A Federated Learning-based Solution for Pneumonia Diagnosis in Remote and Low-Income Areas

The digital transformation in healthcare is a recurring theme around the world. However, the continental extension of Brazil makes it difficult to offer basic healthcare services in remote locations. Due to the scarcity of diagnosis services in these areas, this article proposes leveraging Federated Learning as a way to reduce costs and mitigate these problems, helping with pre-diagnosis. The experiments carried out showed that just using the federated approach can increase the model's predictive capacity and reduce training time. The model developed using Federated Learning increased the model's accuracy by 14%, while managing to reduce the model's loss in the validation set by 1.0354.