Development and Deployment of an Explainable Machine Learning Model for Preoperative Prediction of Thigh Liposuction Volume in Female Patients.

Abstract

Accurate preoperative estimation of liposuction volume is essential for achieving optimal aesthetic outcomes and minimizing surgical complications. However, conventional assessment methods are primarily subjective and rely heavily on the surgeon's experience, which may lead to inconsistent results. This study aimed to develop and externally validate a machine learning model to predict thigh liposuction volume in female patients using routine preoperative clinical and anthropometric data. A retrospective cohort of 518 female patients was analyzed, including 423 cases for model development and 95 for external validation. Eight key predictors were selected using the Boruta algorithm. A linear regression model with Elastic Net regularization was trained and compared with 5 other machine learning models. Model performance was evaluated by mean absolute error (MAE), root mean squared error (RMSE), mean absolute percentage error (MAPE), and Pearson correlation. Shapley Additive Explanations (SHAP) analysis was used for interpretability, and a web-based prediction tool was developed for clinical deployment. The final model demonstrated strong predictive performance (external MAE=349.96 mL; r=0.93, P<0.001) and maintained accuracy across Body Mass Index (BMI) categories. SHAP analysis identified body fat rate, skinfold thickness, and thigh circumference as top predictors. This explainable and externally validated model enables individualized, data-driven prediction of liposuction volume, supporting safer and more precise surgical planning in clinical practice.