Related Experiment Video
Updated: May 25, 2026
06:01
A Thrombotic Stroke Model Based On Transient Cerebral Hypoxia-ischemia
Published on: August 18, 2015
An Explainable Machine Learning Model Integrating Transcranial Doppler and Clinical Data to Predict Outcomes After
Xiaoqiong Chen1, Qun Huang1, Xiao Yang2
1Department of Ultrasound, Zhangzhou Municipal Hospital of Fujian Province, Zhangzhou Affiliated Hospital of Fujian Medical University, Zhangzhou, Fujian, China.
World Neurosurgery
|May 23, 2026
Summary
An interpretable machine learning model using transcranial Doppler (TCD) hemodynamics and clinical factors can predict functional outcomes after endovascular thrombectomy (EVT) for acute ischemic stroke. Key predictors include MFV_Index, PSV_Ratio, smoking status, and age, aiding in personalized treatment strategies.
Area of Science:
- Neurology
- Medical Imaging
- Machine Learning
Background:
- Endovascular thrombectomy (EVT) improves outcomes for acute ischemic stroke with large vessel occlusion (AIS-LVO).
- A significant number of patients still experience poor functional outcomes post-EVT, even with successful recanalization.
- Post-procedural cerebral hemodynamics are increasingly recognized as crucial for recovery after EVT.
Purpose of the Study:
- To develop an interpretable machine learning (ML) model for predicting 3-month functional outcomes after EVT.
- To integrate transcranial Doppler (TCD)-derived hemodynamic parameters with clinical features for enhanced prediction.
- To identify key hemodynamic and clinical factors influencing post-EVT recovery.
Main Methods:
- Retrospective cohort study of 176 AIS-LVO patients with successful recanalization (mTICI ≥ 2b) and post-EVT TCD monitoring.
- Bootstrap-based internal validation with 1,000 replicates, employing LASSO-penalized logistic regression for feature selection.
- Evaluation of five ML classifiers (logistic regression, SVM, k-NN, random forest, GBM) using ROC AUC, calibration, and decision curve analyses; interpretability assessed with SHAP.
Main Results:
- Smoking status, PSV_Ratio, age, and MFV_Index were consistently selected as top predictors across bootstrap replicates.
- The random forest model exhibited the best overall performance with good calibration and net benefit.
- SHAP analysis identified MFV_Index and PSV_Ratio as the most influential hemodynamic features, with smoking and age also being significant predictors of unfavorable outcomes.
Conclusions:
- An interpretable ML model integrating post-EVT TCD hemodynamics and clinical data demonstrates promising internal validation for predicting functional outcomes.
- MFV_Index, PSV_Ratio, smoking status, and age are identified as critical predictors.
- The findings underscore the importance of post-reperfusion cerebral hemodynamics and baseline vascular risk factors in predicting stroke recovery.
