Executive Summary : | Alzheimer's disease (AD) is a common neurodegenerative dementia that affects around 44 million people worldwide, with the number expected to triple by 2050 due to an aging population. The disease progresses through three stages: pre-symptomatic, prodromal, and clinical. In AD patients, blood flow is reduced due to reduced retinal vein diameter, sparse vessels, and reduced fractal chambers. Additionally, patients with AD show a difference in the thickness of their choroid compared to healthy controls. Spectral-domain OCT has shown significant thinning of the choroid in AD patients, highlighting the importance of vascular factors in AD pathogenesis.
A low-cost, user-friendly computer-aided system for early detection of AD using smartphone camera-captured fundus images and a hybrid neural network can be developed and tested. The system will collaborate with medical colleges and hospitals with an Ophthalmology department, obtaining approval from an ethical committee. The system will use transfer learning for feature extraction, apply transformation operations to generate additional samples, and analyze the option of reducing images to eliminate redundant information. To avoid problems with non-uniformly sampled data in LSTM networks, the system will be uniformly subsampled. The system will be evaluated against other state-of-the-art models such as FTNN, VGG, and MobileNet V2 with LSTM. |