Related Experiment Video
Updated: Jul 4, 2026
14:56
An Experimental Paradigm for the Prediction of Post-Operative Pain (PPOP)
Published on: January 28, 2010
Developing and validating an interpretable machine learning model for predicting postoperative nausea and vomiting in
Zhensheng Huang1, Shu Wang1, Yahong Liu2
1Department of Anesthesiology, State Key Laboratory of Oncology in South China, Guangdong Provincial Clinical Research Center for Cancer, Sun Yat- sen University Cancer Center, Guangzhou, 510060, China.
BMC Anesthesiology
|July 3, 2026
Summary
This study developed a machine learning model using brain MRI and clinical data to predict postoperative nausea and vomiting (PONV) after lung surgery. The model shows improved accuracy over the Apfel score for personalized risk assessment.
Area of Science:
- Neuroimaging
- Machine Learning
- Surgical Complications
Background:
- Postoperative nausea and vomiting (PONV) is a common complication after lung surgery, impeding Enhanced Recovery After Surgery (ERAS) protocols.
- Existing clinical risk scores, like the Apfel score, have limited precision for individual PONV risk prediction.
- Neuroanatomical biomarkers have not been integrated into current PONV risk assessment models.
Purpose of the Study:
- To develop and validate an interpretable machine learning (ML) model for predicting PONV risk after pulmonary resection.
- To integrate structural brain Magnetic Resonance Imaging (MRI) features with clinical variables for enhanced prediction accuracy.
- To identify novel neuroanatomical predictors of PONV.
Main Methods:
- Retrospective analysis of 416 patients undergoing pulmonary resection.
- Extraction of 1,174 structural brain features from T1-weighted MRI using FreeSurfer.
- Application of feature engineering, including LASSO selection, and training/validation of eleven ML models.
Main Results:
- The optimal ExtraTrees classifier model achieved an Area Under the Receiver Operating Characteristic Curve (AUROC) of 0.82 (internal) and 0.80 (external validation).
- The ML model significantly outperformed the Apfel score in PONV prediction.
- Key predictors included female sex, reduced left parieto-occipital cortical curvature, and decreased right occipital pole curvature/cuneus folding index.
Conclusions:
- Integrating neuroimaging biomarkers with clinical data significantly improves PONV prediction accuracy after lung surgery.
- Identified neuroanatomical features in visual and vestibular regions offer insights into PONV susceptibility.
- The interpretable ML model provides a tool for personalized PONV risk stratification within ERAS pathways.
