Predicting aphasia type from brain damage measured with structural MRI.

摘要: Chronic aphasia is a common consequence of left-hemisphere stroke. Since the early insights by Broca and Wernicke, studying relationship between loci cortical damage patterns language impairment has been one concerns aphasiology. We utilized multivariate classification in cross-validation framework to predict type chronic from spatial pattern brain damage. Our sample consisted 98 patients with five types (Broca's, Wernicke's, global, conduction, anomic), classified based on scores Western Aphasia Battery (WAB). Binary lesion maps were obtained structural MRI scans (obtained at least 6 months poststroke, within 2 days behavioural assessment); after normalization, lesions parcellated into disjoint set areas. The proportion areas was used classify patients' type. To create this parcellation, we relied atlases; our classifier (support vector machine - SVM) could differentiate different kinds using any parcellations. In sample, best accuracy when novel parcellation that combined two previously published atlases, first atlas providing segmentation grey matter, second segment white matter. For each type, computed relative importance for distinguishing it other types; findings consistent reports locations implicated aphasia. Overall, results revealed automated distinguish atlas-defined