Sacco, Ralph L.

Introduction: In order to increase the efficiency and effectiveness of neurovascular clinical research studies, increase data quality, facilitate data sharing, help educate new clinical investigators and reduce study start-up time, the National Institute of Neurological Disorders and Stroke (NINDS) convened a Working Group (WG) that developed Version 1.0 (published 2010) Stroke-specific Common Data Elements (CDEs). Since their initial publication, intervening advances in science and initial experience with the CDEs identified a need to update them and refine guidance on their deployment. Hypothesis/Objective: The NINDS has updated guidance on uniform data structures for use in cerebrovascular research in epidemiology, clinical trials and imaging studies in order to advance the prevention, acute treatment and recovery from cerebrovascular disease. Methods: The NINDS convened experts in research and data element design drawing strongly from investigators in the NIH StrokeNet and other NINDS clinical research projects. Results: Stroke CDE leadership developed a revised process for classifying Stroke CDEs among the four hierarchical categories of Core, Supplemental - Highly Recommended, Supplemental and Exploratory. Due to the heterogeneity of stroke conditions and study types, the classification of Supplemental - Highly Recommended was used for study type (clinical trial or observational), disease type (e.g., ischemic stroke, intracerebral hemorrhage, subarachnoid hemorrhage) and disease phase (primary prevention, acute, recovery and secondary prevention). Conclusion: The second iteration of NINDS CDE recommendations for neurovascular disease is an important step towards more efficient study start-up time and improved data sharing. The updated CDEs were released on the NINDS CDE website in May 2015. The information at this meeting will include examples of how the Stroke CDEs may be used by a research study, an explanation of the new CDE classifications, and examples of navigating and selecting CDEs from the NINDS CDE website. Support: This project was funded by HHSN271201200034C.

Objectives: Identification of gene variants of stroke subtypes is important for the development of tailored ischemic stroke therapies among various ethnic groups. Valid and reliable determination of ischemic stroke subtype is essential for achieving this goal and to standardize a classification scheme across multi-center studies and different populations. Causative Classification System for Ischemic Stroke (CCS) is a novel computerized subclassification tool developed to improve reliability and accuracy of classifying stroke types. The CCS algorithm relies on both phenotypic and causative stroke variables. A Hispanic subset of the SiGN, an important and distinct target population with greater risk of certain stroke subtypes, was evaluated with Trial of Org 10172 in Acute Stroke Treatment (TOAST) and CCS and the agreement between the two classification systems was analyzed. Methods: Over 6000 subjects at 15 sites across US and Europe were enrolled, with TOAST and CCS locally adjudicated. Blood collection and central data quality control (10% central readjudication) were performed on all participants. A subset of Hispanics was analyzed for the purpose of this study and the agreement between the TOAST and CCS were assessed by kappa statistic. Findings: Hispanics (n=595, 10.9%) compared to non-Hispanics (n=5457) were more likely to be younger (63.7 vs. 64.0), male (55% vs. 46%) and have fewer of the traditional stroke risk factors HTN (54% vs. 64%), Afib (11% vs. 14%), DM(23% vs. 25%), CAD(16% vs. 20%) and smoking(19% vs. 22%). While the TOAST showed no differences between stroke subtypes for Hispanic vs. non-Hispanics, in CCS, Hispanics were classified with more of large vessel (22% vs. 20%), cardioembolic (37% vs. 30%) and small vessel strokes (13% vs. 9%) and fewer with undetermined etiology (28% vs. 40%) as compared to non-Hispanics. TOAST and CCS offered moderate correlation across all stroke types in Hispanics: kappa of 0.66 for large artery atherosclerosis, 0.58 for cardioembolic, and 0.58 for small artery occlusion. Conclusion: CCS offers a more sensitive and accurate system for subphenotyping of strokes in Hispanics who tended to have relatively fewer risk factors and unclassified strokes. Further studies correlating the two classification systems and their relation to genotyping data are warranted.

Background: Fundamental advances in stroke care will require pooling imaging phenotype data from multiple centers, to complement the current aggregation of genomic, environmental, and clinical information. Sharing clinically acquired MRI data from multiple hospitals is challenging due to inherent heterogeneity of clinical data, where the same MRI series may be labeled differently depending on vendor and hospital. Furthermore, the de-identification process may remove data describing the MRI series, requiring human review. However, manually annotating the MRI series is not only laborious and slow but prone to human error. In this work, we present a recurrent convolutional neural network (RCNN) for automated classification of the MRI series. Methods: We randomly selected 1000 subjects from the MRI-GENetics Interface Exploration study and partitioned them into 800 training, 100 validation and 100 testing subjects. We categorized the MRI series into 24 groups (see Table). The RCNN used a modified AlexNet to extract features from 2D slices. AlexNet was pretrained on ImageNet photographs. Since clinical MRI are 3D and 4D, a gated recurrent unit neural network was used to aggregate information from multiple 2D slices to make the final prediction. Results: We achieved a classification accuracy (correct/total cases) of 99.8%, 98.5% and 97.5% on the training, validation and testing set, respectively. The averaged F1-score (percent overlap between predicted cases and actual cases) over all categories were 99.8% 98.2% and 94.4% on the training, validation and testing set. Conclusion: We showed that automated annotation of MRI series by repurposing deep-learning techniques used for photographic image recognition tasks is feasible. Such methods can be used to facilitate high throughput curation of MRI data acquired across multiple centers and enable scientifically productive collaboration by researchers and, ultimately enhancing big data stroke research.

