, Vancouver, BC, Canada, 8 Department of Neurology, Schirmer MD, Hong S, Dalca AV, University of Maryland School of Medicine and Veterans Affairs Maryland Health Care System, Baltimore, MD, United States, 9 Donahue KL, Giese A-K, School of Medical Sciences, University of Campinas (UNICAMP) and the Brazilian Institute of Neuroscience and Etherton MR, Rist PM, Nardin M, Neurotechnology (BRAINN), Campinas, Brazil, 10 Department of Neurosurgery, Geisinger, Danville, PA, United States, 11 Marinescu R, Wang C, Research Institute of Neurointervention, Paracelsus Medical University, Salzburg, Austria, 12 Division of Emergency Regenhardt RW, Leclerc X, Lopes R, Medicine, Washington University School of Medicine, St. Louis, MO, United States, 13 Department of Neurology, Washington Benavente OR, Cole JW, Donatti A, University School of Medicine and Barnes-Jewish Hospital, St. Louis, MO, United States, 14 Institute of Biomedicine, Griessenauer CJ, Heitsch L, Sahlgrenska Academy at University of Gothenburg, Gothenburg, Sweden, 15 Department of Neurology, Neurovascular Holmegaard L, Jood K, Research Group (NEUVAS), Institut Hospital del Mar d’Investigacions Mèdiques (IMIM), Universitat Autonoma de Barcelona, Jimenez-Conde J, Kittner SJ, Barcelona, Spain, 16 Department of Neurosciences, Experimental Neurology and Leuven Research Institute for Neuroscience Lemmens R, Levi CR, McArdle PF, and Disease (LIND), KU Leuven – University of Leuven, Leuven, Belgium, 17 VIB, Vesalius Research Center, Laboratory of McDonough CW, Meschia JF, Neurobiology, Department of Neurology, University Hospitals Leuven, Leuven, Belgium, 18 School of Medicine and Public Phuah C-L, Rolfs A, Ropele S, Health, University of Newcastle, Newcastle, NSW, Australia, 19 Department of Neurology, John Hunter Hospital, Newcastle, Rosand J, Roquer J, Rundek T, NSW, Australia, 20 Division of Endocrinology, Diabetes and Nutrition, Department of Medicine, University of Maryland School Sacco RL, Schmidt R, Sharma P, of Medicine, Baltimore, MD, United States, 21 Department of Pharmacotherapy and Translational Research and Center for Slowik A, Sousa A, Stanne TM, Pharmacogenomics, University of Florida, Gainesville, FL, United States, 22 Department of Neurology, Mayo Clinic, Strbian D, Tatlisumak T, Thijs V, Jacksonville, FL, United States, 23 Centogene AG, Rostock, Germany, 24 Department of Neurology, Clinical Division of Vagal A, Wasselius J, Woo D, Wu O, Neurogeriatrics, Medical University of Graz, Graz, Austria, 25 Henry and Allison McCance Center for Brain Health, Center for Zand R, Worrall BB, Maguire JM, Genomic Medicine, Massachusetts General Hospital, Boston, MA, United States, 26 Department of Neurology and Evelyn F. Lindgren A, Jern C, Golland P, McKnight Brain Institute, Miller School of Medicine, University of Miami, Miami, FL, United States, 27 Institute of Kuchcinski G and Rost NS (2021) MRI Cardiovascular Research, Royal Holloway University of London (ICR2UL), Egham, United Kingdom, 28 Ashford and St. Radiomic Signature of White Matter Peter’s Hospitals, Chertsey and Ashford, United Kingdom, 29 Department of Neurology, Jagiellonian University Medical Hyperintensities Is Associated With College, Krakow, Poland, 30 Division of Neurocritical Care and Emergency Neurology, Department of Neurology, Helsinki Clinical Phenotypes. University Central Hospital, Helsinki, Finland, 31 Department of Clinica Neuroscience, Institute of Neuroscience and Front. Neurosci. 15:691244. Physiology, Sahlgrenska Academy at University of Gothenburg, Gothenburg, Sweden, 32 Department of Neurology, doi: 10.3389/fnins.2021.691244 Sahlgrenska University Hospital, Gothenburg, Sweden, 33 Stroke Division, Florey Institute of Neuroscience and Mental Frontiers in Neuroscience | www.frontiersin.org 1 July 2021 | Volume 15 | Article 691244 Bretzner et al. Radiomics of White Matter Hyperintensities Health, Department of Neurology Austin Health, Heidelberg, VIC, Australia, 34 Department of Radiology, University of Cincinnati College of Medicine, Cincinnati, OH, United States, 35 Department of Clinical Sciences Lund, Radiology, Lund University, Lund, Sweden, 36 Department of Radiology, Neuroradiology, Skåne University Hospital, Malmö, Sweden, 37 Department of Neurology and Rehabilitation Medicine, University of Cincinnati College of Medicine, Cincinnati, OH, United States, 38 Department of Neurology, Geisinger, Danville, PA, United States, 39 Department of Neurology and Public Health Sciences, University of Virginia, Charlottesville, VA, United States, 40 Faculty of Health, University of Technology Sydney, Ultimo, NSW, Australia, 41 Department of Neurology and Rehabilitation Medicine, Skåne University Hospital, Lund, Sweden, 42 Department of Clinical Sciences Lund, Neurology, Lund University, Lund, Sweden Objective: Neuroimaging measurements of brain structural integrity are thought to be surrogates for brain health, but precise assessments require dedicated advanced image acquisitions. By means of quantitatively describing conventional images, radiomic analyses hold potential for evaluating brain health. We sought to: (1) evaluate radiomics to assess brain structural integrity by predicting white matter hyperintensities burdens (WMH) and (2) uncover associations between predictive radiomic features and clinical phenotypes. Methods: We analyzed a multi-site cohort of 4,163 acute ischemic strokes (AIS) patients with T2-FLAIR MR images with total brain and WMH segmentations. Radiomic features were extracted from normal-appearing brain tissue (brain mask–WMH mask). Radiomics-based prediction of personalized WMH burden was done using ElasticNet linear regression. We built a radiomic signature of WMH with stable selected features predictive of WMH burden and then related this signature to clinical variables using canonical correlation analysis (CCA). Results: Radiomic features were predictive of WMH burden (R2 = 0.855 ± 0.011). Seven pairs of canonical variates (CV) significantly correlated the radiomics signature of WMH and clinical traits with respective canonical correlations of 0.81, 0.65, 0.42, 0.24, 0.20, 0.15, and 0.15 (FDR-corrected p-valuesCV 1−6 < 0.001, p-valueCV 7 = 0.012). The clinical CV1 was mainly influenced by age, CV2 by sex, CV3 by history of smoking and diabetes, CV4 by hypertension, CV5 by atrial fibrillation (AF) and diabetes, CV6 by coronary artery disease (CAD), and CV7 by CAD and diabetes. Conclusion: Radiomics extracted from T2-FLAIR images of AIS patients capture microstructural damage of the cerebral parenchyma and correlate with clinical phenotypes, suggesting different radiographical textural abnormalities per cardiovascular risk profile. Further research could evaluate radiomics