This research presents a proposal focused on the implementation of robustness techniques to strengthen convolutional neural networks (CNNs) in the detection of prostate cancer in the face of adverse threats, taking into account the serious public health situation in Peru. The vulnerabilities of deep learning models in the face of antagonistic examples are addressed, with the aim of improving robustness and accuracy in clinical diagnoses. The proposal is based on the perspective of robust optimization and seeks to develop methods to train and evaluate reliable and resilient neural networks to a wide variety of adversarial attacks. In a context where the reliability of medical diagnoses is a constant concern, this proposal seeks to contribute to the improvement of accuracy and safety in the detection of this disease in Peru, where medical negligence and lack of reliability in results, especially in magnetic resonance imaging (MRI) examinations, are significant challenges. Collaboration with medical experts and validation of clinical datasets will be critical to addressing this issue and providing more reliable diagnoses to patients in a public health environment affected by recurrent crises.