Localization of specific language pathways using diffusion-weighted imaging tractography for presurgical planning of children with intractable epilepsy

    loading  Checking for direct PDF access through Ovid



To examine whether diffusion-weighted imaging (DWI) tractography can detect multiple white matter pathways connected to language cortices, we employed a maximum a posteriori probability (MAP) classification method, which has been recently validated for the corticospinal tract.


DWI was performed in 12 normally developing children and 17 children with intractable focal epilepsy who underwent subsequent two-stage epilepsy surgery with intracranial functional mapping. First, whole-brain DWI tractography was performed to identify unique pathways originating from Broca's area, premotor area, and Wernicke's area on functional magnetic resonance imaging (fMRI) of normal children and intracranial electrical stimulation mapping (ESM) of children with epilepsy. Group averaging of these pathways based on fMRI was performed to construct the probability maps of language areas in standard MRI space. These maps were finally used to design a DWI-MAP classifier, which can automatically sort individual fibers originating from fMRI language areas as well as ESM language areas.


In normally developing children, the DWI-MAP classifier predicted language-activation areas on fMRI with up to 77% accuracy. In children with focal epilepsy, the DWI-MAP classifier also showed high accuracy (up to 82%) for the fibers terminating in proximity to essential language areas determined by ESM. Decreased volumes in DWI-MAP–defined pathways after epilepsy surgery were associated with postoperative language deficits.


This study encourages further investigations to determine if DWI-MAP analysis can serve as a noninvasive diagnostic tool during pediatric presurgical planning by estimating not only the location of essential language cortices, but also the underlying fibers connecting these cortical areas.

Related Topics

    loading  Loading Related Articles