Deep Learning for Prostate Cancer Diagnosis Using Computed Tomography: A comprehensive Review of CNN Architectures, Clinical Applications, Challenges, and Future Directions

Abstract

Prostate cancer is one of the most common malignancies and a leading cause of cancer-related death among men worldwide. Although magnetic resonance imaging and prostate-specific membrane antigen positron emission tomography/computed tomography are central to prostate cancer diagnosis and staging, computed tomography remains widely accessible and important for staging and radiotherapy planning, particularly in resource-limited settings. This review critically examines convolutional neural network applications in prostate computed tomography imaging and compares them with evidence from magnetic resonance imaging, positron emission tomography/computed tomography, and ultrasound. We conducted a structured narrative review of relevant literature published between 2021 and 2026, supplemented by seminal studies on established convolutional neural network architectures. The review identified a clear imbalance in current applications: computed tomography research predominantly focuses on automated segmentation of the prostate and surrounding organs for radiotherapy planning and on lesion segmentation in positron emission tomography/computed tomography, whereas direct diagnostic classification remains comparatively limited. Reported segmentation performance is generally strong, whereas diagnostic classification studies are smaller and more heterogeneous. Explainable artificial intelligence methods may improve model interpretability, although their reliability remains variable. Future progress requires larger multicentre datasets linked to histopathology, standardized performance reporting, rigorous external validation, and the integration of computed tomography-based models as adjuncts rather than replacements for established diagnostic pathways.