Rectal Cancer MRI Radiomics and Clinical Data for T-Staging

A multicenter study of 313 rectal cancer patients used MRI-derived tumor microenvironment habitats and peritumoral radiomic features to differentiate T1-2 from T3 stages. The research developed a nomogram integrating habitat features, optimal 2mm peritumoral radiomics (PERI2mm), and clinical predictors like age and CEA. The combined model achieved an AUC of 0.881 on an external test cohort of 130 patients.

Use Cases

• Train a classifier using MRI-derived habitat clusters and PERI2mm peritumoral radiomic features to predict pathological T-stage.
• Analyze the correlation between clinical predictors, including age and elevated CEA, and radiomic habitat features for staging accuracy.
• Validate model calibration using Brier scores and intercept metrics reported for the training and test cohorts.
• Apply decision curve analysis (DCA) on the nomogram's predicted probabilities to assess clinical net benefit across risk thresholds.

Strengths

• Multicenter design with 313 total patient records, split into training (183) and external test (130) cohorts for robust validation.
• Model performance is quantified with multiple metrics, including a test AUC of 0.881 and a test F1 score of 0.851 for the final nomogram.
• Radiomic analysis includes specific features: three intratumoral habitat clusters and peritumoral features extracted at 1mm, 2mm, and 3mm extensions.

Limitations

• Dataset is small in file size (125.5 KB), contained within a DOCX document, limiting direct computational analysis of raw features.
• Retrospective study design may introduce selection bias, and the patient sample size of 313 may limit generalizability.
• The raw columnar data and sample records are unavailable, preventing immediate feature engineering or model replication.

Provenance

Source

figshare, author Liang Zhang.

Collection Method

Retrospective multicenter study analyzing patient MRI scans and clinical records.

Time Range

null

Freshness

Last updated March 2026.

Geography

Multicenter study, specific locations unknown.