Individualized Diagnosis of Preclinical Alzheimer's Disease using Deep Neural Networks

Abstract

The early diagnosis of Alzheimer's Disease (AD) plays a central role in the treatment of AD. Particularly, identifying the preclinical AD (pAD) stage could be crucial for timely treatment in the elderly. However, screening participants with pAD requires a series of psychological and neurological examinations. Thus, an efficient diagnostic tool is needed. Here, we recruited 91 elderly participants and collected 1 minute of resting-state electroencephalography data to classify participants as normal aging or diagnosed with pAD. We used deep neural networks (Deep ConvNet, EEGNet, EEG-TCNet, and cascade CRNN) in the within-and cross-subject paradigms for classification and found individual variations of classification accuracy in the cross-subject paradigm. Further, we proposed an individualized diagnostic strategy to identify neurophysiological similarities across participants and the proposed approach considering individual characteristics improved the diagnostic performance by approximately 20%. Our findings suggest that considering individual characteristics would be a breakthrough in diagnosing AD using deep neural networks.