In June 2016, Kavya Kopparapu was looking for a new project that would put in practice her computer science skills. She immediately thought of her grandfather who was showing symptoms of diabetic retinopathy: a diabetes complication that affects eyes, and eventually, can cause blindness. Her grandfather has been diagnosed and treated, but unfortunately, his vision slightly deteriorated. Although it can be treated (if diagnosed early), most patients never receive care.
The statistics are alarming: Of 415 million diabetics worldwide, 1/3 will develop the disease. 15% will not be diagnosed. Of patients with severe forms, 1/2 will go blind in 5 years.
As you can imagine, the biggest challenge is clearly the lack of diagnosis. That was the beginning of Kopparapu’s AI project: Eyeagnosis, a smartphone app that has the potential to replace a long and expensive diagnostic procedure to… a quick photo.
Kopparapu and her team have trained an AI to recognize signs of diabetic retinopathy in photos of eyes. To do it, they used a machine learning architecture known as a convolutional neural network (CNN) which requires vast sets of data (photos) to work well. They haven’t built it from scratch but rather have used a model developed by Microsoft Researchers called ResNet-50.
For the dataset, (all the photos required to teach the AI) she found the NIH’s EveGene database which included 34,000 retinal scans. As you can imagine, many of these pictures were not very well exposed. As Kopparapu pointed out: “That was actually a good thing. It’s very representative of the real-world conditions you’d get with using a smartphone.”
Fielding Hejtmancik, an expert in visual diseases at the National Institutes of Health (NIH) said that “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!”
Kopparapu is really an ambitious student. In high school, she took classes in computer science, then computer vision, and finally Artificial Intelligence. However, she quickly realized that… there were almost no girls in those classes. Consequently, she started an organization to empower girls to pursue computer science.
Eveagnosis began with a lot of “google” and emails – to ophthalmologists, computational pathologists, biochemists, epidemiologists, neuroscientists, physicists, and expert in machine learning. Then Kopparapu formed a plan with all the information she disposed of.
Not surprisingly, the AI System has already acquired the accuracy of a human pathologist. Instead of competing with them, humans will have to cooperate with machines and work in tandem with them. Eyeagnosis is the beginning of what is going to happen on a larger scale: “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’.
The next step for Kopparapu is to show off that her app is reliable. To accomplish this endeavor, she will need a lot of clinical data showing that the AI works well.
Thanks so much for reading, that’s the end of Article 90. Share it with your friends and don’t forget to follow me on social media!