Abstract P383: AI-Guided Detection and Volumetric Analysis of Embolic and Watershed Infarcts on Diffusion-Weighted MRI Enables Measurement of the Absolute and Relative Ischemic Infarct Burden

Abstract

Acute ischemic strokes frequently present with multiple restricted diffusion lesions scattered throughout the brain parenchyma, particularly embolic and watershed infarcts. The morphological variety and complexity of such infarcts are high in terms of their potential number, shape, volume, and spatial distribution. In current clinical practice, there is a limited description of the burden and variability of such infarcts. Here, we demonstrate an artificial-intelligence guided approach using high-performance computing, for the fast (minutes) and precise (sub-milliliter) detection and quantification of each infarcted brain volume regardless of size and morphology. Our methodology is based on processing and analyzing the original DICOM data by filtering, segmentation, and morphological neighborhood operations to identify and triage 3D-connected volumes of interest that represent ischemic stroke. Figure Panel A shows the analysis of a watershed stroke with the identification of 5 ischemic subvolumes with an overall infarct volume of 23.6 ml and a relative infarct volume of 2.0 % of the total brain volume of 1152 ml. In contrast, Figure Panel B shows the analysis of an embolic stroke with the identification of 20 ischemic subvolumes with an overall infarct volume of 51.6 ml and a relative infarct volume of 4.4 % of the total brain volume of 1178 ml. What is novel here is that our approach allows for the immediate measurement of the absolute (in ml) and relative (in terms of % of total brain volume) infarct burdens of complex strokes on initial MRI imaging. Volumetric analysis of stroke may represent a critical future biomarker. We suggest that such an approach might be easily implemented into clinical and research settings to guide clinical diagnoses, therapeutic decisions, and outcome prediction.