Bouts, Mark J.R.J.

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.

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.