Performance of automated CT ASPECTS in comparison to physicians at different levels on evaluating acute ischemic stroke at a single institution in China

Abstract

Background: Our aim was to evaluate the sensitivity and specificity of the automated computer-based Alberta Stroke Program Early CT Score (e-ASPECTS) for acute stroke patients and compare the result with physicians at different levels. Methods: In our center, e-ASPECTS and 9 physicians at different levels retrospectively and blindly assessed baseline computed tomography (CT) images of 55 patients. Sensitivity, specificity, receiver-operating characteristic curves, Bland–Altman plots with mean score error, and Matthews correlation coefficients were calculated. Comparisons were made between the scores by physicians and e-ASPECTS with diffusion-weighted imaging (DWI) being the ground truth. Two methods for clustered data were used to estimate sensitivity and specificity in the region-based analysis. Results: In total, 1100 (55 patients × 20 regions per patient) ASPECTS regions were scored. In the region-based analysis, sensitivity of e-ASPECTS was better than junior doctors and residents (0.576 vs 0.165 and 0.111, p < 0.05) but inferior to senior doctors (0.576 vs 0.617). Specificity was lower than junior doctors and residents (0.883 vs 0.971 and 0.914) but higher than senior doctors (0.883 vs 0.809, p < 0.05). E-ASPECTS had the best Matthews correlation coefficient of 0.529, compared to senior doctors, junior doctors, and residents (0.463, 0.251, and 0.087, respectively). Conclusions: e-ASPECTS showed a similar performance to that of senior physicians in the assessment of brain CT of acute ischemic stroke patients with the Alberta Stroke Program Early CT score method.