AttCORAL: Domain-Adaptive Attention Networks for Early Alzheimers Disease Diagnosis

Abstract

Alzheimer's disease (AD) is one of the most common neurodegenerative disorders characterized by the progressive accumulation of amyloid-beta plaques and tau protein tangles in the brain. Due to the lack of a cure for Alzheimer's disease, early and accurate diagnosis of AD is crucial for effective early interventions to slow disease progression. Magnetic Resonance Imaging (MRI) has emerged as a promising modality for early diagnosis, providing detailed insights into brain structure alterations associated with AD. However, domain shift due to variations in imaging protocols and data distribution among national cohorts remains a challenge for the application of MRI in clinical diagnosis. To address this issue, we propose AttCORAL, a novel Domain-Adaptive Attention Network for Early Alzheimer's Disease Diagnosis, integrating attention mechanisms with Correlation Alignment (CORAL) loss to effectively mitigate domain discrepancy, enhancing the model's robustness and generalization. We evaluate AttCORAL on two large-scale MRI datasets-Alzheimer's Disease Neuroimaging Initiative (ADNI) and the Australian Imaging, Biomarkers, and Lifestyle Study (AIBL)-which differ in acquisition protocols and demographics. While these datasets provide a valuable basis for cross-cohort validation, we acknowledge that further multi-cohort studies are necessary to fully assess global generalizability. To ensure the reliability of our approach, we apply Grad-CAM to visualize the pathological brain regions most informative for our model's predictions. Experimental results demonstrate that AttCORAL significantly outperforms current state-of-the-art studies, highlighting its effectiveness in early diagnosis of AD across diverse imaging domains.