JoVE - Journal of Visualized Experiments
JoVE Visualize
Contact Us

LightGBM is an Effective Predictive Model for Postoperative Complications in Gastric Cancer: A Study Integrating Radiomics with Ensemble Learning

  • 0Department of Anesthesiology and Perioperative Medicine, The First Affiliated Hospital of Nanjing Medical University, Nanjing, 210029, China.
Journal of Imaging Informatics in Medicine +

|

June 28, 2024

|

June 28, 2024

View abstract on PubMed

Summary

This summary is machine-generated.

Accurate prediction of postoperative complications after gastric cancer surgery is crucial. A new model using visceral fat radiomics and machine learning, specifically Light Gradient Boosting Machine (LGBM), shows high accuracy in predicting these complications.

Area Of Science

  • Oncology
  • Medical Imaging
  • Machine Learning

Background

  • Postoperative complications following radical gastrectomy significantly impede patient recovery.
  • Accurate risk prediction is essential for effective perioperative management in gastric cancer patients.

Purpose Of The Study

  • To develop and validate a predictive model for early postoperative complications in gastric cancer patients undergoing radical gastrectomy.
  • To guide perioperative clinical decision-making and improve patient outcomes.

Main Methods

  • Retrospective analysis of 166 patients who underwent radical gastrectomy.
  • Utilized 3D Convolutional Neural Networks (3D-CNN) for visceral fat segmentation and feature extraction from CT scans.
  • Employed LASSO regression for feature selection and ensemble learning with Light Gradient Boosting Machine (LGBM) for prediction model training.
  • Evaluated model performance using fivefold cross-validation.

Main Results

  • The developed LGBM model achieved an Area Under the Curve (AUC) of 0.9232.
  • The model demonstrated a high accuracy of 87.28% (95% CI, 75.61-98.95%) in predicting postoperative complications.
  • The LGBM model outperformed other evaluated machine learning algorithms (XGBoost, Random Forest, GBDT).

Conclusions

  • An effective predictive model for early postoperative complications after radical gastrectomy was successfully constructed using ensemble learning and visceral fat radiomics.
  • The LGBM model offers a promising tool for individualized clinical decision-making, potentially enhancing the early recovery of gastric cancer patients post-surgery.

Keywords: