Federated Learning for Breast Cancer Classification

Abstract

Detecting breast cancer at an early stage and making precise diagnosis are vital in enhancing survival rates and treatment results. Technology, proven to help expedite processes and provide convenience, can also be integrated into breast cancer diagnosis to facilitate early treatment. In this work, the disease and integration of diagnosing mammograms and artificial intelligence (AI) were studied. Recent studies in this field that leveraged on deep learning to perform breast canscer classification were studied, outlining the common techniques and networks used. Public datasets were also analysed, identifying useful information that can enhance the model's performance. By implementing a baseline classification model themselves, the authors also explored using federated learning (FL) to decentralize training of the model and utilize data from different medical institutions. FL experiments were done using the previously studied publicly available datasets to compare the performance between centralized and FL environment, encouraging open-source research development in this field.