Predict potential of acute stroke from OCTA images

This is a purely client-side web-application which can be used to predict the risk of acute stroke in a patient using non-invasive OCTA retinal images.

The following excerpt has been sourced from a paper written by Dr. Luca Giancardo et. al at the School of Biomedical Informatics, University of Texas Health Science Center at Houston (UTHealth) in collaboration with McGovern Medical School, UTHealth and School of Medicine and Public Health, University of Wisconsin-Madison, which was the inspiration behind creating this web-application.

Increased risk of cerebrovascular disease has been documented after prolonged exposure to ionizing radiation on Earth and recently on the International Space Station. During space missions, a precise diagnosis of acute ischemic stroke (AIS) is difficult due to the lack of brain imaging capabilities. Retinal imaging systems, on the other hand, have been deployed in space missions, and may serve as an alternative. In particular, optical coherence tomography angiogram (OCT-A) is a new imaging approach that enables the visualization of microvasculature in the retina, a central nervous system structure with direct connections to the brain. A retinal imaging system that could probe cerebrovascular pathophysiology would enable the prompt delivery of lifesaving treatment for the astronauts on the space mission.

How to use this application?

Please note that this application is still under development. As of now, I am working on calculating the vessel density of the selected regions.

Objective

Predict the potential of acute stroke using a patient's retinal images

Data

There are 4 OCTA images that need to be selected.

1. Superficial layer of the left eye
2. Deep layer of the left eye
3. Superficial layer of the right eye
4. Deep layer of the right eye

The ROSE dataset contains several of such images.

Steps

1. Select images (should not be more than 4)
2. Select regions of interest using the overlay tool. Look at the table below for feedback from the application.
3. Label the images into any of the four labels provided. Note- do not assign one label to multiple images.
4. Apply image processing algorithms to pre-process your data.
5. Run machine learning model on the data to and get the prediction.