to predict the progression of WMH and for the follow-up of stroke patients’ brain health. Keywords: stroke, cerebrovascular disease (CVD), MRI, radiomics, machine learning, brain health INTRODUCTION neurodegeneration beyond normal brain aging (Brugulat-Serrat et al., 2019). White matter hyperintensities (WMH) are a cardinal Structural injury of the brain has been shown to occur at manifestation of small vessel disease (SVD) (Wardlaw et al., the macrostructural level, in the form of WMH, but also at 2013). Increased WMH burden is associated with incident the microstructural level. Advanced diffusion tensor imaging ischemic stroke and worse clinical outcome (Arsava et al., (DTI) studies have shown an age-related loss of parenchymal 2009). Beyond ischemic stroke, WMH are also associated microstructural integrity in normal-appearing white matter with vascular cognitive impairment and dementia (Au (NAWM) (Etherton et al., 2017a). Furthermore, perfusion- et al., 2006). WMH prevalence increases with age but is weighted imaging (PWI)-based research has also revealed age- also directly influenced by individual small vessel risk factors: related alterations of the blood-brain barrier with increased the aggregation of cardiovascular risk factors leads to an contrast agents’ leakage (Topakian et al., 2010). However, such increased WMH burden (Wardlaw et al., 2015). Hence, microstructural injuries are not visualized with conventional WMH are an imaging biomarker of brain health suggestive of structural MRI sequences, and as DTI and PWI require special Frontiers in Neuroscience | www.frontiersin.org 2 July 2021 | Volume 15 | Article 691244 Bretzner et al. Radiomics of White Matter Hyperintensities acquisition times, the outlined imaging biomarkers are not Data Collection and Neuroimaging currently used in clinical routine for SVD patients. Consequently, Pre-processing we are in need of conventional MRI-based methodologies that Clinical data were acquired within the SiGN study and better quantify SVD and brain health to ensure a widespread comprised information on age, sex, hypertension (HTN), history application and translation to clinical practice. of smoking, diabetes mellitus (DM), atrial fibrillation (AF). Radiomic analyses cover a broad ensemble of high- Among the 6,627 patients included across 20 sites, FLAIR throughput quantification methods applicable to digitalized images were available for 6,389 patients. Axial T2-FLAIR medical images (Gillies et al., 2015). These methods images were acquired between 2003 and 2011 within 48 h automatically extract high-dimensional data, called of the initial stroke. They had a mean in-plane resolution of radiomic features, by describing a given region of 0.7 mm (range: 0.3–1.0 mm) and a through-plane resolution interest by its size, shape, histogram, and relationship of 6.2 mm (range: 3.0–30.0 mm). Total brain, ventricle, and between voxels. Because these techniques can capture WMH segmentations were accomplished using deep learning slight differences in intensities and patterns that would methods described in detail previously (Schirmer et al., 2019; remain undetected to a human reader, radiomics bear the Dubost et al., 2020). Briefly, total brain segmentation was potential to describe neuroimaging beyond what meets done using a tailored 2D-convolutional neural network for the naked eye, and thus might help to phenotype SVD clinical T2-FLAIR data. T2-FLAIR image intensities were (Lambin et al., 2012). Conceivably, they may identify early normalized and scaled. Successively, WMH and ventricles were underlying brain injury at the individual level with rapid automatically segmented using distinct convolutional neural clinical translatability and thus enhance personalized care network frameworks. A total of 1,353 patients were excluded in stroke and SVD. after final quality control of all T2-FLAIR images and respective The aim of the current study was to assess the structural segmentations; this control process is described in great detail integrity of the brain using a texture analysis approach in a previous publication (Schirmer et al., 2019). To capture and to understand the infra-radiological footprint of the underlying processes of SVD in brain parenchyma not WMH by exploring its relationship with cardiovascular overtly affected by WMH, we computed masks for normal- risk profiles. To do so, we analyzed 4,163 T2 FLAIR appearing brain parenchyma by subtracting ventricles and WMH images from a large multi-site international collaborative masks from total brain masks, resulting in 5,031 masks. To effort studying stroke and WMH. We sought to (a) remove any stray voxels that could impact radiomics extraction, build a robust radiomic signature of the subvisible a morphological opening operation with a 3 × 3 voxel manifestations of WMH and (b) to apply canonical kernel was performed on the final masks. Among those 5,031 correlation analysis (CCA) to investigate the relation patients, 868 were excluded for missing major clinical data between this latent textural expression in relation to (age and sex), remaining missing values (89 patients included, sociodemographic information and cardiovascular risk factors, see Supplementary Table 2) were imputed using medians of providing a potentially novel approach to improve SVD distributions. As a result, a total of 4,163 patients were included and stroke care. across 17 different sites. Radiomic Feature Extraction MATERIALS AND METHODS Radiomic features were extracted using the open-source toolbox PyRadiomics V2.2.0. The full list of the radiomics Participants extraction parameters can be found online at https://github.com/ We reviewed all ischemic stroke patients included in MBretzner/WMH_radiomicSign. the MRI-GENetics Interface Exploration (MRI-GENIE) Briefly, to account for the discrepancy in voxel sizes and study, a large international multi-site collaboration of to reduce unwanted variance that could be originating from 20 sites gathering clinical, MRI imaging, and genetic differences between centers and scanners, all features were data, built on top of the NINDS Stroke Genetics extracted in-plane from down-sampled 1 mm × 1 mm × 6 mm Network (SiGN) study. Both study design, data collection T2-FLAIR images. Quantization was set to a fixed bin width protocols, and populations have been previously described of 5. Features extraction was performed outside of WMH on (Giese et al., 2020). native and pre-filtered images. Filters included Laplacian of Gaussian (LoG) filters (with sigmas of 1, 2, and 3 mm), wavelet decompositions, and 2D Local Binary Patterns (2D-LBD). Ethics For each patient, 118 features were computed including mask The MRI-GENIE project