Region-specific agreement in ASPECTS estimation between neuroradiologists and e-ASPECTS software.

Area of Science:

• Neuroradiology
• Medical Imaging Analysis
• Machine Learning in Medicine

Background:

• The Alberta Stroke Program Early CT Score (ASPECTS) is crucial for assessing ischemic changes in acute ischemic stroke.
• Current manual ASPECTS scoring faces limitations due to modest interobserver agreement.
• Machine learning (ML) offers a potential solution to improve the objectivity and reliability of ASPECTS assessment.

Purpose of the Study:

• To compare the performance of an ML-based software, e-ASPECTS, against manual scoring by experienced neuroradiologists.
• To evaluate the agreement between e-ASPECTS and manual ASPECTS scoring on a per-region basis in acute ischemic stroke patients.

Main Methods:

• Retrospective review of 178 non-contrast CT scans from acute ischemic stroke patients.
• Consensus manual scoring of ASPECTS by two independent neuroradiologists.
• Evaluation of all scans using the e-ASPECTS machine learning software.
• Calculation of interobserver agreement using kappa statistics and intraclass correlation coefficients.

Main Results:

• Overall agreement between manual ASPECTS and e-ASPECTS was fair (κ=0.248).
• Regional agreement varied significantly, with better performance in subcortical and cortical regions (e.g., lentiform, M3) compared to others (e.g., M1).
• Prevalence-adjusted bias-adjusted kappa indicated substantial agreement for several regions, particularly cortical areas.

Conclusions:

• Manual scoring and e-ASPECTS demonstrate fair agreement on a per-region basis for acute ischemic stroke assessment.
• Further investigation is warranted to validate e-ASPECTS performance, potentially using follow-up scans or MRI as a gold standard.