Alzheimer’s disease classification from hippocampal atrophy based on PCANet-BLS

Alzheimer’s Disease (AD) is a neurodegenerative disease of unknown etiology that progresses progressively and is currently incurable. It is more common in the elderly that seriously affects the physical and mental health of patients, thus early detection is very important for the prevention of AD progression. By using PCANet and Broad Learning System (BLS), we propose a novel method to identify Alzheimer’s patients according to the clinical symptom of hippocampal atrophy, which is the most important indicator of AD. T1-weighted magnetic resonance images (MRIs) are used in this study, containing 207 patients with AD, 209 patients with mild cognitive impairment (MCI) and 109 cognitively normal (CN) cohorts from ADNI dataset. The left and right hippocampus are segmented from MRI at the first step, then the PACNet is applied to extract features from these images, finally the BLS is used to distinguish the different types of patients. Compared with the traditional machine learning methods, PCANet is able to extract the most informative features inside pictures effectively, while BLS can reach over 95% accuracy rate with lower time consuming. Experimental results have revealed that our method improves the performance of computer-aided diagnosis of Alzheimer’s disease in both accuracy and speed of classification task.