Automated Calculation of Alberta Stroke Program Early CT Score: Validation in Patients With Large Hemispheric Infarct

Abstract

Background and Purpose- We compared the Alberta Stroke Program Early CT Score (ASPECTS), calculated using a machine learning-based automatic software tool, RAPID ASPECTS, as well as the median score from 4 experienced readers, with the diffusion-weighted imaging (DWI) ASPECTS obtained following the baseline computed tomography (CT) in patients with large hemispheric infarcts. Methods- CT and magnetic resonance imaging scans from the GAMES-RP study, which enrolled patients with large hemispheric infarctions (82-300 mL) documented on DWI-magnetic resonance imaging, were evaluated by blinded experienced readers to determine both CT and DWI ASPECTS. The CT scans were also evaluated by an automated software program (RAPID ASPECTS). Using the DWI ASPECTS as a reference standard, the median CT ASPECTS of the clinicians and the automated score were compared using the interclass correlation coefficient. Results- The median CT ASPECTS for the clinicians was 5 (interquartile range, 4-7), for RAPID ASPECTS 3 (interquartile range, 1-6), and for DWI ASPECTS 3 (2-4). Median error for RAPID ASPECTS was 1 (interquartile range, -1 to 3) versus 3 (interquartile range, 1-4) for clinicians (P<0.001). The automated score had a higher level of agreement with the median of the DWI ASPECTS, both for the full scale and when dichotomized at <6 versus 6 or more (difference in intraclass correlation coefficient, P=0.001). Conclusions- RAPID ASPECTS was more accurate than experienced clinicians in identifying early evidence of brain ischemia as documented by DWI.

Keywords: brain; cerebral infarction; machine learning; magnetic resonance imaging; tomography.