Skip to cookie consent Skip to main content

AI Tool, Spurred by Research Partially Funded by the 2018 Innovation Discovery Grant, Aims to Faster Diagnose Preventable Blindness in Babies

5 minute read
Image that reads "Where are they now?"
Jayashree Kalpathy-Cramer, PhD, 2018 Recipient of IDG Award

Babies weighing less than 3.3 pounds at birth can be affected by retinopathy of prematurity, a leading cause of preventable childhood blindness that happens when abnormal blood vessels grow in the retina. In the U.S., there are about 40,000 cases of ROP every year, resulting in blindness in about 600 infants. Most cases are mild, but the more severe cases are known as plus disease in ROP, which is characterized as retinal vessel tortuosity (twisted form) and dilation. 

Plus disease requires treatment with laser photocoagulation or injected drugs, but only if found early enough to avert blindness. That’s why diagnosis is key. Concerning is that the routine screening is complicated, and there is a shortage of ophthalmologists willing and able to manage it in the U.S. and worldwide. To solve this unmet need, Jayashree Kalpathy-Cramer, PhD, is finding a simpler way to screen for the condition, and to identify clinical decision support tools that give these babies a chance at protecting vision. 

Finding a less complicated way to diagnose this condition and treat it has become the decades-long research venture for Kalpathy-Cramer, who served as the director of the Quantitative Translational Imaging in Medicine Lab at Mass General Hospital from 2011-2022 and now serves in an adjunct capacity at Mass General Brigham. Was there a way, she hypothesized, to diagnose plus disease of ROP with an artificial intelligence tool to provide immediate, expert-level clinical diagnoses to medical images taken as part of ROP screening in neonatal units? A tool like this could be extremely helpful in saving sight.

Working collaboratively

Together with research partners from Oregon Health and Science University, Northeastern University and University of Illinois Chicago in 2018, Kalpathy-Cramer developed such a tool, a deep learning algorithm to diagnose plus disease in ROP. To test the algorithm, the software system would have to be trained and evaluated on a multi-institutional retinal image dataset from these and other institutions acquired as part of the NIH-funded i-ROP study. The images and the algorithm were tested against eight ROP experts, each of whom had more than 10 years of clinical experience. Data were collected and analyzed from December 2016 to September 2017. 

The outcome: The algorithm diagnosed plus disease in ROP with comparable or better accuracy than the human experts—an amazing finding that demonstrated its great potential in disease detection, monitoring, and prognosis in infants at risk.

“Our goal always was to be able to identify babies most at risk for the disease,” says Kalpathy-Cramer, “and to be assured that it (the software) works well in all populations.”

Support for further study came from the Innovation Discovery Grants of Mass General Brigham Innovation. Kalpathy-Cramer was a 2018 recipient of the grant, and funds were complemented by the NIH to further the data collection. More data collection is critical to validate the findings and to receive FDA approval for its use. 

The technology was licensed to Boston AI and subsequently to Siloam Vision, a startup founded by two of the co-inventors, J. Peter Campbell and R.V. Paul Chan. In furthering the work, Siloam Vision recently received a Small Business Innovation Research grant, funding that encourages domestic small businesses to engage in federal research and development with the potential for commercialization. 

Global reach

Kalpathy-Cramer is hopeful the tool will be most useful in underserved communities like India, Nepal and Mongolia, where the disease is more prevalent. She cautions that there will have to be other considerations for adapting the technology to certain populations as there are ethnic differences in eye structure. Other opportunities for the tool may include developing a mobile phone app with a camera system, especially beneficial for distant locations. 

Even with its great diagnostic success rate, the algorithm is currently being revised so that it becomes close to 100 percent accurate, according to Kalpathy-Cramer, who is now Chief of Artificial Medical Intelligence in Ophthalmology at the University of Colorado School of Medicine. There, she is charged with translating novel AI methods into effective patient care practices, including age-related macular degeneration and glaucoma at the Sue Anschutz-Rodgers Eye Center. To date, the original algorithm has been modified to also include a quantitative disease severity score. 

“This whole process started with the IDG grant,” says Kalpathy-Cramer, who presented her research at the 2019 World Innovation Forum. “We knew nothing about commercialization and thanks to Chris Coburn, Chief Innovation Officer at Mass General Brigham and Heonick Ha Jeong, PhD, Associate Director, Business Development & Licensing at Mass General Brigham, we learned so much about how to find our way.” 

Ha Jeong admires Kalpathy-Cramer’s ambitions. “Jayashree is a fantastic researcher with lots of passion. She wants to help babies and we can’t wait to see how she moves this project even further.”