Functional anomaly mapping reveals local and distant dysfunction caused by brain lesions

Abstract

The lesion method has been important for understanding brain-behavior relationships in humans, but has previously used maps based on structural damage. Lesion measurement based on structural damage may label partly damaged but functional tissue as abnormal, and moreover, ignores distant dysfunction in structurally intact tissue caused by deafferentation, diaschisis, and other processes. A reliable method to map functional integrity of tissue throughout the brain would provide a valuable new approach to measuring lesions. Here, we use machine learning on four dimensional resting state fMRI data obtained from left-hemisphere stroke survivors in the chronic period of recovery and control subjects to generate graded maps of functional anomaly throughout the brain in individual patients. These functional anomaly maps identify areas of obvious structural lesions and are stable across multiple measurements taken months and even years apart. Moreover, the maps identify functionally anomalous regions in structurally intact tissue, providing a direct measure of remote effects of lesions on the function of distant brain structures. Multivariate lesion-behavior mapping using functional anomaly maps replicates classic behavioral localization, identifying inferior frontal regions related to speech fluency, lateral temporal regions related to auditory comprehension, parietal regions related to phonology, and the hand area of motor cortex and descending corticospinal pathways for hand motor function. Further, this approach identifies relationships between tissue function and behavior distant from the structural lesions, including right premotor dysfunction related to ipsilateral hand movement, and right cerebellar regions known to contribute to speech fluency. Brain-wide maps of the functional effects of focal lesions could have wide implications for lesion-behavior association studies and studies of recovery after brain injury.

Keywords: Aphasia; Functional anomaly mapping; Lesion; Lesion-behavior mapping; Machine learning; Stroke; Support vector machine.