has been approved by the statistics, shape features, first-order histogram statistics, GLCM MGH Institutional Review Board (IRB, Protocol #: (Gray Level Co-occurrence Matrix) features, GLRLM (Gray 2001P001186 and Protocol #: 2003P000836), as well Level Run Length Matrix), GLDM (Gray Level Dependence as ethics boards of the collaborating institutions. All Matrix), and NGTDM (Neighboring Gray Tone Difference participants or health care proxy provided signed Matrix) features. Exhaustive and didactic descriptions and informed consent. formulas of every radiomic feature and filter can be found online Frontiers in Neuroscience | www.frontiersin.org 3 July 2021 | Volume 15 | Article 691244 Bretzner et al. Radiomics of White Matter Hyperintensities at https://pyradiomics.readthedocs.io/en/latest/features.html. The shrinkage ability of the ElasticNet regressor As a result, we extracted 763 rotation invariant radiomic was leveraged to select the most predictive features features per patient. of the WMH burden. The radiomic signature of the WMH was built with the features that were consistently selected across each of the 30 repetitions Machine Learning Approach to Build the and therefore represented the most robust and stable Radiomic Signature of the WMH predictors of WMH burden. To account for cerebral size differences, each WMH volume was divided by the corresponding brain volume to obtain a Understanding the Textural Footprint of percentage of WMH per total brain volume. As the resultant distribution was highly skewed, it was transformed using a Box- Clinical Phenotypes Association of clinical variables and the radiomic signature Cox transform and is referred to as “WMH burden” in the of WMH burden were done via CCA, which allows studying next paragraphs. two multivariate variable sets concomitantly (Hotelling, 1936; To address the high dimensionality of the data, prediction Wang et al., 2020). Indeed, traditional analyses explore of the WMH burden was done using an ElasticNet relationships between many to one variable, whereas CCA linear regressor. Since ElasticNet coefficient estimates can study complex many-to-many correlations, truly leveraging are not scale-invariant, we standardized predictors, i.e., the power of multivariate datasets. CCA can be conceived radiomics variables, to be 0 centered and have variances as similar to principal component analyses in the way that of the same order. each side of the data (here clinical and radiomics) undergoes Radiomics-based predictions of WMH burden were a factorization into a latent representation of the variables, performed in 30-times repeated nested fivefold stratified cross- called canonical variates (CV). The canonical correlation score validation scheme, resulting in a total of 24,990 out-of-sample of a canonical function represents the correlation between predictions. Predictions were plotted against ground truth values, the two CV that composes it. To extract each canonical and R2 values were computed with standard deviation. function, CCA finds combinations of factors of the two sets To better understand the role of each class of radiomics so that they are maximally correlated. Canonical loadings and to rule out an association based solely on the size of represent the correlations between variables and their latent the extraction mask and thus reflecting only atrophy, an representation (CV) and can be interpreted as the relative ancillary prediction of the WMH burden was performed contribution of variables to the variates: a variable with a large using only the radiomics features that only reflected the loading has more impact on a variate than a variable with a size and the shape of the analyzed brains. As NIHSS has smaller loading. been shown to be a surrogate marker of stroke volume, Radiomic features and continuous clinical variable (age) and to assess a hypothetical impact of stroke lesions normality was assessed using the Shapiro–Wilks test and, if on radiomics features, residual of the WMH burden needed, were transformed using the R toolbox BestNormalize predictions were studied per NIHSS score when available (Peterson and Cavanaugh, 2020). Significance of canonical (Tong et al., 1998). To explore the potential impact of data correlations was determined via permutation testing (1,000 heterogeneity across all sites, a second ancillary analysis was permutations) and assessed using Wilks’ Lambda computed performed predicting WMH burden in a leave-one-site-out with Rao’s F-approximation, p-values were corrected for cross-validation scheme. multiple testing with Benjamini–Hochberg procedure (Hotelling, 1936; Wang et al., 2020). Explained variances of the canonical functions were calculated and figured TABLE 1 | Demographic and clinical characteristics of the study population in a scree plot. Loadings were calculated to discover and (n = 4,163). characterize the impact of clinical and radiomic features on Age Mean (SD) 62.8 (15.0) each canonical function and thus to provide support for the interpretation of the relationship between the radiomics and Female n (%) 1,748 (42.0%) clinical domains. Hypertension n (%) 2,825 (67.9%) Overall, the goal of CCA is to find underlying representations Diabetes mellitus n (%) 687 (16.5%) that best describe the correlations between the two multi- Atrial fibrillation n (%) 595 (14.3%) dimensional datasets. Thus, this technique permits the Coronary artery disease n (%) 772 (18.5%) estimation of the sources of maximal covariance between History of smoking n (%) 1,331 (32.0%) the clinical and the radiomics domains, highlighting the Prior stroke n (%) 539 (12.9%) subvisible contribution of cardiovascular risk factors to WMH volume Median (IQR) 4.2 mL (1.4–11.2) T2 FLAIR imaging. NIHSS* Median (IQR) 3 (1–6) NIHSS, NIH Stroke Scale. *NIHSS was available for 2,234 (53.7%) patients included. Code Availability “Prior stroke” refers to a stroke preceding the one that led to the Radiomic features extraction, feature selection, and inclusion in the study. machine learning analyses were performed in python Frontiers in Neuroscience | www.frontiersin.org 4 July 2021 | Volume 15 | Article 691244 Bretzner et al. Radiomics of White Matter Hyperintensities 3.7.6 using the toolbox scikit-learn (Pedregosa et al., 2011). CCA was performed in R V1.3.1056 using the toolboxes CCA, vegan (González et al., 2008; Oksanen et al., 2019). The complete codes used to perform the radiomics extraction as well as the extraction parameters and the data analysis are available here: https://github.com/MBretzner/WMH_ radiomicSign. RESULTS Population All patients included in MRI-GENIE have suffered an ischemic stroke. Population demographics are shown in Table 1. The mean