A Secure and Collaborative Approach to Breast Tumor Detection Using Federated Learning and CNNs

Breast cancer is a major health problem affecting many women worldwide, causing a multitude of deaths, and benefits from early identification for successful treatment. In this study, an original federated learning system combined with Convolutional Neural Networks (CNNs) is outlined to solve the challenge of data privacy in joint medical research concerning inter-institutional breast tumour recognition. Obtaining an accuracy of 95.4%, a precision of 94.8%, a recall of 96.1%, and an F1-score of 95.4%, the federated model outperformed the baseline centralized model's accuracy of 94.7%. This proposed system makes it possible for a wide variety of institutions to partner on the training of a global model that keeps patient data secure, thus noticeably lessening privacy issues and ensuring conformity with standards such as GDPR and HIPAA. The federated model demonstrated its efficacy by dealing with data heterogeneity, ensuring higher performance across datasets from diverse institutions reflecting varying picture quality and demographics. The ROC Area Under the Curve (AUC) of the federated model was 0.97 in opposition to 0.95 for the centralized model, demonstrating better discriminatory precision. This study draws attention to the opportunities for federated learning in health applications, which require strict data privacy, as well as a wide array of accurate, high-quality datasets to foster effective diagnostic models. The next initiatives will seek to improve communication efficiency and investigate fresh CNN designs to boost performance. This study presents a privacy-preserving method that can be scaled up for breast tumour identification, which in turn promotes enhanced multi-institutional relationships within healthcare.