AI-driven Method Developed to Diagnose Neurodegenerative Diseases

A combination of OCT, adaptive optics, and neural networks has the potential to enable better diagnosis and monitoring for eye and brain diseases, such as glaucoma, that damage neurons. The combination is part of an AI-drive process developed by biomedical engineers at Duke University as leaders of a multi-institution consortium that can easily and precisely track the number and shape of retinal ganglion cells in the eye.

Ganglion cells are one of the primary neurons in the eye that process and send visual information to the brain. In many neurodegenerative diseases, such as glaucoma, ganglion cells degenerate and disappear, which causes irreversible blindness. OCT is a light-based imaging technology that can image beneath layers of eye tissue to diagnose and track the progression of such diseases. However, the technique is only sensitive enough to show the thickness of the cell layer — it lacks the sensitivity necessary to reveal individual ganglion cells. This hinders early diagnosis or the ability to rapidly track disease progression, as large quantities of ganglion cells need to disappear before physicians can see the changes in thickness.

In their work, the collaborating researchers used adaptive optics OCT (AO-OCT) to image with the sensitivity required to view individual ganglion cells. As an independent technology, adaptive optics minimizes the effect of optical aberrations that occur when examining the eye. These aberrations are a major limiting factor in the ability to achieve high-resolution in OCT imaging.

An image generated by AO-OCT (top), and the result of WeakGCSeg algorithms to identify and trace the shapes of the ganglion cells in the eye (bottom). Courtesy of Duke Biomedical Engineering.

Though AO-OCT makes it easier to diagnose neurodegenerative disease, however, the higher resolution generates an amount of data that create a bottleneck to image analysis using the method, said Sina Farsiu, a professor of biomedical engineering at Duke.

A solution to the problem introduced in the researchers’ new paper uses a highly adaptive and easy-to-train deep learning-based algorithm to identify and trace the shape of ganglion cells from AO-OCT scans.

The developed algorithm, the researchers said, is the first to perform this kind of identification and tracing. The researchers incorporated the algorithm in an approach, WeakGCSeg, to analyze AO-OCT data from retinas of healthy subjects as well as those with glaucoma. The framework efficiently and accurately segmented ganglion cells from both samples and identified which samples came from the glaucomatous eyes based on the number and size of ganglion cells present.

"Our experimental results showed that WeakGCSeg is actually superior to human experts, and it's superior to other state-of-the-art networks that can process volumetric biomedical images," said Soltanian-Zadeh, a postdoctoral researcher in Farsiu’s lab. In addition to diagnostics, the team believes the approach will make it easier to conduct clinical trials of therapies for neurodegenerative diseases.

In a study to test for a therapy for glaucoma, for example, WEAKGCSeg can be used to see if the therapy has slowed down cell degeneration compared to the control group used in the study. With OCT alone, the first sign of any change would require hundreds or thousands of cells dying, which can take months or years.

"With our technique, you would be able to quantify the earliest change," Farsiu said.

A next step for the method involves applying it to a larger cohort of patients, with collaborators from the FDA, Indiana University, and the University of Maryland. The team also hopes to extend WeakGSeg to different cell types, such as photoreceptors, and additional diseases of the eye, such as retinitis pigmentosa. As studies have shown that changes in the ganglion cell layer are associated with diseases of the central nervous system such as Alzheimer’s disease, Parkinson’s disease, and ALS, Farsiu said, the developed technique can both further study this connection and potentially discover biomarkers for improved diagnosis and treatment.

The research was published in Optica (www.doi.org/10.1364/OPTICA.418274).