We aim to develop an AI-based model to detect Acute Ischemic Stroke, Large Vessel Occlusion and cerebral collateral status using Non-Contrast CT (NCCT) and CT Angiography (CTA).

Stroke is a leading cause of morbidity and mortality in the world. Access to specialized care (thrombolysis and mechanical thrombectomy) for stroke is limited to tertiary care hospitals, mostly in cities. Increasingly younger people are being affected by stroke leading to more loss of DALYs. By making life saving ischemic stroke therapies accessible in rural hospitals we can drastically improve outcomes after ischemic strokes in the developing world.

This retrospective observational study assessed diagnostic accuracy in patients with suspected acute ischemic stroke (AIS) in the anterior circulation at a University hospital in Western India from October 2022 to May 2025. Data sets were created for each patient with the following inputs (Labels): demographics, vascular risk factors and imaging. NCCT was semi-automatically segmented for CT ASPECTS using ITK SNAP (Version 4.2.2). Outcomes (output) were Clot Burden Score (on CTA) and cerebral collaterals (classified as good or poor using TAN, MAAS, MITEFF and Multiphase CTA scores).

Of 156 patients 105 (67.3%) were male (Mean age 62.3±12.5y). Vascular risk factors included hypertension (56.6%), dyslipidemia (53.2%), hyperhomocysteinemia (40.9%), diabetes (36.4%). Median NIHSS was 8 (IQR 4-14). Median CT ASPECTS was 9 (IQR 8-10). Median Clot Burden Score was 9 (IQR 8-10). Cerebral collaterals were good in 87 (55.8%). Seven machine learning models were trained. X G Boost model performed best with 85% sensitivity, 90% specificity and AUC of 0.88 for AIS detection; 87% sensitivity and 85% specificity for LVO and 82% accuracy for collateral status (good vs poor).