Prognostic estimation for acute ischemic stroke patients undergoing mechanical thrombectomy within an extended...

Area of Science:

• Neurology
• Medical Imaging
• Machine Learning in Medicine

Background:

• Mechanical thrombectomy (MT) is crucial for acute ischemic stroke with large vessel occlusion (AIS-LVO), even in extended therapeutic windows.
• However, predicting patient outcomes post-MT remains challenging, with only 40-50% achieving favorable results.
• Machine learning (ML) shows potential for outcome prediction but often lacks interpretability.

Purpose of the Study:

• To develop and select an optimal ML model for predicting outcomes in extended-window MT for AIS-LVO.
• To enhance the interpretability of the chosen model using the Shapley additive explanation (SHAP) approach.

Main Methods:

• Retrospective analysis of 260 AIS-LVO patients undergoing extended-window MT.
• Development and validation of four ML classifiers and one logistic regression model using pre-treatment variables.
• Comparative validation and testing of models, with SHAP analysis for feature interpretation.

Main Results:

• Extreme gradient boosting (XGBoost) demonstrated superior predictive performance (AUC 0.93 validation, 0.77 testing).
• SHAP analysis identified key predictors: ischemic core volume, baseline NIHSS score, ischemic penumbra volume, ASPECTS, and age.
• The XGBoost model proved effective in predicting prognosis for the studied patient cohort.

Conclusions:

• XGBoost is the most effective model for predicting prognosis in extended-window MT for AIS-LVO.
• SHAP interpretation increases clinical confidence in ML models for stroke outcome prediction.
• This study supports the integration of interpretable ML into clinical decision-making for stroke management.