Cerebrovascular CTA radiomics for objective collateral grading in acute ischemic stroke

Comprehensive Summary

In this study, Rallios et al. developed an artificial intelligence tool that uses CT angiography (CTA) scans to automatically grade collateral circulation in patients with acute ischemic stroke caused by large vessel occlusion. Collateral circulation is important because it can help keep brain tissues alive after the vessel is occluded. It is also important in the determination of whether the patient can benefit from thrombectomy. The authors trained their model on CTA scans from 343 patients in the MR CLEAN trial and tested it on both an internal test set of 69 patients and an external set of 140 patients. Their pipeline first used deep learning to segment the brain’s blood vessels, then extracted radiomic features from those vessels, and finally used a random forest model to classify collateral status as sufficient or insufficient. This vessel-based model performed better than the traditional middle cerebral artery mask-based model. The AUROC of this model is 0.88 compared to 0.82 in internal testing and 0.83 compared to 0.66 in external testing. The addition of the circle of Willis features improved the performance of the model to 0.87 AUROC. The results indicate that AI can help in the more objective assessment of the patient's collateral status through standard CTA imaging.

Outcomes and Implications

This paper is significant in that it demonstrates the potential for AI to assist in speeding up decision-making in stroke imaging. In acute stroke management, collateral grading is one of the factors to be considered in decision-making. Manual grading is subjective in nature, and there is a significant variation from individual to individual. The potential for this tool to be useful in decision-making is significant, especially in acute cases, since it uses machine learning and radiomics from CTA scans. It can be particularly useful in borderline cases of thrombectomy or in later stages, when it is critical to know whether collateral blood flow is good or poor. However, this study is in its early stages, looking retrospectively at cases with proximal large vessel occlusions, and while they suggest that further prospective studies are needed to determine whether this tool can improve clinical workflows, decision-making, and patient outcomes, this tool has significant potential in acute stroke management in the near future.