DICE-Net: A Novel Convolution-Transformer Architecture for Alzheimer Detection in EEG Signals

Objective: Alzheimer’s disease (AD) is a progressive neurodegenerative disorder that affects a significant percentage of the elderly. EEG has emerged as a promising tool for the timely diagnosis and classification of AD or other dementia types. This paper proposes a novel approach to AD EEG classification using a Dual-Input Convolution Encoder Network (DICE-net). Approach: Recordings of 36 AD, 23 Frontotemporal dementia (FTD), and 29 age-matched healthy individuals (CN) were used. After denoising, Band power and Coherence features were extracted and fed to DICE-net, which consists of Convolution, Transformer Encoder, and Feed-Forward layers. Main results: Our results show that DICE-net achieved an accuracy of 83.28% in the AD-CN problem using Leave-One-Subject-Out validation, outperforming several baseline models, and achieving good generalization performance. Significance: Our findings suggest that a convolution transformer network can effectively capture the complex features of EEG signals for the classification of AD patients versus control subjects and may be expanded to other types of dementia, such as FTD. This approach could improve the accuracy of early diagnosis and lead to the development of more effective interventions for AD.