Alzheimer's disease (AD) is the most common form of dementia. It is an irreversible, progressive brain disorder marked by a decline in cognitive functioning with no treatment. It is characterized by a massive decrease in brain size due to the accumulation of proteins (amyloid-beta and tau ) in the neurons. Eyes extend the brain as both the retina and brain grow from the same neural tube. Postmortem studies in AD also highlighted the collection of these proteins in the retina. More recently, high-resolution visual imaging techniques, including optical coherence tomography (OCT), have been proposed as tools for evaluating structural changes in the retina of AD patients. It is
Conventional diagnostic methods from medical images greatly depend on physicians' professional experience and knowledge. Artificial intelligence (AI) has improved the performance of many challenging tasks when working with high-resolution, complex imaging data. Artificial neural networks are a subset of AI inspired by a simplification of neurons and their connections in the brain. Deep learning (DL) is a multi-layer structure of neural networks that mimics human learning by analyzing data with a given logical structure.
This project is planned for three years, focusing on using deep learning-based analysis of retinal OCT scans for AD detection. Even though this technique is widely used to detect many other retinal diseases from OCT images, there is no application in AD.
Retinal scans obtained from OCTS devices are two-dimensional and three-dimensional images. Despite their high performance, DL architectures are black-box models. Trusting their predictions is an important factor in using them for decision-making in medicine. Therefore, this research aims to train the model with retinal images and develop algorithms that will help clinicians to review and visualize the decision process. The power of explainability tools can also help to highlight relevant patterns and even discover new ones.
It is not easy to collect sufficient, high-quality, and uniformly annotated data to build high-accuracy models. Research studies exclude old patients with multiple retinal illnesses. Therefore, only a minimal subset of collected data could be used in the studies. We also plan to investigate various learning methods to increase learning with small AD datasets by transferring knowledge from other studies and datasets.
This research aims to create an efficient and explainable model that will learn to classify stages of Alzheimer's disease from the currently used retinal scans.
.jpg)