CT perfusion plus clinical machine-learning model predicts 9-month disability after acute ischemic stroke

An interpretable XGBoost model using baseline multimodal CT perfusion and clinical data predicted 9-month poor functional outcome (modified Rankin Scale 3-6) after stroke intervention with AUC 0.956 (95% CI 0.917-0.995) in 371 patients.

• Retrospective single-center cohort of 371 acute ischemic stroke patients treated with endovascular or surgical intervention; model integrated CT perfusion parameters, quantitative CT density, laboratory markers, and clinical variables.
• XGBoost achieved AUC 0.956 (95% CI 0.917-0.995), sensitivity 0.932, specificity 0.897, F1 0.938, and Brier score 0.0713; calibration and decision curve analysis favored this model.
• SHAP analysis ranked door-to-intervention interval, affected-side Hounsfield units, and cerebral blood volume below 42% as the strongest predictors; external validation is required before clinical use.

Automated summary

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