MR Texture Analysis and Machine Learning for Differentiating Benign and Malignant Prostate Lesions

Purpose: To evaluate the diagnostic performance of magnetic resonance (MR) texture analysis combined with machine learning algorithms in distinguishing benign from malignant prostate lesions, based on histopathological confirmation following fusion-guided biopsy. Methods: In this retrospective single-center study, 70 patients who underwent prostate MRI and subsequent fusion biopsy between January and June 2023 were included. Texture features were extracted from T2-weighted, diffusion-weighted (b=1400), and apparent diffusion coefficient (ADC) images using LIFEx software. Feature selection was performed using LASSO regression, and predictive modeling was carried out via logistic regression and artificial neural networks. Model performance was evaluated using 10-fold cross-validation and reported in terms of AUC, sensitivity, specificity, and accuracy. Results: Logistic regression and neural network models demonstrated the highest diagnostic performance with ADC-derived features (AUC: 0.931 and 0.903; accuracy: %94 and %88, respectively). DWI features achieved AUC values of 0.944 and 0.861, with accuracies of %82 and %76. T2-weighted features yielded AUC values of 0.681 and 0.667; accuracy: %65 and %53. Conclusion: MR texture analysis, particularly from ADC and diffusion-weighted sequences, combined with machine learning algorithms, demonstrates strong diagnostic performance in differentiating benign from malignant prostate lesions. These approaches may enhance non-invasive diagnostic accuracy and support clinical decision-making in prostate cancer evaluation.