Artificial Intelligence software to interpret brain computed tomography in patients with stroke: external validation of e-ASPECTS software for Interpreting Brain CT in Stroke

Objective: The purpose of this study was to test e-ASPECTS software in stroke patients. Marketed as a decision-support tool, e-ASPECTS may detect ischemia or hemorrhage on CT imaging and quantify ischemic extent using ASPECTS.

Methods: Using CT scans from nine stroke studies, the software was compared with masked experts. e-ASPECTS results were assessed in patients with/without MCA ischemia, and the dataset was enriched with non-MCA ischemia, hemorrhage, and mimics to simulate a "front door" hospital population. With final diagnosis as reference standard, diagnostic accuracy was tested for ischemic features, hyperattenuated arteries, and hemorrhage.

Results: 4,100 patients were included (51% women, median age 78, NIHSS 10, onset to scan 2.5 hours); final diagnoses were ischemia (78%), hemorrhage (14%), or mimic (8%). Most e-ASPECTS results (69%) were within one point of expert ratings. Diagnostic accuracy was 71% for ischemic features and 85% for hemorrhage, with more false positives than experts (ischemia 12% vs 2%; hemorrhage 14% vs <1%).

Conclusions: On independent testing, e-ASPECTS showed moderate agreement with experts but overcalled stroke features. Future prospective trials are needed to evaluate the impact of AI software on patient care and outcomes before widespread implementation.