Federated GAN based biomedical image augmentation and classification for Alzheimer’s disease

Abstract

Federated Learning (FL) is a distributed machine learning approach that can de velop a global or customized model from scattered datasets on edge devices using federated datasets. ‘Federated GAN Based Biomedical Image Augmentation and Classification for Alzheimer’s Disease’ will focus on augmenting the medical images using Federated Generative Adversarial Network. Due to patient-doctor confiden tiality, the scarcity of data in the medical sector is a massive hindrance to the advancement of machine learning models in this sector. Our study aims to augment the existing medical data, in this case, magnetic resonance imaging(MRI) images of the brain, and test that augmented dataset on existing classification models to eval uate our generated MRI images’ quality. Generative Adversarial Networks (GANs) have been utilized in order to synthesize realistic and varied Alzheimer’s disease affected MRI images in order to cover the data shortage in the actual medical image distribution and identify Alzheimer’s disease using Federated Learning. We expect our proposed model to successfully augment the medical images and be over 90% accurate at detecting the medical condition.