MRI-based radiomics for preoperative T-staging of rectal cancer: a retrospective analysis

Abstract

Puropose: Preoperative T-staging in rectal cancer is essential for treatment planning, yet conventional MRI shows limited accuracy (~ 60-78). Our study investigates whether radiomic analysis of high-resolution T2-weighted MRI can non-invasively improve staging accuracy through a retrospective evaluation in a real-world surgical cohort.

Methods: This single-center retrospective study included 200 patients (January 2024-April 2025) with pathologically confirmed rectal cancer, all undergoing preoperative high-resolution T2-weighted MRI within one week prior to curative surgery and no neoadjuvant therapy. Manual segmentation was performed using ITK‑SNAP, followed by extraction of 107 radiomic features via PyRadiomics. Feature selection employed mRMR and LASSO logistic regression, culminating in a Rad-score predictive model. Statistical performance was evaluated using ROC curves (AUC), accuracy, sensitivity, specificity, and Delong's test.

Results: Among 200 patients, 95 were pathologically staged as T2 and 105 as T3-T4 (55 T3, 50 T4). After preprocessing, 26 radiomic features were retained; key features including ngtdm_contrast and ngtdm_coarseness showed AUC values > 0.70. The LASSO-based model achieved an AUC of 0.82 (95% CI: 0.75-0.89), with overall accuracy of 81%, sensitivity of 78%, and specificity of 84%.

Conclusion: Radiomic analysis of standard preoperative T2-weighted MRI provides a reliable, non-invasive method to predict rectal cancer T-stage. This approach has the potential to enhance staging accuracy and inform personalized surgical planning. Prospective multicenter validation is required for broader clinical implementation.

Keywords: Artificial Intelligence; MRI; Oncologic Imaging; Radiomics; Rectal cancer.

Fig. 1

Workflow diagram of the study. The process begins with eligibility assessment of 200 patients, followed by manual segmentation of T2-weighted MRI scans using ITK-SNAP, feature extraction with PyRadiomics, and feature selection via mRMR and LASSO regression. The resulting radiomic signature (Rad-score) was then evaluated using ROC analysis and compared to expert radiologist assessment

Fig. 2

Heatmap illustrating the association between individual radiomic features and pathological T stage, measured by the AUC of three-class ROC curves (T2, T3, T4). Color intensity reflects the discriminative power of each feature (AUC range: 0–1). Features such as ngtdm_contrast, ngtdm_strength, ngtdm_coarseness, glszm_grayLevelVariance, and glszm_sizeZoneNonUniformity exhibit strong predictive ability (AUC > 0.70)

Fig. 3

Barplot showing the normalized coefficients of radiomic features selected by LASSO regression for T-stage classification. Features with positive coefficients (in red) are associated with advanced tumors (T3–T4), while those with negative coefficients (in blue) are associated with early-stage tumors (T2). Features are ordered by their contribution to the predictive model

Fig. 4

Boxplot of the radiomic feature ngtdm_coarseness across pathological T stages (T2, T3, T4). The distribution highlights statistically significant differences between T2 and T4 (p < 0.01) and a near-significant trend between T2 and T3, based on Wilcoxon rank-sum test. This suggests that ngtdm_coarseness is a relevant marker of tumor invasiveness and may support non-invasive T-stage differentiation

Fig. 5

Comparison of ROC curves for the radiomic model (blue), clinical assessment by expert radiologists (orange), and the combined model integrating both (green). The radiomic model alone achieved an AUC of 0.82, outperforming clinical assessment (AUC 0.69). The combined model demonstrated further improvement, reaching an AUC of 0.85

Fig. 6

Confusion matrix summarizing the classification performance of the radiomic model in distinguishing T2 from T3–T4 rectal tumors. Values represent the number of true positives, true negatives, false positives, and false negatives, along with per-class accuracy. The model achieved a global accuracy of 88%, with sensitivity and specificity of 90% and 94%, respectively