🇺🇸 Ellen Xu

The disease

Kawasaki disease inflames the blood vessels of small children. The Society for Science describes it as the leading cause of acquired heart disease in young children, and the danger lies in the coronary arteries, which can be permanently damaged if the illness is not treated in time.

The reason a photograph is a plausible input is that several of the signs clinicians look for appear on the surface of the body rather than in a laboratory result. The difficulty is that those signs overlap with the ordinary illnesses of childhood, so recognition depends on a clinician assembling a pattern rather than reading a single test.

Timing is the whole argument for a triage tool. A condition that is treatable but easy to mistake for something ordinary creates a window in which the useful intervention is not a new therapy but earlier recognition, and that is the gap the project was aimed at.

The algorithm

Xu's approach was to use smartphone photographs as that input. An algorithm trained on such images can act as a prompt — a second opinion offered at the point where a parent or a clinician is deciding whether what they are looking at is worth escalating.

The award description states the purpose and the method without publishing the training data, the accuracy, the validation or any clinical trial. Those omissions matter for a diagnostic tool, and they are recorded here as gaps. What this page can state is what the competition organiser published: an algorithm using smartphone photos to help diagnose the disease, judged third best among 40 finalists.

Third place, and the money

The Regeneron Science Talent Search, founded in 1942, reached its 82nd year in 2023. Nearly 2,000 students entered and 40 reached the final; $1.8 million went to finalists and $3.1 million in total including school awards. Regeneron has backed the competition since 2017 under a ten-year, $100 million commitment.

Neel Moudgal of Michigan took first place and $250,000 for a model predicting RNA structure; Emily Ocasio of Virginia took second and $175,000 for AI analysis of racial disparity in homicide coverage. Xu's $150,000 was third. Maya Ajmera, the Society's chief executive, said she was "in awe of their creativity and conviction."

Why phone-based diagnosis is a live field

A smartphone camera is the most widely distributed medical sensor on earth. Systems that read visible signs — skin lesions, eye disease, jaundice in newborns — have been built for exactly that reason, and the appeal is the same each time: the hardware is already in the patient's home.

The hard part is never the first demonstration. It is proving that the result holds across skin tones, cameras, lighting and populations, and that the tool changes what a clinician does without generating a flood of false alarms. Those are the questions that separate a prize-winning algorithm from a deployed one, and none of them is answered in this record.

What is documented — and what is not

Firm: third place, the $150,000, her age, her home city, the disease and the method, published by the Society for Science in March 2023.

Open: performance figures, validation, regulatory status, any publication, and what Xu has done since. This page makes no claim that the algorithm is available or approved for use, because no source gathered here says so.