age was 62.8, and there were 42% females, median WMH volume was 4.2 mL [interquartile range (IQR): 1.4–11.2]. Admission NIH stroke scale (NIHSS) was available for 2,234 (53.7%) patients; median NIHSS was 3 (IQR: 1–6). Comparison of included and excluded patients’ available clinical characteristics is shown in Supplementary Table 2. Briefly, excluded patients were younger, had less hypertension, less coronary artery disease (CAD), less AF, but had more prior strokes. There was no difference in sex or diabetes status. Importantly NIHSS scores did not differ. Building the Latent Radiomic Signature of the WMH Burden The coefficient of determination of the repeated out-of- sample cross-validated predictions of the WMH burden was R2 = 0.855 ± 0.011 (Figure 1). The average (SD) FIGURE 1 | Repeated out-of-sample cross-validated predictions of WMH number of selected features per repetition was 150.3 (5.6). burden. (A) Predictions of the WMH burden resulted in a coefficient of These features represented the most relevant ones in the determination of R2 = 0.855 ± 0.011. Predicted and true WMH burdens show prediction of WMH burden. To reduce the redundancy and negative values due to the Box-Cox transformation of the WMH burden distribution. The bottom panels provide an illustrative example of a radiomics multicollinearity of radiomic features, we built a signature of extraction mask (B) and a WMH mask (C). the WMH burden by only including the features that were systematically selected in every repetition. This step resulted in the automatic selection of 68 features, which are referred to as the “radiomic signature of WMH.” These features are listed in Clinical Phenotypes Captured by Supplementary Table 1. Radiomics Prediction performance of the WMH burden using radiomics Aiming to discover possible links between clinical phenotypes that only capture the shape and size of the extraction mask and textural features of the radiomic signature of WMH burden, but not voxel intensities was substantially lower with an R2 we performed a CCA. of 0.41 ± 0.03. The analysis of the residual of the WMH The CCA could identify seven canonical functions (CF 1– burden predictions per available NIHSS score showed no trend 7) correlating the radiomics with clinical variates. All seven suggesting an impact of stroke lesions on the predictions canonical functions were significant (False discovery rate (Supplementary Figure 1). corrected p-values CF1−6 < 10−3 ; CF7 = 0.012) with respective Results of the predictions recursively holding one site canonical correlations of 0.81, 0.65, 0.42, 0.24, 0.20, 0.15, and 0.15. out are reported in Supplementary Table 3. Distributions Figure 2 contains the scree plot of the explained variance of each of patients Age and WMH volume per site are reported CF and the correlation plot of the clinical and radiomic variates in Supplementary Figure 2. Briefly, when the center held- of the first canonical function with patients points colored out was showing large clinical differences (mainly younger according to their age. Loadings of the clinical and the five patient and/or neglectable loads of WMH), prediction most impactful radiomic variables (highest loadings) of the first performances were lower. two canonical functions are reported in Table 2. The bi-loading Frontiers in Neuroscience | www.frontiersin.org 5 July 2021 | Volume 15 | Article 691244 Bretzner et al. Radiomics of White Matter Hyperintensities plot in Figure 3 provides a graphical interpretation relationship between the most impactful variables of the first two canonical functions. Loadings of the clinical variate of all canonical functions are shown in Table 3; loadings of the radiomics variate are presented in Supplementary Table 1. Variables that share the same direction along a given function have a positive covariance, whereas variables that show opposing directions have negative covariance. The magnitude of the loading reflects the strength of the association. DISCUSSION Radiomic features, extracted outside of the visible WMH, captured latent characteristics of WMH and could accurately predict WMH burden. Upon further analysis, seven distinct combinations of radiomics features were associated seven distinct combinations of clinical traits relevant to WMH, such as age, sex, hypertension, history of smoking, DM, and CAD. Therefore, the methods presented here provide new tools to help to understand and quantify the microstructural portion of the parenchymal deterioration due to SVD in stroke and give a radiological snapshot of brain health. Importantly, our analyses relied on basic T2-FLAIR images, as commonly acquired in clinical routine and thus do not require any advanced, more costly additional imaging sequences. White matter hyperintensities represent a cardinal feature among radiological manifestations of brain aging and SVD. However, DTI-(Etherton et al., 2016) and PWI-based studies suggested (Promjunyakul et al., 2018) that WMH represent an end-stage macrostructural injury, embodying a surreptitious disease altering brain parenchyma. Our results support the hypothesis of WMH penumbra in cerebral SVD with a continuum between visible and invisible parenchymal damage (Maillard et al., 2011; Wardlaw et al., 2013). A major caveat of traditional advanced imaging biomarkers is their acquisition. Indeed, DTI sequences are rarely acquired routinely because of long scanning times, and PWI necessitates the injection of Gadolinium-based contrast agents. In contrast, our method can capture parenchymal microstructural integrity and hence, promises to replace additional dedicated imaging as a candidate approach to FIGURE 2 | Scree plot of the explained variance per canonical function and follow-up SVD progression in the clinic. correlation plot between the first clinical and radiomic variates. (A) Scree plot Cerebral atrophy has been shown to be associated with of the explained variance by canonical functions. (B) Correlation plot of the WMH burden (Aribisala et al., 2013). However, little is first clinical and radiomics canonical variates. Each dot represents a patient and is colored according to age. The first canonical function mainly known about the relationship between the texture of the represented age. There was a very strong correlation between the clinical and brain and WMH accumulation. By showing a twofold the radiomics variates of r = 0.81. improvement of predictions leveraging exhaustive textural information compared to predictions restricted to radiomics describing only shape and size, our findings suggest that, beyond atrophy, textural analysis of the brain might better By means of our CCA, we estimated the associations between document WMH related