Consecutive large vessel occlusion patients with pre-EVT MRI were identified from two centers (2011-2019). Acute infarcts were extracted through a deep learning-enabled pipeline from DWI and spatially normalized. Brains were parcellated (atlas-defined 94 cortical regions, 14 subcortical nuclei, 20 white matter tracts) and reduced to ten principal lesion patterns using unsupervised dimensionality reduction techniques (non-negative matrix factorization). Binary HT, defined as ECASS PH1 or PH2, was modeled via Bayesian regression, using lesion patterns as inputs, and controlling for total lesion volume, age, sex, initial NIH Stroke Scale (NIHSS), thrombolysis, good reperfusion (TICI 2b-3), acute stenting, last known well-to-puncture time, and other risk factors.

These data from a large, multicenter cohort with precise MRI-defined infarcts underscore the risk of specific brain regions infarcted for HT after EVT. Understanding this pathophysiology can inform not only current clinical practice but also the development of future novel therapeutics to prevent HT as more patients with large cores undergo EVT.