Abstract

Alzheimer’s Disease (AD) is a form of dementia which causes memory, thinking, and behavior disorders in humans. Effective early diagnosis and treatment of AD is of fundamental importance as it can reduce disease progression, allow more effective management of symptoms, facilitate timely patient access to advice and support, and lower associated costs of health care. Given that Alzheimer’s typically progresses in stages over an extended period of time, we propose that automated analysis of time sequential data may enhance disease prediction. We present a novel time-series Alzheimer’s Disease Prediction Model (ADPM) comprising Random Forest (RF) region of
interest (ROI) selection and Gated Recurrent Units (GRU) prediction. Experiments show that our methodology achieves higher classification accuracy in comparison to existing algorithms, and can facilitate prediction of early onset AD. Furthermore, testing demonstrates that random forest ROI selection can identify disease-relative brain regions across different image modalities (MRI, PET, DTI).