Introduction: White matter hyperintensity volume (WMHv) is an important and highly heritable cerebrovascular phenotype; however, manual or semi-automated approaches to clinically acquired MRI analysis hinder large-scale studies in acute ischemic stroke (AIS). In this work, we develop a high-throughput, fully automated WMHv analysis pipeline for clinical fluid-attenuated inversion recovery (FLAIR) images to facilitate rapid genetic discovery in AIS. Methods: Automated WMHv extraction from multiple subjects relies on significant pre-processing of medical scans, including co-registration of the images. To reduce the effects of anisotropic voxel sizes, each FLAIR image is upsampled using bi-cubic interpolation. Brain extraction is performed using RObust Brain EXtraction (ROBEX). Images are then registered to an in-house FLAIR template using Advanced Normalization Tools (ANTs). The spatial covariation of WMH is learned through principal component analysis (PCA) of manual outlines from 100 subjects. Areas of leukoaraiosis are identified and separated from other lesions using the PCA modes. Volumes are then computed using non-interpolated slices for each subject. Standard deviation (SD) in WMHv (9 subjects; 6 raters each) is calculated as a measure of variability. Good agreement between automated and manual outlines is assessed in 358 subjects (automated WMHv within 3SD of manual WMHv). Results: As part of the MRI - Gen etics I nterface E xploration (MRI-GENIE) study, WMHv were calculated on a set of 2703 FLAIR images of patients from 12 independent AIS cohorts (sites). Results are shown in Figure 1. Comparing manual and automated WMHv shows that 88% of the automated WMHv fall within 3 SD from the manual WMHv, suggesting good agreement. Conclusion: WMHv segmentation using a fully-automated pipeline for analysis of clinical MRIs is both feasible and accurate. Ongoing analysis of the extracted WMHv is expected to advance current knowledge of risks and outcomes in AIS.

Introduction: Even though neuroinflammation is increasingly recognized as an essential contributor to ischemic brain injury, the exact underlying mechanisms remain unclear. Higher expression of the vascular cell adhesion molecule 1 (VCAM-1) increases leukocyte-brain endothelium interaction and has been associated with larger infarct size following acute ischemic stroke (AIS). The activation of the TGF-β signaling pathways can down-regulate VCAM-1 expression and ameliorate deleterious tissue outcome during neuroinflammation. Hypothesis: We sought to investigate whether genetic variation in the TGF-beta pathway and adhesion molecule genes is associated with acute stroke lesion size. Methods: We completed genome-wide association (GWA) testing and diffusion-weighted imaging lesion volume (DWIv) analysis in a discovery cohort of 532 AIS patients of European ancestry enrolled within 48 hours of symptom onset. An independent European ancestry cohort of 724 AIS patients with GWA data and automated DWIv served as replication cohort. GWA testing per SNP was performed using linear regression modeling of natural log-transformed DWIv adjusted for age, sex and relevant principal components. We selected 42 inflammatory genes in the TGF-beta pathway for a gene-based analysis using VEGAS (Versatile Gene-based -Association Study) software. A pre-specified discovery phase Bonferroni-corrected threshold was set at p<0.001. In the replication phase, 14 SNPs overlapping were tested at the Bonferroni-corrected p-value threshold of p<0.004. Results: Of all genes in the TGF-beta pathway, VCAM1 (p=0.0006) was significantly associated with DWIv in the discovery AIS cohort (age: 66 ± 14.9 years, sex: 63.4% male, DWIv: 2.2cm 3 (IQR: 0.6-11.7cm 3 )). A single SNP within the VCAM1 gene boundaries (rs3176876 (BETA=0.2341, p=0.003)) was significantly associated with DWIv in the replication AIS cohort (age: 65 ± 14.1 years, sex: 68.6% male, DWIv: 4cm 3 (IQR: 1.3-17.2cm 3 )). Conclusion: The genetic variant rs3176876 in the VCAM1 gene is associated with larger infarct lesion size measured on brain MRI of AIS patients. These findings suggest genetic contribution to the pro-inflammatory mechanisms in acute cerebral ischemia and warrant further investigation.

