CTSI Pilot Awardee Aims to Create Smarter, More Accessible AI for Detecting Eye Diseases

By: Joslyn Cassano  |  Published: April 22, 2022

Delia Cabrera DeBuc, Ph.D.

Artificial intelligence (AI) is changing the way ophthalmic conditions are diagnosed. Miami CTSI pilot awardee Delia Cabrera DeBuc, Ph.D., wants to make sure that AI is capturing the whole picture for every patient.

“We’re trying to see if we can develop an algorithm that allows us to accurately detect diseases of the eye regardless of a patient’s race, ethnicity, or gender,” said Cabrera DeBuc, a research associate professor at University of Miami’s Bascom Palmer Eye Institute.

Over her 18-year career at Bascom Palmer, Dr. Cabrera DeBuc’s work has focused primarily on the study of diabetic retinopathy, a diabetes-related condition that can lead to blindness.

With emergence of Artificial Intelligence (AI) in diagnosis and treatment of conditions like diabetic retinopathy, Dr. Cabrera DeBuc’s work has turned to ensuring AI delivers unbiased results based on a diverse dataset.

AI is an important tool for the future of healthcare. According to Dr. Cabrera DeBuc, advanced diagnostic technology has the potential to address physician shortages and provide healthcare information to patients in hard-to-reach areas. Using AI, primary care physicians will be able to provide diagnoses that currently require a visit to a specialist.

However, AI is only as “intelligent” as humans teach it to be through algorithms. Currently, AI-based ophthalmic diagnostics are operating with racial, ethnic, and gender bias, says Dr. Cabrera DeBuc.

“There is a critical gap in the knowledge base because the role of demographic bias is rarely discussed and has been relatively unexplored,” she says. “If you are training your algorithm from a predominantly Caucasian population, when you try to classify an image from a patient who is African American or Hispanic, the algorithm is not going to be able to identify properly that particular disease, because you didn’t train the algorithm with balanced ground truth.”

With support through a CTSI pilot award in 2020, Dr. Cabrera DeBuc and her team are collecting large amounts of data that will help better train AI to recognize and diagnose ophthalmic conditions in a diverse patient population.

The development of AI unbiased algorithms could mitigate the disparity of accuracy across subpopulations . “Your algorithm properly trained with balanced data will reign the right classification,” says Dr. Cabrera DeBuc.

The CTSI funding has been critical to this piece of her work. “Artificial intelligence in any field is quite expensive. It requires a lot of resources to make it happen. The most important is a computational component, a supercomputer,” she says.

Over the years, Dr. Cabrera DeBuc’s team has identified other areas of diagnostics for improvement. According to her, the current tools for diagnosing neurodegenerative diseases have limitations that can be addressed through ophthalmologic technology.

She is studying ways to use biomarkers in the eye to provide early diagnosis of neurological diseases like Alzheimer’s disease, a project funded by the Alzheimer’s Association and the Finker Frenkel Legacy Foundation.

Using images of the retina to detect diseases like Alzheimer’s could provide patients with a more comfortable experience and an earlier diagnosis.

“We need a non-invasive and low-cost methodology for differential diagnosis of a symptom like cognitive decline commonly reported in people with chronic conditions and neurocognitive disorders ,” says Dr. Cabrera DeBuc. “We need a methodology that can be ubiquitous, and that can detect disease before you may have a clinical manifestation of the problem. That’s why the eye is so good. It’s a beautiful approach to see what’s going on in the brain.”

UM computational neuroscientist Odelia Schwartz, Ph.D., statistician Emma Jingfei Zhang, Ph.D., retina and vitreous specialist William E. Smiddy, M.D., and neuro-ophthalmologist Carlos Mendoza-Santiesteban, M.D., are working with Dr. Cabrera DeBuc on advancing this aspect of her research and applying for NIH funding.

Rishav Sapahia, an AI Engineer in Cabrera DeBuc’s lab, is advancing the development of algorithms and the emerging AI framework to mitigate AI bias in retinal diagnostics. As a graduate student in the Computer Science Department, Rahul Dass, a U-Link predoctoral fellow, has also facilitated the AI data structuring as part of his research activities last summer.

The Cabrera DeBuc lab is also working on leveraging AI and telemedicine to develop a low-cost eye screening for primary care and community settings.