This study is a multicenter retrospective analysis that included a total of 106 patients diagnosed with pancreatic cancer, who underwent surgical treatment and were confirmed by postoperative pathological examination. The patients were enrolled from four medical institutions, spanning the period from January 2019 to December 2024. The patients were randomly assigned to a training set (n = 74) and a validation set (n = 32). Multiparametric MRI sequences, including T2-weighted imaging (T2WI), diffusion-weighted imaging (DWI), and dynamic contrast-enhanced (DCE) sequences, were used for radiomic feature extraction to build a traditional radiomics model. Contrast-enhanced T1-weighted imaging (CE-T1WI) was segmented into distinct subregions, or "habitats," using k-means clustering based on voxel intensity and entropy values to capture tumor heterogeneity. Radiomic features were subsequently extracted from these subregions to develop a habitat model. Feature selection was performed using the intraclass correlation coefficient (ICC) to ensure robustness and the least absolute shrinkage and selection operator (LASSO) algorithm to identify the most relevant features. Clinical and pathological features related to PNI, such as tumor size, CA19-9 levels, and tumor differentiation, were also included to establish a combined predictive model.