BACKGROUND: Prostate cancer (PCa) is the most common malignant tumor that endangers the life and health of middle-aged and elderly men. PURPOSE: To evaluate the significance of radiomic features from intralesional and perilesional regions in bi-parametric magnetic resonance imaging (MRI) for diagnosing PCa. MATERIAL AND METHODS: A total of 211 patients with suspected PCa who accepted prostate MRI scans were enrolled in this study. The region of interest (ROI) corresponding to the original lesion was manually delineated to define the intralesional ROI on bp-MRI maps. The original lesion ROI was then expanded by 2 mm, 4 mm, 6 mm, and 8 mm, while excluding the intralesional area to create the perilesional ROI. Features were extracted from each ROI, and a radiomics model was developed using logistic regression. The combined model integrated features from both intralesional and perilesional regions. Its predictive performance was assessed using receiver operating characteristic (ROC) curves and area under the curve (AUC) to evaluate its diagnostic efficacy for PCa. RESULTS: The comparison revealed that perilesional 4 mm model had the best performance among all perilesional models, its AUCs of 0.934 and 0.894 in the training testing set, respectively, outperformed the combined model of other regions. The clinical model, combined model for intralesional regions, and INTRAPERI model achieved AUCs of 0.911, 0.925, 0.931 in the training sets and 0.770, 0.867, 0.905 in the testing sets. The predictive performance of the INTRAPERI model is better than the clinical model and intralesional model. CONCLUSION: The radiomic model combining intralesional and perilesional features from bi-parametric MRI shows strong predictive value for PCa and may enhance clinical decision-making.