Kopparapu, who grew up in Virginia, has been interested in science her whole life, and added computers to her list of interests after attending a programming workshop held by the National Center for Women and Information Technology.
While taking high school classes in computer science, computer vision, and artificial intelligence, Kopparapu was typically one of the only girls in the class, and to help combat this, founded the nonprofit Girls Computing League; one of her sponsors is Amazon Web Services. She also holds coding workshops for underprivileged kids, gave a speech at the March for Science in Washington D.C. this spring, organized a student-led research symposium at her high school, and had the opportunity to present her work on her diabetic retinopathy diagnostic system at the O’Reilly Artificial Intelligence conference last month.
“I went home and taught myself Java, HTML, Python, C. My mom had to tear me away from the computer. I’d forget to eat,” Kopparapu said.
Kopparapu’s grandfather, who lives in India, began displaying symptoms of the disease in 2013. It can often go undetected, and while he was eventually diagnosed and treated, his vision did deteriorate. According to Kopparapu, of the total 415 million people with diabetes, one-third will develop diabetic retinopathy, and even though medication and surgery can stop or even reverse eye damage if caught in time, 50% of those will go undiagnosed; half the patients who have severe forms will go blind in five years.
Kopparapu explained, “The lack of diagnosis is the biggest challenge. In India, there are programs that send doctors into villages and slums, but there are a lot of patients and only so many ophthalmologists.”
She wondered if there was an easy, inexpensive way to diagnose the disease, and came up with the idea for Eyagnosis, a machine-learning system/smartphone app with a 3D printed lens that could possibly turn a lengthy, expensive diagnostic procedure into nothing more than a quick photo session. Kopparapu got to work, spending a lot of time on Google and emailing doctors and researchers, before formulating a plan. She teamed up with her brother and a fellow classmate, and they used a machine-learning architecture called a convulutional neural network (CNN) to set up the diagnostic AI behind the Eyeagnosis. Neural nets parse large sets of data and look for similar patterns – as the design resembles the visual system of the human brain, CNNs are excellent for classification.
Kopparapu said, “It’s kind of funny that we’re using a system based on how the retinal system works to diagnose a retinal disease.”
She used ResNet-50, an off-the-shelf CNN developed by Microsoft researchers, to build her network, and utilized the 34,000 retinal scans found in the EyeGene database from the National Institutes of Health (NIH) as training data, so she and her team could teach the AI system to recognize signs of the disease in eye photos and give a preliminary diagnosis. Many of the images in the database were poorly exposed or blurry, but according to Kopparapu, that wasn’t bad news.
Her team trained the ResNet-50 to detect diabetic retinopathy as accurately as a real pathologist could; it also highlights microaneurysms and blood vessels in each image, which normally means having to inject fluorescent dye into someone’s eye.
“But that was actually a good thing. It’s very representative of the real-world conditions you’d get with using a smartphone,” she explained.
Kopparapu said, “We’re trying to make it as easy as possible for an ophthalmologist to look at all that info and say ‘Here’s my final diagnosis.’”
In the fall, Aditya Jyot Eye Hospital in Mumbai agreed to test Kopparapu’s Eyeagnosis app, and in November, she sent the first 3D printed prototype to the hospital, and the system has already made accurate diagnoses for five patients there. Once it’s fitted to a smartphone with a 3D printed mount, the system’s 3D printed lens takes retinal scans – the phone’s off-center, scattered flash is focused to illuminate a person’s retina. Some experts believe that the system has “commercial potential,” but larger companies may not think it has potential for a profit margin, and that it still needs some work.
“What she’s going to need is a lot of clinical data showing that [Eyeagnosis] is reliable under a variety of situations: in eye hospitals, in the countryside, in clinics out in the boonies of India,” said J. Fielding Hejtmancik, a visual diseases expert with NIH.
J. Fielding Hejtmancik, a visual diseases expert with NIH, said, “The device is ideal for making screening much more efficient and available to a broader population. These kids have put things together in a very nice way that’s a bit cheaper and simpler than most [systems designed by researchers]—who, by the way, all have advanced degrees!”
Discuss in the Diagnostic Eye System forum thread at 3DPB.com.[Sources: TechCrunch, IEEE Spectrum]