Data Sheet 1_Machine learning based on body composition radiomics for predicting early rec

A multicenter study by Yongjie Zhou, last updated in 2026, developed a machine learning model using preoperative CT body composition radiomics to predict early recurrence in colorectal cancer. The dataset includes 917 patients from three institutions, partitioned into training and external test sets. An 11-feature radiomics signature was identified, with Random Forest models achieving AUCs up to 0.807.

Use Cases

• Predicting early recurrence in colorectal cancer patients based on body composition radiomic features.
• Comparing the performance of eight different machine learning algorithms for a clinical prediction task.
• Stratifying patient risk for personalized postoperative surveillance using an individualized radiomics risk score.
• Interpreting model predictions using SHAP analysis to identify primary drivers like intermuscular adipose tissue features.

Strengths

• Multicenter design with 917 patients from three independent institutions.
• External validation performed on two separate test sets (n=263 and n=106).
• Model interpretability achieved using SHAP analysis, identifying key feature contributions.
• The optimal Random Forest model demonstrated AUCs of 0.807, 0.776, and 0.750 across cohorts.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• The dataset is small at 425.9 KB, indicating limited scope.

Provenance

Source

figshare

Collection Method

Data collected from a multicenter study of patients who underwent radical resection for colorectal cancer.

Freshness

Last updated 2026-06-04 04:27:25; freshness should be verified.