Introduction: White matter hyperintensity (WMH) is a highly heritable trait and a significant contributor to stroke risk and severity. Vascular risk factors contribute to WMH severity; however, knowledge of the determinants of WMH in acute ischemic stroke (AIS) is still limited. Hypothesis: WMH volume (WMHv) varies across AIS subtypes and is modified by vascular risk factors. Methods: We extracted WMHv from the clinical MRI scans of 2683 AIS subjects from the MRI-Genetics Interface Exploration (MRI-GENIE) study using a novel fully-automated, volumetric analysis pipeline. Demographic data, stroke risk factors and stroke subtyping for the Causative Classification of Stroke (CCS) were performed at each of the 12 international study sites. WMHv was natural log-transformed for linear regression analyses. Results: Median WMHv was 5.7cm 3 (interquartile range (IQR): 2.2-12.8cm 3 ). In univariable analysis, age (63.1 ± 14.7 years, β=0.04, SE=0.002), prior stroke (10.2%, β=0.66, SE=0.08), hypertension (65.4%, β=0.75, SE=0.05), diabetes mellitus (23.1%, β=0.35, SE=0.06), coronary artery disease (17.6%, β=0.04, SE=0.002), and atrial fibrillation (14.6%, β=0.48, SE=0.07) were significant predictors of WMHv (all p<0.0001), as well as smoking status (52.2%, β=0.15, SE=0.05, p=0.005), race (16.5% Non-Caucasian, β=0.25, SE=0.07) and ethnicity (8.2% Hispanic, β=0.30, SE=0.11) (all p<0.01). In multivariable analysis, age (β=0.04, SE=0.002), prior stroke (β=0.56, SE=0.08), hypertension (β=0.33, SE=0.05), smoking status (β=0.16, SE=0.05), race (β=0.42, SE=0.06), and ethnicity (β=0.34, SE=0.09) were independent predictors of WMHv (all p<0.0001), as well as diabetes mellitus (β=0.13, SE=0.06, p=0.02). WMHv differed significantly (p<0.0001, unadjusted) across CCS stroke subtypes: cardioembolic stroke (8.0cm 3 , IQR: 4.2-15.4cm 3 ), large-artery stroke (6.9cm 3 , IQR: 3.1-14.7cm 3 ), small-vessel stroke (5.8cm 3 , IQR: 2.5-13.5cm 3 ), stroke of undetermined (4.7cm 3 , IQR: 1.6-11.0cm 3 ) or other (2.55cm 3 , IQR: 0.9-8.8cm 3 ) causes. Conclusion: In this largest-to-date, multicenter hospital-based cohort of AIS patients with automated WMHv analysis, common vascular risk factors contribute significantly to WMH burden and WMHv varies by CCS subtype.

Background and Purpose— We evaluated deep learning algorithms’ segmentation of acute ischemic lesions on heterogeneous multi-center clinical diffusion-weighted magnetic resonance imaging (MRI) data sets and explored the potential role of this tool for phenotyping acute ischemic stroke. Methods— Ischemic stroke data sets from the MRI-GENIE (MRI-Genetics Interface Exploration) repository consisting of 12 international genetic research centers were retrospectively analyzed using an automated deep learning segmentation algorithm consisting of an ensemble of 3-dimensional convolutional neural networks. Three ensembles were trained using data from the following: (1) 267 patients from an independent single-center cohort, (2) 267 patients from MRI-GENIE, and (3) mixture of (1) and (2). The algorithms’ performances were compared against manual outlines from a separate 383 patient subset from MRI-GENIE. Univariable and multivariable logistic regression with respect to demographics, stroke subtypes, and vascular risk factors were performed to identify phenotypes associated with large acute diffusion-weighted MRI volumes and greater stroke severity in 2770 MRI-GENIE patients. Stroke topography was investigated. Results— The ensemble consisting of a mixture of MRI-GENIE and single-center convolutional neural networks performed best. Subset analysis comparing automated and manual lesion volumes in 383 patients found excellent correlation (ρ=0.92; P <0.0001). Median (interquartile range) diffusion-weighted MRI lesion volumes from 2770 patients were 3.7 cm 3 (0.9–16.6 cm 3 ). Patients with small artery occlusion stroke subtype had smaller lesion volumes ( P <0.0001) and different topography compared with other stroke subtypes. Conclusions— Automated accurate clinical diffusion-weighted MRI lesion segmentation using deep learning algorithms trained with multi-center and diverse data is feasible. Both lesion volume and topography can provide insight into stroke subtypes with sufficient sample size from big heterogeneous multi-center clinical imaging phenotype data sets.