damage of the brain. Therefore, one the radiomic signature of WMH and SVD risk factors. The hypothesis would be that radiomics might be able to capture influence of cardiovascular risk factors on brain tissue was early-stage infra-radiological abnormalities prior to their previously investigated in neuropathology and advanced imaging evolution into irreversible cerebral loss, potentially bearing studies yet was rarely described by analyzing the texture of implications for future studies targeting WMH progression conventional imaging (Gouw et al., 2011; Wardlaw et al., 2013). prevention. Our work complement and support previous studies on MRI Frontiers in Neuroscience | www.frontiersin.org 6 July 2021 | Volume 15 | Article 691244 Bretzner et al. Radiomics of White Matter Hyperintensities TABLE 2 | Clinical and most impactful radiomic loadings of the first two canonical functions. Clinical loadings Radiomics loadings CF 1 CF 2 CF 1 CF 2 AF 0.310 0.005 LoG-1mm histogram 10 percentile −0.254 −0.128 Age 0.990 0.008 LoG-1mm GLSZM large area high gray level emphasis −0.747 −0.008 CAD 0.260 0.097 LoG-1mm GLSZM large area low gray level emphasis −0.743 −0.005 DM 0.127 0.009 LoG-2mm GLDM gray level non uniformity −0.514 0.671 Hypertension 0.381 −0.057 LoG-2mm GLRLM run variance −0.241 −0.124 Female sex 0.089 −0.993 LoG-2mm GLRLM short run low gray level emphasis 0.734 0.097 Smoking 0.069 0.180 LoG-3mm GLRLM gray level non uniformity normalized 0.300 −0.167 LoG-3mm GLRLM short run low gray level emphasis 0.767 0.073 Original histogram 10 percentile −0.733 −0.071 Original GLRLM run length non uniformity 0.662 0.221 Original GLRLM run length non uniformity normalized 0.658 0.051 Original GLRLM run variance −0.801 −0.013 Original shape major axis length −0.263 0.696 Original shape maximum 2D diameter column −0.162 0.745 Original shape mesh volume −0.608 0.581 Original shape minor axis length 0.046 0.709 Original shape sphericity −0.759 −0.172 Original shape surface volume ratio 0.778 0.044 Wavelet-LH GLSZM Small area high gray level emphasis −0.413 −0.161 AF, atrial fibrillation; CAD, coronary artery disease; DM, diabetes mellitus; LoG, Laplacian of Gaussian; GLCM, Gray Level Co-occurrence matrix; GLRLM, Gray Level Run Length Matrix; GLDM, Gray Level Dependence Matrix; NGTDM, Neighboring Gray Tone Difference Matrix. Loadings assess the relative contribution of a variable to a canonical function. textural analysis applied to SVD by Valdés et al. (2017) on microbleed, which is, along with WMH, radiological hallmarks gadolinium-enhanced T2-FLAIR, Bernal et al. (2020) on dynamic of SVD (Wardlaw et al., 2013). Radiomics presented here spectral gadolinium-enhanced T1 weighted imaging, Tozer et al. could therefore portray a representation of a pathological (2018) on T1 and T2-FLAIR cognitive textural biomarkers, and brain aging process in stroke patients, depicting atrophic and Shu et al. (2020) and Shao et al. (2018) who could predict heterogeneous parenchyma. the progression of WMH using radiomics extracted from, The second canonical function illustrated sex differences in respectively, T1-FLAIR and T2-FLAIR images. Our analyses tissue aspects in T2-FLAIR. The association of the radiomic were based on a large collection of clinical T2-FLAIR images, signature with sex was mainly driven by shape radiomics a routine MRI sequences acquired during both acute screening capturing differences in brain size. This finding remains, and follow-up of patients with stroke and cerebrovascular disease. however, independent from age-related atrophy since canonical Therefore, it argues for the overall clinical relevance of radiomics functions analyze the unexplained variance from the previous in stroke and SVD. function. Nevertheless, the female sex was also associated with Age was the clinical aspect correlating most strongly with the greater linear edge density (GLRLM after LoG filtering), which radiomic signature of WMH burden and is a well-established might indicate some sex-specific textural differences in the predictor of WMH (Rost et al., 2010; Giese et al., 2020). loss of microstructural integrity, as suggested in DTI with Similarly, blood-brain barrier studies using PWI highlight previous findings reporting sex-specific fractional anisotropy an age-associated increased leakage of contrast agents within values (Etherton et al., 2017b). WMH, but also beyond, in NAWM, showcasing a possible The third canonical function captured a profile representing preclinical pathogenic step leading to cognitive decline (Wardlaw mainly patients with a history of smoking, and, to a lesser extent, et al., 2017). Our findings also suggest the presence of age- diabetes, which shared common textural features describing related subvisible abnormalities that can notably be quantified more high spatial frequency changes in intensities which could on structural T2-FLAIR images. Radiomic features describing represent diffuse and fine heterogeneity throughout the brain. atrophy (brain size and lower sphericity) and T2-FLAIR The fourth canonical function characterized a specific relation heterogeneity, were the most strongly correlated with age. between hypertension and some textural features highlighting On the first canonical function, age was the main variable, inhomogeneity on a lower spatial frequency after wavelet however, HTN, AF, and CAD were also moderately represented, decomposition, thus describing a patchy texture. Since no other painting the picture of vascular pathological brain aging. The cardiovascular risk factor was represented on this dimension, it heterogeneity and hyperintensities of the parenchyma could describes an age-independent specific textural manifestation of have maybe captured lacunes, enlarged perivascular spaces, or hypertension on T2-FLAIR. Frontiers in Neuroscience | www.frontiersin.org 7 July 2021 | Volume 15 | Article 691244 Bretzner et al. Radiomics of White Matter Hyperintensities available sample size, allowing us to truly leverage both machine learning methodologies and multivariate modeling to select and characterize relevant radiomic variables in a data-driven fashion. In fact, to date and to the best of our knowledge, this is the largest radiomics study performed on any topic. Previous work on radiomics of SVD studied smaller datasets (<250 participants) and thus did not permit powerful unsupervised feature selections (Valdés et al., 2017; Tozer et al., 2018; Bernal et al., 2020). Another added value of the present study is its multicentric design. Our study is the first to explore radiomics of SVD in a large and multicentric population including diverse ischemic stroke cohorts with patients presenting a large spectrum of age and WMH burdens. While the heterogeneity of our dataset could be perceived as a limitation for prediction performances, we think it is on the contrary strength of our study. Indeed, our algorithm was trained on very diverse patients therefore theoretically increasing its chance of success if applied to an external dataset. By implementing multiple measures, such as down- sampling and intensity normalization, to prevent differences FIGURE 3 | Bi-loading plot of the variables projected over the two first originating from acquisition parameters discrepancies, we could canonical functions. A bi-loading plot graphically represents both the reach homogeneous results across all centers while capturing correlation of variables with canonical functions, and the correlation between relevant sources of variance, as depicted by the low error of our variables of each set. The position of a variable relative to an axis describes WMH burden predictions. Another source of unwanted variance the strength of the correlation of that variable with the axis, the closer to the in radiomics analyses is segmentation. Indeed, underestimation outer circle, the stronger the correlation. Clinical variables (red dot) and radiomic features (blue triangle) are positively correlated if close or negatively of the WMH burden, brain parenchyma, or ventricles could correlated if diagonally opposed. Blue tags were positioned next to correlated have impacted our radiomics based WMH burden prediction. radiomic features representing common textural concepts. On T2 FLAIR However, we here built upon previous results obtained with state- images, younger patients had larger brains and more homogeneous brain of-the-art deep learning-based, fully automated segmentation tissue (left side of the plot) whereas older patients had more atrophic and methods that could produce consistent outlines of brains, WMH, heterogeneous brains (right side of the plot). and ventricles from T2-FLAIR (Schirmer et al., 2019; Dubost et al., 2020). Preventive measures we implemented, especially down-sampling and intensity normalization, may have come at Diabetes mellitus was mainly represented in the fifth the cost of losing pertinent information. However, that impact dimension, correlating with textural features that illustrated might have been mitigated thanks to our large sample size. overall less hyperintense parenchyma and especially those We thus emphasize the capital importance of international obtained after filtering with LoG filters. Since those filters are collaborations, such as the MRI-GENIE consortium, to gather known to act as blob detectors, they potentially captured isolated large datasets, especially in the era of quantitative imaging and islets of damage. personalized medicine. The sixth canonical function related the presence of CAD and AF to a more homogeneous texture, which was, however, Limitations and Future Directions combined with a high impulse response to the LoG filters of 1, We acknowledge several limitations; first, stroke lesion outlines 2, and 3 mm sigma that could signify the presence of spots of were not available and thus not accounted for. Overall, the subvisible damage of varying size of presupposed embolic origin. median size of ischemic stroke lesions in this cohort is expected On the contrary, the seventh canonical function pictured the to be small, as the median NIHSS was 3. Moreover, the radiomic differences separating AF from CAD and DM patients, where analysis conducted here provides a single value per radiomic AF patients seemed to exhibit more patches of high spatial variable per patient, averaging the textural presentation over the frequency intensity changes, which could represent zones of whole extraction zone and thus largely decreasing the impact subtly lesioned brain. of small lesions. Regarding large lesions, the corresponding Diabetes and AF were represented by several dimensions perturbated radiomic value could have been assimilated to an meaning that the diseases in question could manifest in several outlier and then mitigated by the ElasticNet model, which distinctive aspects or stages in our data. Conditional factors includes an L1 regularization that improves its robustness to that could influence such diversity in presentations include extreme values. The absence of clear trend in the analysis of the relative control of disease by treatment or lifestyle, the the residual of WMH burden prediction per NIHSS score is in patient’s stage of disease severity, genetic predispositions, and favor of a limited impact of stroke lesions on the predictions. endophenotypes of varying severity. Other SVD imaging features were also not accounted for, such As with any work on radiomics, the main pitfall remains the as microbleeds or enlarged perivascular space, which have been curse of dimensionality, which refers to a very high number of previously linked to radiomic features (Valdés et al., 2017; Tozer variables. Consequently, one of the strengths of our study was the et al., 2018). Frontiers in Neuroscience | www.frontiersin.org 8 July 2021 | Volume 15 | Article 691244 Bretzner et al. Radiomics of White Matter Hyperintensities TABLE 3 | Clinical loadings for all seven canonical functions. Clinical loadings Canonical function 1 2 3 4 5 6 7 AF 0.310 0.005 0.102 0.020 0.313 −0.596 0.663 Age 0.990 0.008 0.096 0.080 −0.045 0.034 0.017 CAD 0.260 0.097 0.169 −0.157 −0.214 −0.744 −0.521 DM 0.127 0.009 −0.443 −0.117 0.781 −0.050 −0.401 Hypertension 0.381 −0.057 0.067 −0.916 0.055 0.058 0.030 Female sex 0.089 −0.993 −0.015 0.056 −0.004 −0.030 0.039 Smoking 0.069 0.180 −0.903 −0.055 −0.347 −0.132 0.081 AF, atrial fibrillation; CAD, coronary artery disease; DM, diabetes mellitus. A high loading coefficient implies a higher contribution to a canonical function. For instance, age was the most important variable when establishing the clinical variate of the first canonical function. Secondly, we could not study the relationship between manuscript. Requests to access the data should be directed to NR: radiomics and other known WMH biomarkers such nrost@partners.org. as dyslipidemia. Thirdly, we here suggest a novel biomarker to assess the structural integrity of the brain on routine T2-FLAIR ETHICS STATEMENT imaging. However, as with every new biomarker, the results The studies involving human participants were reviewed presented here would need external validation, especially to and approved by the MRI-GENIE project and by the MGH appreciate the robustness of the features included in the radiomic Institutional Review Board (IRB, Protocol #: 2001P001186 signature of WMH. and Protocol #: 2003P000836), as well as ethics boards of Fourthly, a substantial number of