Document Type

Original Study


Artificial intelligence models are the most common methods of diagnosing various diseases today. In the meantime, the use of deep learning for faster diagnosis with less error, which is usually less expensive than other methods, is noteworthy. However, these methods are expensive due to the problems of accessing data, including care data such as hospitals and clinics. This article presents a novel federated learning-based approach for ultrasound mammography image classification. In this method, the clients possess their own datasets and are trained to use them. Then, the server, with the help of the trained clients, obtains the best parameters for its model. The most important innovation of this article is the use of a shared biomedical model in which the internal features of the models are shared among each other. The proposed method has an average of 9.99% higher accuracy compared to similar studies.