Multiparametric MRI radiomics for preoperative prediction of tumor budding grade in rectal cancer
4 affiliations
- 1Department of Radiology, Huzhou Central Hospital, Fifth School of Clinical Medicine of Zhejiang Chinese Medical University
- 2Department of Radiology, Changxing Hospital of Traditional Chinese Medicine
- 3Department of Pathology, Huzhou Central Hospital, Fifth School of Clinical Medicine of Zhejiang Chinese Medical University
- 4School of Medicine and Nursing, Huzhou Normal University
Abstract
Background: Tumor budding (TB) is a histopathological marker of tumor aggressiveness and an established prognostic factor in colorectal cancer. In rectal cancer, however, TB is usually assessed on biopsy or resection material, and a reliable result may not be available when preoperative management is being considered.
Purpose: To develop and internally validate a radiomics model for preoperative discrimination of low-grade versus intermediate/high-grade TB in patients with rectal cancer who proceed to upfront surgery without neoadjuvant treatment. The primary model was selected according to the highest repeated cross-validation AUC, with alternative feature sets and algorithms evaluated exploratorily.
Methods: This retrospective study included 139 patients with pathologically confirmed rectal cancer who had received no neoadjuvant treatment. TB was assessed on postoperative resection specimens. T1WI, T2WI, arterial-phase contrast-enhanced T1WI, and DWI were analyzed. Radiomics features were extracted from the tumor core, a combined tumor-plus-2-mm expansion, and an isolated external 2-mm ring. Seven feature sets and four machine-learning algorithms were evaluated; the primary model was selected according to the highest repeated cross-validation AUC. Feature filtering, imputation, selection, and scaling were performed within the training portion of every split. Model performance was estimated using stratified five-fold cross-validation repeated five times and patient-level out-of-fold (OOF) probabilities.
Results: The primary core-ROI Elastic Net model achieved a repeated cross-validation AUC of 0.825 (range 0.809–0.839 across repetitions), with PR-AUC 0.781, sensitivity 0.604, specificity 0.851, and Brier score 0.173. Its patient-level mean OOF AUC was 0.829 (2,000 paired bootstrap replicates; 95% CI, 0.755–0.896). At a probability threshold of 0.50, patient-level sensitivity was 0.607 and specificity was 0.867. In conditional paired bootstrap comparisons, the core model showed higher AUCs than the clinical, peritumoral-ring, clinical-plus-expanded, clinical-plus-core, and clinical-plus-core-plus-ring models (BH-adjusted P = 0.004–0.005). The difference versus the expanded ROI was smaller and did not reach statistical significance after adjustment (P = 0.052).
Conclusion: Multiparametric MRI radiomics features from the tumor core showed promising internal discrimination of low-grade versus intermediate/high-grade TB among patients who underwent upfront surgery without neoadjuvant treatment. The model may support preoperative TB risk stratification in this population, but external validation is required before clinical use.
Related
- Multiparametric MRI radiomics model for preoperative prediction of IDH1 R132H immunopositivity status in adult diffuse Glioma
- Multiparametric MRI-based habitat and peritumoral radiomics integrating semi-automated segmentation for preoperative prediction of high-grade prostate cancer: a dual-center study
- Multiparametric MRI radiomics and clinical features for preoperative prediction of high-grade meningioma: a two-centre retrospective study
- Multiparametric MRI radiomics combined with clinical-radiologic features improves preoperative prediction of muscle invasion in bladder cancer
- Multiparametric MRI-based radiomics combined with deep learning for preoperative risk stratification of endometrial carcinoma
BibTeX
@article{chen2026multiparametric,
title = {{Multiparametric MRI radiomics for preoperative prediction of tumor budding grade in rectal cancer}},
author = {Peng Chen and Yanyan Zhou and Dan Wu and Chengmeng Zhang and Xiangyi He and Ru Pan and Dong Xia and Jinge Li and Jue Hou and Jian Shen and Hao Chen},
journal = {Frontiers in Oncology},
year = {2026},
month = oct,
doi = {10.3389/fonc.2026.1984365},
url = {https://www.frontiersin.org/articles/10.3389/fonc.2026.1984365/full},
note = {CC BY 4.0},
}
RIS
TY - JOUR
TI - Multiparametric MRI radiomics for preoperative prediction of tumor budding grade in rectal cancer
AU - Peng Chen
AU - Yanyan Zhou
AU - Dan Wu
AU - Chengmeng Zhang
AU - Xiangyi He
AU - Ru Pan
AU - Dong Xia
AU - Jinge Li
AU - Jue Hou
AU - Jian Shen
AU - Hao Chen
PY - 2026
DA - 2026/10/14
JO - Frontiers in Oncology
DO - 10.3389/fonc.2026.1984365
UR - https://www.frontiersin.org/articles/10.3389/fonc.2026.1984365/full
AB - Background: Tumor budding (TB) is a histopathological marker of tumor aggressiveness and an established prognostic factor in colorectal cancer. In rectal cancer, however, TB is usually assessed on biopsy or resection material, and a reliable result may not be available when preoperative management is being considered. Purpose: To develop and internally validate a radiomics model for preoperative discrimination of low-grade versus intermediate/high-grade TB in patients with rectal cancer who proceed to upfront surgery without neoadjuvant treatment. The primary model was selected according to the highest repeated cross-validation AUC, with alternative feature sets and algorithms evaluated exploratorily. Methods: This retrospective study included 139 patients with pathologically confirmed rectal cancer who had received no neoadjuvant treatment. TB was assessed on postoperative resection specimens. T1WI, T2WI, arterial-phase contrast-enhanced T1WI, and DWI were analyzed. Radiomics features were extracted from the tumor core, a combined tumor-plus-2-mm expansion, and an isolated external 2-mm ring. Seven feature sets and four machine-learning algorithms were evaluated; the primary model was selected according to the highest repeated cross-validation AUC. Feature filtering, imputation, selection, and scaling were performed within the training portion of every split. Model performance was estimated using stratified five-fold cross-validation repeated five times and patient-level out-of-fold (OOF) probabilities. Results: The primary core-ROI Elastic Net model achieved a repeated cross-validation AUC of 0.825 (range 0.809–0.839 across repetitions), with PR-AUC 0.781, sensitivity 0.604, specificity 0.851, and Brier score 0.173. Its patient-level mean OOF AUC was 0.829 (2,000 paired bootstrap replicates; 95% CI, 0.755–0.896). At a probability threshold of 0.50, patient-level sensitivity was 0.607 and specificity was 0.867. In conditional paired bootstrap comparisons, the core model showed higher AUCs than the clinical, peritumoral-ring, clinical-plus-expanded, clinical-plus-core, and clinical-plus-core-plus-ring models (BH-adjusted P = 0.004–0.005). The difference versus the expanded ROI was smaller and did not reach statistical significance after adjustment (P = 0.052). Conclusion: Multiparametric MRI radiomics features from the tumor core showed promising internal discrimination of low-grade versus intermediate/high-grade TB among patients who underwent upfront surgery without neoadjuvant treatment. The model may support preoperative TB risk stratification in this population, but external validation is required before clinical use.
SN - 2234-943X
ER -