Refined AI-ASPECTS with expanded atlas and optimized thresholds improves stroke outcome prediction over conventional AI

A refined AI-ASPECTS model using an expanded middle cerebral artery atlas and optimized lesion thresholds (2%–29%) improved correlation with NIHSS (r= -0.414) and functional outcome prediction (AUC 0.723) compared to conventional AI (AUC 0.654) and expert readers (AUC 0.641) on…

• The refined AI (Ref-AI-ASPECTS) demonstrated superior agreement with expert scores (ICC 0.81) versus conventional AI (ICC 0.67) on external validation DWI.
• Prospective real-world DWI data (AUC 0.710) and expanded CT data (AUC 0.696) supported generalizability of the refined framework.
• The study was limited by the need for further prospective external validation, as noted by the authors.

Automated summary

RadPigeon summaries are original and for information only. They are not clinical advice.