The Study of Intelligent Scoring Tools for Acute Posterior Circulation Ischemic Stroke.
Authors
Abstract
Background and purpose Posterior Circulation Acute Stroke Prognosis Early Computed Tomography Scores (PC-ASPECTS) is crucial for diagnosing, treating, and predicting the prognosis of acute ischemic stroke in patients with posterior circulation involvement. However, physicians take longer to score patients' pc-ASPECTS and inter-rater reliability is low among different physicians. To address this issue, we developed an intelligent scoring model using artificial intelligence technology to enhance the accuracy and consistency of these scores. Methods Retrospective clinical and imaging data from multiple stroke centers were used to train and validate a convolutional neural network (CNN)-based model. The model identified early ischemic changes in predefined posterior circulation regions. Performance was evaluated using standard metrics (e.g., AUC, sensitivity, specificity) and compared to manual scoring by clinicians. Results A total of 674 patients with complete data were included in the study, 536 patients (mean age, 56 years ± 12 [SD]; 298 [55.6%] female) were included for model development (training: 300; validation: 129; and internal test set: 107). Another 138 patients (mean age, 59 years ± 14; 90 [65.2%] female) were included in an external test set to evaluate model's performance and generalizability. The PC-ASPECTS intelligent scoring model demonstrated strong discriminative ability across all regions (AUC range: 0.687-0.805). It significantly improved inter-rater consistency (kappa: 0.317 to 0.711) and reduced scoring time compared to clinicians (2-5 seconds vs. 25-90 seconds, p< 0.05). Conclusions The PC-ASPECTS intelligent scoring model developed in this study demonstrated commendable performance. Utilizing this prediction model, the consistency of PC-ASPECTS scoring among clinical physicians was improved and efficiency was significantly enhanced.