Integrating MR and Ultrasound Images for AI-Enhanced Prostate Cancer Detection: A Comparative Study with Clinicians

This study presents a novel deep learning framework designed to enhance prostate cancer detection using transrectal ultrasound (TRUS) images alone, by integrating information from MRI images during model training. The multimodal deep neural network was pre-trained on MRI data to identify clinically significant prostate cancer and then used to train an unimodal TRUS-only network. In tests with 102 patients, the multimodal-guided TRUS model achieved a sensitivity of 80% and specificity of 70%, significantly outperforming an unguided baseline model (54% sensitivity, 48% specificity) and expert clinicians (35% sensitivity, 92% specificity). These findings suggest that this approach could improve prostate cancer biopsy diagnoses.