A distributed deep learning approach for blood sample-based early detection of dementia

Alzheimer’s Disease (AD), the prevailing form of dementia, is a neurological condition that significantly impacts individuals globally, leading to devastating effects. The early detection of AD is of paramount importance in mitigating its impact. Numerous methodologies have been suggested for diagnosing AD through medical imaging techniques such as positron emission tomography (PET) and magnetic resonance imaging (MRI). Nevertheless, it is anticipated that utilizing blood biomarkers would enhance the identification of individuals with AD and cognitive impairments. This paper introduces an innovative distributed deep-learning methodology for the early identification of AD through the analysis of blood samples. This study aims to investigate the application of federated learning (FL) in the analysis of blood samples to predict the likelihood of getting AD. Our study employed a dataset of many blood samples characterized by various features. A generative adversarial network (GAN) has been applied to regenerate data from original data to improve model generalization, increase diversity, and reduce overfitting. Our experimental results demonstrate that the proposed approach maintains high accuracy and provides better privacy. The accuracy, recall, specificity, and F1 score achieved were 85.1%, 75.5%, 93.8%, and 84.9% for original data and 89.8%, 87.8%, 91.3%, and 89.9% for regenerated data.