Development and validation of an interpretable machine learning model integrating baseline multimodal CT perfusion and clinical data for predicting 9-month functional outcomes in acute ischemic stroke

 

Predicting failure to recover is the sign of PURE INCOMPETENCE!

You are supposed to deliver 100% recovery protocols! If you don't know that; get the hell out of stroke!  I'd fire anyone doing this type of crapola research!

Development and validation of an interpretable machine learning model integrating baseline multimodal CT perfusion and clinical data for predicting 9-month functional outcomes in acute ischemic stroke

• G

  Guizhi Zhang ^1,2

• Y

  Yuanli Zhang ^1,2

• X

  Xin Chen ³

• B

  Bo Han ³

• X

  Xiang Li ⁴

• Y

  Yang Wang ^1,2

• 

  Bo Jiang ^3*

• 1. Department of Medical Technology, Chongqing Three Gorges Medical College, Chongqing, China

• 2. Research Centre of Oral Materials and Technology, Chongqing Three Gorges Medical College, Chongqing, China

Abstract

Background and purpose: 

Accurate early prediction of long-term functional outcomes in acute ischemic stroke (AIS) remains challenging. We aimed to develop and validate an interpretable machine learning model integrating baseline multimodal CT perfusion and clinical data to predict 9-month poor functional outcome [modified Rankin Scale (mRS) 3–6] in AIS patients undergoing endovascular or surgical intervention.

Methods: 

This retrospective study included 371 AIS patients who underwent endovascular or surgical intervention. We integrated clinical variables, laboratory markers, CT perfusion parameters (Tmax, cerebral blood flow, and cerebral blood volume), and quantitative CT density measurements [Hounsfield units on the affected and healthy sides, normalized water uptake (NWU), Alberta Stroke Program Early CT Score (ASPECT score)]. Feature selection was performed using LASSO regression followed by SHAP-based refinement. Four algorithms—Logistic Regression, Random Forest, Support Vector Machine, and XGBoost—were compared. Model performance was evaluated using area under the receiver operating characteristic curve (AUC), sensitivity, specificity, F1-score, Brier score, calibration curves, and decision curve analysis (DCA). SHAP analysis was used to enhance model interpretability.

Results: 

Eleven predictors were selected by LASSO; after SHAP-based refinement, ten remained in the final model. The XGBoost model achieved the best performance, with an AUC of 0.956 (95% CI 0.917–0.995), sensitivity of 0.932, specificity of 0.897, F1-score of 0.938, and Brier score of 0.0713 (95% CI 0.0438–0.1135). SHAP analysis identified door-to-intervention interval (DTI) (mean SHAP = 0.95) as the most influential predictor, followed by HU_affected (0.88), CBV < 42% (0.82), NLR (0.78), blood glucose (0.74), age (0.70), gender (0.68), ASPECT score (0.66), HU_healthy (0.50), and NWU (0.48). Calibration curves demonstrated excellent agreement between predicted and observed outcomes, and DCA confirmed superior net benefit of the XGBoost model across clinically relevant thresholds (0.01–0.60).

Conclusion: 

We developed an interpretable XGBoost-based model integrating baseline multimodal CT perfusion and clinical data that accurately predicts 9-month functional outcomes in AIS patients undergoing endovascular or surgical intervention. DTI, HU_affected, and CBV < 42% emerged as the dominant predictors, highlighting the prognostic importance of time-to-reperfusion and baseline tissue injury. After external validation, this model may serve as a practical tool for early risk stratification and individualized rehabilitation planning.