A T2-weighted MRI-based nomogram integrating radiomics, deep learning, and clinical factors for preoperative prediction of tumor deposits in T3–T4 rectal cancer: a dual-center retrospective study

Background Tumor deposits (TDs) are adverse pathological features in rectal cancer, but their preoperative identification on conventional MRI remains challenging. This study aimed to develop and validate a practical T2-weighted MRI-based nomogram for predicting TDs positivity in patients with T3–T4 rectal cancer. Methods This retrospective dual-center study included 405 patients with T3–T4 rectal adenocarcinoma who underwent preoperative oblique axial T2-weighted imaging. Patients from institution A formed the development cohort (n = 289; TDs-positive, n = 72), and those from institution B formed the independent testing cohort (n = 116; TDs-positive, n = 30). Radiomics features were selected using least absolute shrinkage and selection operator (LASSO) logistic regression, and a radiomics model based on a support vector machine and a deep learning model based on a Vision Transformer were developed from T2WI. A combined nomogram was constructed by integrating the radiomics risk score, deep learning risk score, and CA19-9, the only independent clinical predictor. All model development procedures, including feature selection, hyperparameter tuning, and threshold determination, were performed exclusively within the development cohort, and the independent testing cohort was used only once for final evaluation. Discrimination, calibration, and clinical utility were evaluated. Results CA19–9 was the only independent clinical risk factor for TDs positivity and was incorporated with the radiomics and deep learning risk scores into the nomogram. The nomogram achieved AUCs of 0.884 (95% CI, 0.841–0.918) in the development cohort and 0.854 (95% CI, 0.776–0.913) in the independent testing cohort, which were higher than those of the clinical, radiomics, and deep learning models, although the difference between the nomogram and the radiomics model was not statistically significant in the testing cohort (P = 0.230). In the independent testing cohort, sensitivity, specificity, and negative predictive value were 76.7% (95% CI, 59.1%–88.2%), 83.7% (95% CI, 74.5%–90.0%), and 91.1% (95% CI, 82.8%–95.6%), respectively. Calibration curves showed good agreement, and decision curve analysis demonstrated favorable clinical utility. Conclusions In this preliminary dual-center study with a single external validation cohort, a practical T2WI-based nomogram integrating radiomics features, Vision Transformer-derived deep learning features, and CA19–9 may assist preoperative prediction of TDs positivity and individualized risk stratification in T3–T4 rectal cancer, pending prospective multicenter validation.

Authors

Institutions

Publication Details

Journal
Frontiers in Oncology
Published
2026-11-17
DOI
https://doi.org/10.3389/fonc.2026.1974120
Primary Topic
Radiomics and Machine Learning in Medical Imaging
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

A T2-weighted MRI-based nomogram integrating radiomics, deep learning, and clinical factors for preoperative prediction of tumor deposits in T3–T4 rectal cancer: a dual-center retrospective study

黄子星, Kai Liao, Wei Chen, Ke Li et al.
3 links
Frontiers in Oncology
Radiomics and Machine Learning in Medical Imaging
article

A T2-weighted MRI-based nomogram integrating radiomics, deep learning, and clinical factors for preoperative prediction of tumor deposits in T3–T4 rectal cancer: a dual-center retrospective study

黄子星, Kai Liao, Wei Chen, Ke Li, Wei Zhang, Jian Guan, Dan Liu
article en

Abstract

Background Tumor deposits (TDs) are adverse pathological features in rectal cancer, but their preoperative identification on conventional MRI remains challenging. This study aimed to develop and validate a practical T2-weighted MRI-based nomogram for predicting TDs positivity in patients with T3–T4 rectal cancer. Methods This retrospective dual-center study included 405 patients with T3–T4 rectal adenocarcinoma who underwent preoperative oblique axial T2-weighted imaging. Patients from institution A formed the development cohort (n = 289; TDs-positive, n = 72), and those from institution B formed the independent testing cohort (n = 116; TDs-positive, n = 30). Radiomics features were selected using least absolute shrinkage and selection operator (LASSO) logistic regression, and a radiomics model based on a support vector machine and a deep learning model based on a Vision Transformer were developed from T2WI. A combined nomogram was constructed by integrating the radiomics risk score, deep learning risk score, and CA19-9, the only independent clinical predictor. All model development procedures, including feature selection, hyperparameter tuning, and threshold determination, were performed exclusively within the development cohort, and the independent testing cohort was used only once for final evaluation. Discrimination, calibration, and clinical utility were evaluated. Results CA19–9 was the only independent clinical risk factor for TDs positivity and was incorporated with the radiomics and deep learning risk scores into the nomogram. The nomogram achieved AUCs of 0.884 (95% CI, 0.841–0.918) in the development cohort and 0.854 (95% CI, 0.776–0.913) in the independent testing cohort, which were higher than those of the clinical, radiomics, and deep learning models, although the difference between the nomogram and the radiomics model was not statistically significant in the testing cohort (P = 0.230). In the independent testing cohort, sensitivity, specificity, and negative predictive value were 76.7% (95% CI, 59.1%–88.2%), 83.7% (95% CI, 74.5%–90.0%), and 91.1% (95% CI, 82.8%–95.6%), respectively. Calibration curves showed good agreement, and decision curve analysis demonstrated favorable clinical utility. Conclusions In this preliminary dual-center study with a single external validation cohort, a practical T2WI-based nomogram integrating radiomics features, Vision Transformer-derived deep learning features, and CA19–9 may assist preoperative prediction of TDs positivity and individualized risk stratification in T3–T4 rectal cancer, pending prospective multicenter validation.

Frontiers in OncologyVol. 16
Army Medical University (CN), Sichuan University (CN), West China Hospital of Sichuan University (CN), Southwest Hospital (CN), Chinese People's Armed Police Force (CN)
Good health and well-being
Radiomics and Machine Learning in Medical Imaging
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon