CNN-Based Deep Learning for MRI-Assisted Classification of Benign and Malignant Prostate Nodules: A Diagnostic Performance Study

Background: Prostate cancer (PCa) is one of the most common malignancies among men worldwide, and accurate differentiation between benign and malignant nodules remains challenging. Magnetic resonance imaging (MRI) provides valuable soft-tissue contrast but still suffers from interpretive variability. Deep learning–based computer-aided diagnostic (CAD) systems may help improve diagnostic accuracy and consistency. Therefore, this study aimed to evaluate a convolutional neural network (CNN)–based deep learning system for computer-aided classification of benign versus malignant prostate nodules using multiparametric MRI (mpMRI). Methods: In this retrospective study, 80 patients with histopathologically confirmed prostate nodules (39 malignant, 41 benign) were enrolled between January 2021 and June 2024. Univariate and multivariate analyses were conducted to identify key imaging risk factors associated with malignancy. The diagnostic performance of conventional MRI was compared with that of five CNN-assisted models (CNN-1 to CNN-5) based on sensitivity, specificity, and overall accuracy. Results: Significant differences (<i>p</i> < 0.001) were observed between malignant and benign groups in imaging features, including signal distribution (χ² = 31.473), lesion margins (χ² = 19.776), lesion volume (<i>t</i> = 19.421), short-axis diameter (<i>t</i> = 10.337), long-axis diameter (<i>t</i> = 9.071), and the product of diameters (<i>t</i> = 6.548). Multivariate logistic regression identified these parameters as independent malignancy predictors, with odds ratios (ORs) ranging from 2.818 to 3.277 across training and validation cohorts. These variables were incorporated into a malignancy risk score model. Among all CNN models, CNN-3, characterized by three max-pooling layers, achieved the highest diagnostic sensitivity and accuracy. Receiver operating characteristic (ROC) analysis further confirmed its superior performance, demonstrating the largest area under the curve (AUC) and outperforming both conventional MRI and other CNN variants. Conclusion: The CNN-based deep learning diagnostic system significantly enhances the classification accuracy of prostate nodules on mpMRI. The CNN-3 model enables automated lesion detection and feature extraction, improving early diagnosis and risk stratification. It shows strong potential to support clinical decision-making in prostate cancer management.