Extension of voxel-based lesion mapping to multidimensional neurophysiological data

Focal brain lesions cause neurophysiological changes in local and distributed neural systems. While electroencephalography (EEG) has a long history in post-stroke neurophysiological assessment, the observed changes have rarely been linked to specific lesion locations, leaving neuroanatomical-neurophysiological relationships after stroke unclear. Current data-driven methods, such as voxel-based lesion symptom mapping (VLSM), relate lesion locations to single-feature "symptoms" but currently cannot associate anatomical injury with multidimensional data such as EEG, with its rich spatiotemporal information. To overcome this limitation, we introduce MD-VLM, an extension of VLSM to multidimensional "symptoms" that identifies relationships between lesion locations and neurophysiology. MD-VLM is data-agnostic, compatible with various lesion (e.g., lesion maps, lesion network maps) and neurophysiological (e.g., channel-level or source-localized EEG) inputs, and uses robust statistics to test for the existence of significant neuroanatomical-neurophysiological relationships. We demonstrate MD-VLM's feasibility by applying it to EEG from chronic stroke patients performing a cued-movement task. MD-VLM revealed significant associations between frontal white-matter lesions and reduced ipsilesional parietal cue-evoked responses, consistent with damage to known fronto-parietal networks. MD-VLM is a novel data-driven extension to VLSM for multidimensional "symptoms". Applying MD-VLM to link lesions to neurophysiological data can improve mechanistic understanding of post-stroke neurological impairments and guide future biomarker development.