Related Experiment Video
Updated: Jun 23, 2026
07:13
Comparison of Predictive Performance of Three Lymph Node Staging Systems in Colorectal Signet Ring Cell Carcinoma Based on Machine Learning Model
Published on: April 18, 2025
Machine learning based on body composition radiomics for predicting early recurrence in colorectal cancer: a
Yongjie Zhou1, Miaoping Zhou2, Yongming Tan3
1Department of Radiology, Jiangxi Cancer Hospital & Institute, Jiangxi Clinical Research Center for Cancer, The Second Affiliated Hospital of Nanchang Medical College, Nanchang, China.
Frontiers in Nutrition
|June 22, 2026
Summary
An interpretable machine learning model using CT body composition radiomics can predict early recurrence in colorectal cancer (CRC). This tool improves risk stratification and patient surveillance beyond current staging methods.
Area of Science:
- Radiology
- Oncology
- Machine Learning
Background:
- Early recurrence (ER) in colorectal cancer (CRC) significantly impacts patient outcomes.
- Current pTNM staging lacks the ability to assess the host's systemic pathophysiological status.
- Predicting ER is crucial for effective patient management and treatment strategies.
Purpose of the Study:
- To develop and validate an interpretable machine learning (ML) model for predicting ER in CRC patients.
- To utilize preoperative CT body composition radiomics as a basis for the predictive model.
- To enhance risk stratification for colorectal cancer patients.
Main Methods:
- A multicenter study involving 917 CRC patients undergoing radical resection.
- Extraction of 1,896 radiomic features from skeletal muscle (SM), subcutaneous adipose tissue (SAT), intermuscular adipose tissue (IMAT), and visceral adipose tissue (VAT) at the L3 level on CT.
- Feature selection using LASSO and Boruta algorithms, evaluation of eight ML algorithms, and integration with clinical factors. SHAP analysis for interpretability.
Main Results:
- An 11-feature radiomics signature was identified, with the Random Forest model achieving AUCs of 0.807, 0.776, and 0.750 in training and test sets.
- SHAP analysis highlighted IMAT (46.1%) and SM (42.9%) features as key predictors, with SM textural uniformity indicating potential adverse muscle quality.
- The radiomics risk score effectively stratified patients for recurrence-free and overall survival (p < 0.05) and demonstrated superior clinical benefit over pTNM staging.
Conclusions:
- An interpretable ML model based on CT body composition radiomics shows promise for predicting ER in CRC.
- This model offers a quantitative tool for improved risk stratification and individualized postoperative surveillance in colorectal cancer patients.
