Deep learning with guided attention for early diagnosis of Alzheimer's disease

Abstract

Alzheimer's Disease (AD) is one of the most common forms of neurodegenerative disease that involves the accumulation of amyloid beta plaques and tau tangles. The early diagnosis of AD is crucial as it helps patients to start preventive interventions to slow the disease's progression. We created a Guided-Attention Feature Extraction Deep Learning Network (GADL) for the early diagnosis of Alzheimer's disease (AD). We applied a GADL for the prediction of mild cognitive impairment (MCI) progression to AD and classification between MCI and cognitively normal (CN). We trained the model with magnetic resonance imaging images in the Alzheimer's Disease Neuroimaging Initiative (ADNI) database by subject-level data splitting and verified its generalizability in the Australian Imaging Biomarkers and Lifestyle Flagship Study of Aging (AIBL) database. Our method outperformed other subject-level studies with an accuracy of 80.29% for the prediction of MCI progression to AD and 83.70% for CN versus MCI classification in the ADNI dataset. The accuracies of our models when they were applied to the AIBL dataset are recorded as 79.38% and 79.83%, respectively. These results prove the high performance of our models in terms of its generalizability. The evaluation results showed that the proposed approach has competitive performance in comparison with recent studies in terms of its performance and generalizability. These results suggest that deep learning with guided attention can be an effective early diagnosis technique and a prognostic tool for Alzheimer's disease.