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

Area of Science:

• Neuroimaging analysis
• Artificial intelligence in medicine
• Stroke diagnostics

Background:

• Growing interest in AI for neuroimaging evaluation in acute ischemic stroke.
• Need for improved accuracy and interobserver agreement in ASPECTS scoring.

Purpose of the Study:

• To evaluate the impact of Brainomix e-ASPECTS software on interobserver agreement and accuracy.
• To compare AI-assisted interpretation with standard interpretation for anterior circulation large vessel occlusion (LVO).

Main Methods:

• 60 patients with anterior circulation LVO and successful revascularization were included.
• 16 readers (senior/junior neuroradiologists, vascular neurologists) interpreted CT scans with and without e-ASPECTS assistance.
• Interclass correlation coefficient and Fleiss' Kappa were used to assess agreement; accuracy was determined against 24-hour CT.

Main Results:

• Interclass correlation for total ASPECTS improved from 0.395 (fair) to 0.574 (good) with e-ASPECTS (P < 0.01).
• Significant improvement in inter-rater agreement for most ASPECTS regions with AI assistance.
• e-ASPECTS demonstrated higher accuracy than readers without assistance for most regions.

Conclusions:

• Brainomix e-ASPECTS software significantly enhances inter-rater agreement and accuracy in ASPECTS score evaluation.
• AI software proved more predictive of final infarct/ASPECTS than human interpretation alone.