In the oldest and most prestigious young adult science competition in the nation, 17-year-old Ellen Xu used a kind of AI to design the first diagnosis test for a rare disease that struck her sister years ago.
With a personal story driving her on, she managed an 85% rate of positive diagnoses with only a smartphone image, winning her $150,000 grand for a third-place finish.
Kawasaki disease has no existing test method, and relies on a physician’s years of training, ability to do research, and a bit of luck.
Symptoms tend to be fever-like and therefore generalized across many different conditions. Eventually if undiagnosed, children can develop long-term heart complications, such as the kind that Ellen’s sister was thankfully spared from due to quick diagnosis.
Xu decided to see if there were a way to design a diagnostic test using deep learning for her Regeneron Science Talent Search medicine and health project. Organized since 1942, every year 1,900 kids contribute adventures.
She designed what is known as a convolutional neural network, which is a form of deep-learning algorithm that mimics how our eyes work, and programmed it to analyze smartphone images for potential Kawasaki disease.
However, like our own eyes, a convolutional neural network needs a massive amount of data to be able to effectively and quickly process images against references.
For this reason, Xu turned to crowdsourcing images of Kawasaki’s disease and its lookalike conditions from medical databases around the world, hoping to gather enough to give the neural network a high success rate.
Xu has demonstrated an 85% specificity in identifying between Kawasaki and non-Kawasaki symptoms in children with just a smartphone image, a demonstration that saw her test method take third place and a $150,000 reward at the Science Talent Search.