Imaging-based outcome prediction in posterior circulation stroke.

Area of Science:

• Neurology
• Radiology
• Artificial Intelligence

Background:

• Posterior circulation (pc)-stroke presents unique challenges for early outcome prediction.
• Conventional methods like pc-ASPECTS may not fully capture prognostic information from CT imaging.
• Early and accurate prediction is crucial for timely intervention and patient management.

Purpose of the Study:

• To develop and evaluate a machine learning (ML) model for early functional outcome prediction in pc-stroke patients.
• To utilize quantitative imaging features from automated pc-ASPECTS assessments.
• To compare the ML model's predictive power against conventional pc-ASPECTS ratings and clinical data.

Main Methods:

• Retrospective analysis of CT scans and clinical data from 172 pc-stroke patients.
• Outcome assessment at 90 days using modified Rankin Scale (mRS) dichotomized at various cut-offs (mRS 2, 3, 4, and survival mRS ≤ 5).
• Random forest algorithms were employed for prediction, with models evaluated using nested fivefold cross-validation.

Main Results:

• Machine learning models significantly outperformed conventional pc-ASPECTS in predicting functional outcomes.
• Imaging-based ML classifiers achieved ROC-AUCs up to 0.87 for survival prediction (mRS ≤ 5).
• The combined clinical data and ML model demonstrated the highest predictive performance, with ROC-AUCs reaching 0.90 for mRS ≤ 2.

Conclusions:

• ML-based evaluation of pc-ASPECTS regions offers superior accuracy in predicting functional outcomes for pc-stroke patients compared to conventional methods.
• This enhanced predictive capability can optimize patient triage, diagnostic workups, and allocation of medical resources.
• Early prediction facilitates timely arrangements for social support and post-discharge care.