Early Diagnosis of Alzheimer's Disease: A Multi-class Deep Learning Framework with Modified k-sparse Autoencoder Classification

Successful, timely diagnosis of neuropsychiatric diseases is key to management. Research efforts in the area of diagnosis of Alzheimer's disease have used various aspects of computer-aided multi-class diagnosis approaches with varied degrees of success. However, there is still need for more efficient and reliable approaches to successful diagnosis of the disease. This research used deep learning framework with modified k-sparse autoencoder (sigma KSA) classification to locate neutrally degenerated areas of the brain magnetic resonance imaging (MRI), low amyloid beta 1-42 imaging in cerebrospinal fluid (CSF) and positron emission tomography (PET) imaging of amyloid; each with a sample of 150 images. Results show a correlation between computational demarcation of infected regions and the images. Degeneration in the studied areas was evidenced by high phosphorylated t-/p-tau levels in CSF, regional hypometabolism fluorodeoxyglucose PET, and the presence of atrophy patterns. The use of sigma KSA algorithm in boosting classification helped to improve the classifier performance. The KSA method with deep learning framework is used for the first time to produce accurate results in diagnosis of Alzheimer's disease.