Alzheimer-OCT Dataset and Detection of Alzheimer's Disease using Deep Learning

This study introduces a new clinical dataset for the detection of Alzheimer's disease (AD), the Alzheimer-OCT (AOCT) Set, and presents deep learning-based analyses. The dataset includes optical coherence tomography (OCT), angiography (OCTA) and fundus images from 31 volunteer participants. To overcome the limited data constraint, transfer learning strategies were employed; 2D slice-based deep learning architectures were compared with the 3D PointNet++ approach, which directly models spatial relationships by converting OCT volumes into point clouds. The results show that the developed 2D and 3D models are able to distinguish retinal structural changes associated with AD and that clinical OCT data may be a potential biomarker for AD diagnosis.