Development and Validation of MRI Radiomics Model for Predicting Perineural Invasion in Rectal Cancer

Background This study aims to explore the application of multiparametric MRI (mp-MRI) based radiomics in evaluating the perineural invasion (PNI) status of rectal cancer. Methods A retrospective analysis was conducted on clinical and MRI data from 423 rectal cancer patients confirmed by surgical pathology across two centers. A total of 343 patients from Center 1 were split into a training set and an internal validation (in-vad) set in an 8:2 ratio, while 80 patients from Center 2 served as an independent external validation (ex-vad) set. Univariate and multivariate analyses were performed on clinical features to construct a clinical model. Radiomic features were extracted using Pyradiomics software, and features were selected and reduced using mRMR and LASSO methods to construct the radiomics model. A combined model integrating clinical and radiomics features was subsequently built, and a nomogram was developed. Results Among all patients, 131 cases (31.0%) were PNI-positive. Multivariate analysis identified mrT (OR = 1.038, P < 0.001) and mrN (OR = 1.038, P < 0.001) as independent predictors of PNI, forming the clinical model. After radiomic feature selection, 30 features were used to build the radiomics model. The area under the curve (AUC) values for the clinical model in the training, in-vad, and ex-vad sets were 0.719, 0.631, and 0.760, respectively. The AUC values for the radiomics model were 0.841, 0.815, and 0.916, while those for the combined model were 0.899, 0.826, and 0.914. The Delong test demonstrated that both the radiomics and combined models outperformed the clinical model across all datasets, with no statistically significant difference between the radiomics and combined models. Conclusions The mp-MRI based radiomics model effectively predicts PNI status in rectal cancer, providing a non-invasive and accurate method for preoperative evaluation.