patients were excluded from the collaborating institutions: (Albrecht-Kossel Institute the analysis because of failed segmentation mainly due to image for Neuroregeneration (AKOS) University of Rostock, quality. While this might bias the analysis, it also highlights the Germany; University of Cincinnati (UC) College of Medicine, challenges of processing clinical imaging in a real-world setting. USA; Geisinger Institute, USA; Helsinki University Central Lastly, radiomics were extracted outside of the WMH Hospital, Finland; IMIM - Hospital del Mar, Spain; Imperial but not specifically within the white matter. Future research college of London, UK; Mayo Clinic Florida, USA; Medical could evaluate the impact of co-registration and resampling University Graz, Austria; Sahlgrenska Academy at University on radiomics of SVD, then benchmark radiomics of NAWM of Gothenburg, Sweden; Skane University Hospital, Sweden; against more traditional DTI metrics in the prediction of clinical Leuven University Hospitals, Belgium; University of British outcomes and therefore provide a more straightforward method Columbia, Canada; University of Maryland School of Medicine, to quantify microstructural integrity. USA; University of Miami, USA; University of Newcastle, Australia). All participants or health care proxy provided signed informed consent. The patients/participants provided their CONCLUSION written informed consent to participate in this study. In a large cohort of ischemic stroke patients, we demonstrated that radiomic features predicted WMH burden and were AUTHOR CONTRIBUTIONS associated with clinical factors. By applying machine learning methods to radiomics analyses of T2-FLAIR images from a MB, AB, MS, SH, ME, PR, RM, CW, RR, XL, RL, GK, and NR: large multi-site ischemic stroke cohort, we could characterize conception and design of the study. MB, AB, MS, SH, ADa, KD, the latent expression of SVD that extends beyond the visible A-KG, ME, PR, MN, OB, JC, ADo, CG, LHo, LHe, KJ, JJ-C, SK, WMH and subsequently uncover links associating cardiovascular RL, CL, PM, CM, JMe, C-LP, AR, SR, JRs, JRq, TR, RSa, RSc, PS, risk factors to distinct textural patterns. Radiomics analysis ASl, ASo, TS, DS, TT, VT, AV, JC, DW, OW, RZ, BW, JM, AL, may hold promise to become a cost-effective tool to quantify CJ, PG, and NR: acquisition and analysis of data. MB, AB, MS, microstructural damage on routinely acquired images in the ME, RM, CW, RR, XL, ReL, CG, JMe, JW, GK, and NR: drafting follow-up of SVD and stroke patients, once externally validated. a significant portion of the manuscript or figures. All authors contributed to the article and approved the submitted version. DATA AVAILABILITY STATEMENT FUNDING Data will be made available upon reasonable request and with The MRI-GENIE study was funded by NIH NINDS approval from local IRBs, to replicate the results presented in this (R01NS086905). MB was supported by the ISITE-ULNE Frontiers in Neuroscience | www.frontiersin.org 9 July 2021 | Volume 15 | Article 691244 Bretzner et al. Radiomics of White Matter Hyperintensities Fundation, the Société Française de Neuroradiologie, the ACKNOWLEDGMENTS Société Française de Radiologie, the Thérèse and René Planiol Fundation. ME was supported by the American Academy of A copy of a public version of this manuscript has been made Neurology and MGH Executive Council on Research. PR was available on https://www.biorxiv.org (Bretzner et al., 2021). supported by NIH K01 HL128791. TT was supported by the Helsinki University Central Hospital, Sigrid Juselius Foundation, Sahlgrenska University Hospital, and University of Gothenburg. SUPPLEMENTARY MATERIAL The funders had no role in design and conduct of the study; collection, management, analysis, and interpretation of the data; The Supplementary Material for this article can be found preparation, review, or approval of the manuscript; and decision online at: https://www.frontiersin.org/articles/10.3389/fnins. to submit the manuscript for publication. 2021.691244/full#supplementary-material REFERENCES Hotelling, H. (1936). Relations between two sets of variates. Biometrika 28, 321– 377. doi: 10.2307/2333955 Aribisala, B. S., Valdés Hernández, M. C., Royle, N. A., Morris, Z., Muñoz Maniega, Lambin, P., Rios-Velazquez, E., Leijenaar, R., Carvalho, S., van Stiphout, S., Bastin, M. E., et al. (2013). Brain atrophy associations with white matter R. G. P. M., Granton, P., et al. (2012). Radiomics: extracting more information lesions in the ageing brain: the Lothian Birth Cohort 1936. Eur. Radiol. 23, from medical images using advanced feature analysis. Eur. J. Cancer 48, 441– 1084–1092. doi: 10.1007/s00330-012-2677-x 446. doi: 10.1016/j.ejca.2011.11.036 Arsava, E. M., Rahman, R., Rosand, J., Lu, J., Smith, E. E., Rost, N. S., et al. (2009). Maillard, P., Fletcher, E., Harvey, D., Carmichael, O., Reed, B., Mungas, D., et al. Severity of leukoaraiosis correlates with clinical outcome after ischemic stroke. (2011). White matter hyperintensity penumbra. Stroke 42, 1917–1922. doi: Neurology 72, 1403–1410. doi: 10.1212/wnl.0b013e3181a18823 10.1161/strokeaha.110.609768 Au, R., Massaro, J. M., Wolf, P. A., Young, M. E., Beiser, A., Seshadri, S., et al. Oksanen, J., Kindt, R., Legendre, P., Hara, B., Simpson, G., Solymos, P., et al. (2006). Association of white matter hyperintensity volume with decreased (2019). vegan: Community Ecology Package. Available online at: https://CRAN. cognitive functioning: the Framingham heart study. Arch. Neurol. 63, 246–250. R-project.org/package=vegan (accessed November 29, 2020). doi: 10.1001/archneur.63.2.246 Pedregosa, F., Varoquaux, G., Gramfort, A., Michel, V., Thirion, B., Grisel, O., Bernal, J., Valdés-Hernández, M. D. C., Escudero, J., Viksne, L., Heye, A. K., et al. (2011). Scikit-learn: machine learning in python. J. Mach. Learn. Res. 12, Armitage, P. A., et al. (2020). Analysis of dynamic texture and spatial spectral 2825–2830. descriptors of dynamic contrast-enhanced brain magnetic resonance images Peterson, R. A., and Cavanaugh, J. E. (2020). Ordered quantile normalization: a for studying small vessel disease. Magn. Reson. Imaging 66, 240–247. doi: semiparametric transformation built for the cross-validation era. J. Appl. Stat. 10.1016/j.mri.2019.11.001 47, 2312–2327. doi: 10.1080/02664763.2019.1630372 Bretzner, M., Bonkhoff, A. K., Schirmer, M. D., Hong, S., Dalca, A., Donahue, Promjunyakul, N.