Radiomics analysis of restaging MRI for detection of pathological complete response in locally advanced rectal cancer.

Purpose: magnetic resonance imaging (MRI)-based radiomics has emerged as a promising approach for non-invasive prediction of treatment response in rectal cancer. This study aimed to develop and validate a machine learning model based on radiomic features extracted from restaging MRI after neoadjuvant therapy in patients with locally advanced rectal cancer (LARC), to identify those achieving pathological complete response (pCR). Methods: In this retrospective single-center study, patients with histologically confirmed rectal cancer treated between 2017 and 2022 were included if they underwent neoadjuvant therapy, staging and restaging MRI, and surgery. Tumor segmentation was performed on oblique axial T2-weighted images. Radiomic features were extracted using PyRadiomics following image resampling and intensity discretization. Feature selection was performed using the least absolute shrinkage and selection operator (LASSO), and a logistic regression model with Elastic Net regularization was trained. Model performance was evaluated using repeated stratified 5-fold cross-validation. Performance metrics included area under the receiver operating characteristic curve (AUC-ROC), accuracy, sensitivity, and specificity. Results: The final cohort included 86 patients (mean age 63 ± 12 years; 65% male) with 33 patients achieving pCR (38.37%) and 53 who did not (61.63%). The model demonstrated moderate discriminative performance, with a mean AUC-ROC of 0.74, accuracy of 67.4%, sensitivity of 60.1%, and specificity of 71.0%. Performance was higher for identifying non-responders compared with patients achieving pCR. Conclusion: MRI-based radiomics from restaging MRI shows potential to predict treatment response in LARC, particularly in identifying non-responders to neoadjuvant therapy. Although performance remains moderate, this approach may support treatment stratification and personalized management. Further validation in larger, multicenter cohorts is warranted.