e-ASPECTS software improves interobserver agreement and accuracy of interpretation of aspects score

Introduction: Artificial intelligence-based software, like Brainomix e-ASPECTS, has gained attention for its potential in improving the evaluation of neuroimaging studies for acute ischemic stroke. This study aimed to assess whether Brainomix e-ASPECTS software improves interobserver agreement and accuracy in detecting ASPECTS regions affected by anterior circulation large vessel occlusion (LVO) compared to standard image interpretation without AI.

Methods: We included 60 consecutive patients with anterior circulation LVO who underwent TICI 3 revascularization within 60 minutes of baseline CT. A total of 16 readers first interpreted CT scans independently, then re-evaluated the same scans two months later with e-ASPECTS assistance. Interclass correlation coefficient (ICC) and interobserver agreement (Fleiss' Kappa) were assessed with and without AI, with 24-hour CT scans as ground truth.

Results: Without e-ASPECTS, the ICC for total ASPECTS was 0.395 (fair); with e-ASPECTS it increased to 0.574 (good) (P < 0.01). Significant improvement in inter-rater agreement was observed for each region except M6 and caudate, and the software showed higher accuracy than the overall reader cohort except for the caudate.

Conclusions: Brainomix e-ASPECTS software significantly improved inter-rater agreement and accuracy in evaluating ASPECTS scores, outperforming the overall group of neuroradiologists and neurologists in predicting final infarct/ASPECTS.