Radiomics and anaesthetic planning: Quantitative imaging as a new frontier in preoperative risk assessment

Table 5 Machine learning algorithms for radiomics-based perioperative risk prediction

Algorithm	Mechanism	Strengths	Perioperative application examples	Typical AUC range
Logistic regression	Linear decision boundary for binary classification	Interpretable; established statistical framework	Surgical site infection prediction; transfusion requirement	0.70-0.82
Random forest	Ensemble of multiple decision trees	Handles high-dimensional data; resistant to overfitting	Sarcopenia detection; postoperative complication prediction	0.80-0.90
Support vector machine	Optimal hyperplane separation in feature space	Effective in high-dimensional spaces; robust with small samples	Myocardial pathology classification; plaque vulnerability	0.78-0.90
LightGBM/XGBoost	Gradient boosting ensemble methods	High accuracy; fast training; handles missing data	Postoperative gastric cancer complications; pancreatic fistula	0.84-0.93
Deep neural networks	Multi-layer non-linear feature learning	Automatic feature extraction; captures complex patterns	Airway difficulty prediction; cardiac event risk	0.80-0.95
Convolutional neural networks	Spatial feature learning from image data	Direct image input; no manual feature engineering	Airway segmentation; organ volumetry; body composition	0.85-0.96

Area under the curve ranges are indicative values compiled across heterogeneous studies and do not represent single validated estimates. AUC: Area under the curve; LightGBM: Light gradient boosting machine; XGBoost: Extreme Gradient Boosting.