For many years, the study of focal brain lesions has played an important role in our effort to localize specific neurological functions to precise anatomic brain regions. Because of the acute onset and distinct anatomical localization of ischemic brain infarction, functional correlation studies have relied heavily on individuals with stroke. Although manual lesion analysis is the gold standard for such analyses, this approach is labor intensive and subject to bias, and it tends to limit the sample size.
Working with Dell Medical School Department of Neurology Chair David Paydarfar, MD, and Satwant Kumar, MBBS, PhD, of The University of Texas Center for Perceptual Systems, pediatric neurology resident Khushboo Verma, MBBS, developed an automated image algorithm to facilitate the characterization of chronic ischemic strokes on T1-weighted magnetic resonance images. Verma used 655 manually defined T1- weighted stroke images from the open-source dataset ATLAS (Anatomical Tracings of Lesions After Stroke) to refine the algorithm. She demonstrated a robust similarity coefficient between the manual and the automated volume calculations.
This automated segmentation model, trained on a large multicentric dataset, may enable accurate automated on-demand processing of MRI scans and quantitative chronic stroke lesion assessment.