Early detection and staging via Magnetic Resonance Imaging (MRI) are essential for the optimal treatment of prostate cancer, the second most prevalent cancer in men. In this context, the critical step of accurate segmentation can greatly benefit from AI-based algorithms. In this work, we developed AI-based models for the segmentation of prostate zones and lesions. We developed two distinct networks, a U-Net, and a Res-U-Net model, leveraging data from 4 publicly available databases comprising MRIs for prostate gland segmentation, 4 datasets for zonal segmentation, and 2 datasets for prostate lesion segmentation. These included T2-Weighted (T2W) and Apparent Diffusion Coefficients (ADC) sequences. For overall prostate gland segmentation, the U-Net model’s performance reached an Intersection over Union (IoU) value of 87% and a Dice Similarity Coefficient (DSC) of 93%. In zonal segmentation, it achieved IoU scores of 83% for the peripheral zone (PZ) and 92% for the central gland (CG), with DSC scores of 91% for both PZ and CG. In lesion segmentation, the model achieved an IoU of 85% and a DSC of 93%.