Introduction: Genome-wide association (GWA) studies have contributed substantially to the understanding of complex vascular traits including ischemic stroke; however, genetic determinants of acute cerebral ischemia remain to be elucidated. Objective: To investigate the genetic architecture of cerebral infarct lesion burden in patients with acute ischemic stroke (AIS). Methods: Twelve international sites contributed 3,301 AIS subjects with acute MRI and genome-wide genotyping to the MRI-GENIE study. In a preliminary analysis of 657 subjects of European ancestry, acute cerebral infarct lesions were outlined on diffusion-weighted images (DWI) using a semi-automated volumetric method. Standard quality control measures were performed per single nucleotide polymorphism (SNP) and per subject. SNPs were imputed to the Haplotype Reference Consortium v1.1 panel. Natural log-transformed DWI volume was used for GWA analysis with allelic dosage per SNP, age, sex and principal components 1-5 of genetic ancestry as covariates. Results: GWA testing for DWI volume in 7.7 million SNPs yielded no signal crossing the Bonferroni-corrected genome-wide significance threshold of p <5*10 -8 . However, several loci passed the nominal significance threshold of p <1*10 -6 (Figure 1), including a locus on chromosome 2 in the SPATA3-AS1 (spermatogenesis associated 3 antisense RNA 1) gene (lead SNP: rs2368999, MAF=36%, p =8.4*10 -7 ) and an intronic locus on chromosome 10 in the ATRNL1 (Attractin like 1) gene (lead SNP: rs592284, MAF=10%, p =9.1*10 -7 ). Conclusion: In this first-to-date, preliminary GWAS of acute cerebral infarct volume in AIS patients, we identified several new, nominally significant loci including a locus in ATRNL1, a gene previously linked to carotid plaque burden. Further analyses of the MRI-GENIE cohort to replicate these findings, to include stroke subtype specific analyses, and to increase the overall statistical power of the study are ongoing.

Background: Current ischemic stroke (IS) subtyping systems are very labor-intensive and costly to implement in large samples. There is a need to develop phenotyping approaches for genetic and other studies that can be implemented using generally available items captured in electronic medical records. Objective: In this report, we seek to validate a simplified approach using results from a genetic association study. Specifically, the strength of association for known genetic signals in the NINDS Stroke Genetics Network study (SiGN) is compared between the established IS subtyping systems (TOAST and Causative Classification System (CCS) subtypes) and phenotyping based on a single item within the CCS. Methods: Non-mutually exclusive “Simple” IS subtypes were defined from single CCS items as follows: 1) Small vessel stroke- lacunar infarct on imaging (with no other imaging or other requirements); 2) large artery stroke - vascular imaging showing > 50% stenosis or <50% stenosis with plaque rupture or thrombosis judged to be due to atherosclerosis in clinically relevant artery; and 3) cardioembolic stroke - persistent or paroxysmal atrial fibrillation. We compared the GWAS signals using these definitions to the published SiGN GWAS results for TOAST and CCS, among those of European ancestry only. Results: The Table shows that, for loci significant in SiGN GWAS study, the “Simple” IS subtypes had similar strengths of association and p-values compared to results from TOAST and the causative and phenotypic CCS subtypes. Conclusions: “Simple” IS subtypes perform as well in GWAS analyses as more complex phenotyping systems and could potentially be derived through automated or semi-automated methods applied to electronic medical record systems. Since the ‘Simple” phenotypes were based on high quality CCS data, the current findings reflect optimal performance. Additional validation studies will be required for implementation in other settings.

Background and Purpose— NINDS (National Institute of Neurological Disorders and Stroke)-SiGN (Stroke Genetics Network) is an international consortium of ischemic stroke studies that aims to generate high-quality phenotype data to identify the genetic basis of pathogenic stroke subtypes. This analysis characterizes the etiopathogenetic basis of ischemic stroke and reliability of stroke classification in the consortium. Methods— Fifty-two trained and certified adjudicators determined both phenotypic (abnormal test findings categorized in major pathogenic groups without weighting toward the most likely cause) and causative ischemic stroke subtypes in 16 954 subjects with imaging-confirmed ischemic stroke from 12 US studies and 11 studies from 8 European countries using the web-based Causative Classification of Stroke System. Classification reliability was assessed with blinded readjudication of 1509 randomly selected cases. Results— The distribution of pathogenic categories varied by study, age, sex, and race ( P <0.001 for each). Overall, only 40% to 54% of cases with a given major ischemic stroke pathogenesis (phenotypic subtype) were classified into the same final causative category with high confidence. There was good agreement for both causative (κ 0.72; 95% confidence interval, 0.69–0.75) and phenotypic classifications (κ 0.73; 95% confidence interval, 0.70–0.75). Conclusions— This study demonstrates that pathogenic subtypes can be determined with good reliability in studies that include investigators with different expertise and background, institutions with different stroke evaluation protocols and geographic location, and patient populations with different epidemiological characteristics. The discordance between phenotypic and causative stroke subtypes highlights the fact that the presence of an abnormality in a patient with stroke does not necessarily mean that it is the cause of stroke.