Enhancing Breast Cancer Detection in Federated Learning by Memorizing at Test Time

Being aware of the fact that we have to save ourselves since nobody cherishes oneself more than the other, it is motivating to make a difference in the medical field. Breast cancer has been rated as the fourth cause of mortality among women especially in the developing countries. However, data sharing restrictions and privacy issues make the identification of cases in the initial stages doable to enhance the survival rates. This applies FL with DP to improve the medical decision-making trust while maintaining patient privacy and uses XAI to explain the model’s decisions. This model can be closed by such technical solutions like adaptive differential privacy, decentralized feature extraction with the Google’s Titans’ architecture, as well as interpretability techniques such as attention-based visualization and LIME. This work aims at fulfilling the role of defining an accurate and socially useful AI system that can contribute to the improvement of breast cancer diagnoses and at the same time illustrating a general guide concerning AI and its use in increasing the effectiveness of the healthcare sector.