A Non-invasive Deep Learning Model for Prostate Cancer Diagnosis with MRI

Prostate cancer (PCa) remains a significant global health concern, with accurate Gleason grade assessment crucial for guiding treatment. Traditional histopathology relies on invasive biopsy and subjective evaluation. This study proposes a novel non-invasive approach for Gleason grade prediction using magnetic resonance imaging (MRI) and deep learning (DL). Our method comprises three main steps: tumor region segmentation using Fuzzy c-means (FCM); relevant feature extraction via modified residual network (ResNet50) model; and Gleason grade classification using a convolutional neural network (CNN). Through extensive experimentation, the proposed CNN model achieved an accuracy of 92.00%, sensitivity of 92.00%, specificity of 92.00%, and an area under the curve receiver operating characteristic (AUC-ROC) of 0.95. These robust results highlight the potential of our DL framework to accurately differentiate between low-grade and high-grade PCa, thereby automating aspects of the diagnostic process, reducing reliance on subjective interpretation, and ultimately improving patient outcomes and treatment decisions.