-O., Dodge, H. H., Lahna, D., Boespflug, E. L., Kaye, J. A., K. L., et al. (2021). MRI radiomic signature of white matter hyperintensities Rooney, W. D., et al. (2018). Baseline NAWM structural integrity and CBF is associated with clinical phenotypes. bioRxiv[Preprint] 2021.01.24.427986, predict periventricular WMH expansion over time. Neurology 90, e2119– Brugulat-Serrat, A., Salvadó, G., Operto, G., Cacciaglia, R., Sudre, C. H., e2126. Grau-Rivera, O., et al. (2019). White matter hyperintensities mediate Rost, N. S., Rahman, R., Sonni, S., Kanakis, A., Butler, C., Massasa, E., et al. gray matter volume and processing speed relationship in cognitively (2010). Determinants of white matter hyperintensity volume in patients with unimpaired participants. Hum. Brain Mapp. 41, 1309–1322. doi: 10. acute ischemic stroke. J. Stroke Cerebrovasc. Dis. 19, 230–235. doi: 10.1016/j. 1002/hbm.24877 jstrokecerebrovasdis.2009.05.007 Dubost, F., de Bruijne, M., Nardin, M., Dalca, A. V., Donahue, K. L., Giese, A.- Schirmer, M. D., Dalca, A. V., Sridharan, R., Giese, A.-K., Donahue, K. L., Nardin, K., et al. (2020). Multi-atlas image registration of clinical data with automated M. J., et al. (2019). White matter hyperintensity quantification in large-scale quality assessment using ventricle segmentation. Med. Image Anal. 63:101698. clinical acute ischemic stroke cohorts – the MRI-GENIE study. Neuroimage doi: 10.1016/j.media.2020.101698 Clin. 23:101884. doi: 10.1016/j.nicl.2019.101884 Etherton, M. R., Wu, O., Cougo, P., Giese, A.-K., Cloonan, L., Fitzpatrick, K. M., Shao, Y., Chen, Z., Ming, S., Ye, Q., Shu, Z., Gong, C., et al. (2018). Predicting et al. (2017a). Integrity of normal-appearing white matter and functional the development of normal-appearing white matter with radiomics in the aging outcomes after acute ischemic stroke. Neurology 88, 1701–1708. doi: 10.1212/ brain: a longitudinal clinical study. Front. Aging Neurosci. 10:393. doi: 10.3389/ wnl.0000000000003890 fnagi.2018.00393 Etherton, M. R., Wu, O., Cougo, P., Giese, A.-K., Cloonan, L., Fitzpatrick, K. M., Shu, Z., Shao, Y., Xu, Y., Ye, Q., Cui, S., Mao, D., et al. (2020). Radiomics et al. (2017b). Structural integrity of normal appearing white matter and sex- nomogram based on MRI for predicting white matter hyperintensity specific outcomes after acute ischemic stroke. Stroke 48, 3387–3389. doi: 10. progression in elderly adults. J. Magn. Reson. Imaging 51, 535–546. doi: 10. 1161/strokeaha.117.019258 1002/jmri.26813 Etherton, M. R., Wu, O., and Rost, N. S. (2016). Recent advances in leukoaraiosis: Tong, D. C., Yenari, M. A., Albers, G. W., O’Brien, M., Marks, M. P., and white matter structural integrity and functional outcomes after acute ischemic Moseley, M. E. (1998). Correlation of perfusion- and diffusion-weighted MRI stroke. Curr. Cardiol. Rep. 18:123. with NIHSS score in acute (<6.5 hour) ischemic stroke. Neurology 50, 864–869. Giese, A.-K., Schirmer, M. D., Dalca, A. V., Sridharan, R., Donahue, K. L., Nardin, doi: 10.1212/wnl.50.4.864 M., et al. (2020). White matter hyperintensity burden in acute stroke patients Topakian, R., Barrick, T. R., Howe, F. A., and Markus, H. S. (2010). Blood– differs by ischemic stroke subtype. Neurology 95, e79–e88. brain barrier permeability is increased in normal-appearing white matter in Gillies, R. J., Kinahan, P. E., and Hricak, H. (2015). . Radiomics: images are patients with lacunar stroke and leucoaraiosis. J. Neurol. Neurosurg. Psychiatry more than pictures, they are data. Radiology 278, 563–577. doi: 10.1148/radiol. 81, 192–197. doi: 10.1136/jnnp.2009.172072 2015151169 Tozer, D. J., Zeestraten, E., Lawrence, A. J., Barrick, T. R., and Markus, González, I., Déjean, S., Martin, P. G. P., and Baccini, A. C. C. A. (2008). An R H. S. (2018). Texture analysis of T1-weighted and fluid-attenuated package to extend canonical correlation analysis. J. Stat. Softw. 23, 1–14. inversion recovery images detects abnormalities that correlate with Gouw, A. A., Seewann, A., van der Flier, W. M., Barkhof, F., Rozemuller, A. M., cognitive decline in small vessel disease. Stroke 49, 1656–1661. doi: 10. Scheltens, P., et al. (2011). Heterogeneity of small vessel disease: a systematic 1161/strokeaha.117.019970 review of MRI and histopathology correlations. J. Neurol. Neurosurg. Psychiatry Valdés, M. H., González-Castro, V., Chappell, F. M., Sakka, E., Makin, S., Armitage, 82, 126–135. doi: 10.1136/jnnp.2009.204685 P. A., et al. (2017). Application of Texture analysis to study small vessel disease Frontiers in Neuroscience | www.frontiersin.org 10 July 2021 | Volume 15 | Article 691244 Bretzner et al. Radiomics of White Matter Hyperintensities and blood-brain barrier integrity. Front. Neurol. 8:327. doi: 10.3389/fneur.2017. Conflict of Interest: AR was employed by the company Centogene GmbH.The 00327 remaining authors declare that the research was conducted in the absence of Wang, H.-T., Smallwood, J., Mourao-Miranda, J., Xia, C. H., Satterthwaite, T. D., any commercial or financial relationships that could be construed as a potential Bassett, D. S., et al. (2020). Finding the needle in a high-dimensional haystack: conflict of interest. canonical correlation analysis for neuroscientists. Neuroimage 216:116745. doi: 10.1016/j.neuroimage.2020.116745 Copyright © 2021 Bretzner, Bonkhoff, Schirmer, Hong, Dalca, Donahue, Giese, Wardlaw, J. M., Makin, S. J., Valdés Hernández, M. C., Armitage, P. A., Etherton, Rist, Nardin, Marinescu, Wang, Regenhardt, Leclerc, Lopes, Benavente, Heye, A. K., Chappell, F. M., et al. (2017). Blood-brain barrier failure Cole, Donatti, Griessenauer, Heitsch, Holmegaard, Jood, Jimenez-Conde, Kittner, as a core mechanism in cerebral small vessel disease and dementia: Lemmens, Levi, McArdle, McDonough, Meschia, Phuah, Rolfs, Ropele, Rosand, evidence from a cohort study. Alzheimers Dement. 13, 634–643. doi: 10. Roquer, Rundek, Sacco, Schmidt, Sharma, Slowik, Sousa, Stanne, Strbian, 1016/j.jalz.2016.09.006 Tatlisumak, Thijs, Vagal, Wasselius, Woo, Wu, Zand, Worrall, Maguire, Lindgren, Wardlaw, J. M., Smith, C., and Dichgans, M. (2013). Mechanisms of sporadic Jern, Golland, Kuchcinski and Rost. This is an open-access article distributed cerebral small vessel disease: insights from neuroimaging. Lancet Neurol. 12, under the terms of the Creative Commons Attribution License (CC BY). The use, 483–497. doi: 10.1016/s1474-4422(13)70060-7 distribution or reproduction in other forums is permitted, provided the original Wardlaw, J. M., Valdés Hernández, M. C., and Muñoz-Maniega, S. (2015). What author(s) and the copyright owner(s) are credited and that the original publication are white matter hyperintensities made of? Relevance to vascular cognitive in this journal is cited, in accordance with accepted academic practice. No use, impairment. J. Am. Heart Assoc. 4:001140. distribution or reproduction is permitted which does not comply with these terms. Frontiers in Neuroscience | www.frontiersin.org 11 July 2021 | Volume 15 | Article 691244