Leptomeningeal Collateral Detection on DSA via Vessel-Graph Neural Networks
Researchers have developed a novel framework to detect individual leptomeningeal collaterals (LMCs) in digital subtraction angiography (DSA) by treating collateral detection as a vessel segment classification problem on a graph. This hybrid graph-pixel architecture integrates a topology-aware graph branch with a dense pixel branch, achieving a PR-AUC of 0.434 in cross-validation, surpassing graph-only and pixel-only baselines. This method offers a more objective and precise quantitative assessment of LMCs compared to current subjective manual grading scales, potentially aiding future biomarker and pattern discovery. AI
IMPACT This research could lead to more objective and precise diagnoses for acute ischemic stroke patients by improving the assessment of blood vessel function.