Related Experiment Video
Updated: Jun 22, 2025
07:21
Systems Analysis of the Neuroinflammatory and Hemodynamic Response to Traumatic Brain Injury
Published on: May 27, 2022
3.1K
Interpretable Machine Learning Models Based on Shapley Additive Explanations for Predicting the Risk of Cerebrospinal
ZongJie Guo1, PeiYang Wang1, SuHui Ye2
1Spine Surgery Center, Department of Spine Surgery, Zhongda Hospital Affiliated to Southeast University, Nanjing, Jiangsu, People's Republic of China.
Spine
|July 4, 2024
Summary
This study developed an interpretable machine learning (ML) model using XGBoost and SHAP to predict cerebrospinal fluid leakage (CSFL) after lumbar fusion surgery, identifying key risk factors for improved patient outcomes.
Area of Science:
- Neurosurgery and Medical Informatics
- Development and validation of predictive models in healthcare
- Application of machine learning in surgical risk assessment
Background:
- Increasing incidence of spinal degeneration necessitates accurate prediction of postoperative complications.
- Cerebrospinal fluid leakage (CSFL) is a significant complication following lumbar fusion surgery.
- Lack of interpretable machine learning models for CSFL risk factor analysis in lumbar fusion.
Purpose of the Study:
- To develop and validate a customized machine learning (ML) framework for forecasting cerebrospinal fluid leakage (CSFL) in lumbar fusion surgery.
- To integrate imaging parameters and utilize the SHapley Additive exPlanations (SHAP) technique for model interpretability.
- To identify significant predictors of CSFL in patients undergoing lumbar fusion.
Main Methods:
- Retrospective collection of clinical and imaging data from 3505 patients undergoing lumbar fusion.
- Formulation and evaluation of six distinct machine learning models: XGBoost, DT, RF, SVM, GaussianNB, and KNN.
- Performance assessment using standard metrics and analysis via the SHAP framework for interpretability.
Main Results:
- Cerebrospinal fluid leakage (CSFL) occurred in 2.71% of patients.
- The XGBoost model demonstrated high accuracy in predicting CSFL (AUC: 0.7343).
- SHAP analysis identified key predictors: ligamentum flavum thickness, zygapophysial joint degeneration, spinal stenosis grade, and surgical parameters.
Conclusions:
- The XGBoost-SHAP model is effective for predicting CSFL risk in lumbar fusion.
- Implementation can aid clinical decision-making, potentially improving patient outcomes and reducing costs.
- Advocates for adopting this interpretable ML approach in clinical settings for enhanced CSFL risk evaluation.
