Школьник из Аризоны занял место в топ-300 страны, создав ИИ для диагностики сердца

Jay Chandra, a former student of Gowan Middle School, recently earned a spot as one of the nation’s top 300 STEM students in the Thermo Fisher Scientific Junior Innovators Challenge, a prestigious competition for middle schoolers centered around technical and scientific excellence. Chandra’s involvement began last year following his submission for the school’s science fair, when he coded and trained an artificial intelligence model to detect “left ventricular ejection fraction via electrocardiogram.” This breakthrough reduces the need for the far more expensive echocardiogram test, which is currently the standard method for detection. “Around a year ago, I was wondering about how I can use artificial intelligence, AI, to solve some problems around the medical field, specifically around the heart,” he said. “I decided on trying to make an artificial intelligent model using machine learning to help estimate people’s left ventricular ejection fraction, or LVEF for short, left ventricular ejection fraction being a measure of when your heart beats, how much blood leaves your heart.” According to Chandra, part of his goal is to bring greater access of in-depth healthcare to underserved communities. “So many people in rural areas and underserved areas in the United States and the world lack access to echocardiogram, which is an ultrasonographic test which needs an ultrasonographer specialist to take an ultrasound of your heart to measure this left ventricular ejection fraction,” Chandra explained. “However, it tends to cost more than a thousand dollars usually and is only in specific hospitals and specific clinics around the country and the world, so it’s not accessible to everybody.” Chandra’s project instead uses the much cheaper electrocardiogram, where electronic leads are attached to the patient’s torso to read electrical pulses from the heart. While useful, this test lacks the ability to measure left ventricular ejection fraction. To complete this project, Chandra spent a few months teaching himself Python, a programming language, and building the AI models that could interpret electrocardiogram data to determine a patient’s left ventricular ejection fraction. This involved looking at data set of more than 100,000 patients who had received both tests. “I trained that on my artificial intelligent model, which is effectively structures as a bunch of interconnecting neurons, which I’ve coded. Kind of like the neurons in your brain, and they only fire when they reach a certain amount of activation,” he said. “If you train it back and forth, using special algorithms, you can eventually achieve a high enough accuracy so that it can be clinically testable.” All of this began last November as a project for his school’s science fair, after which the judges determined that the project was good enough to advance to the county science fair and, eventually, the state science fair. “I got second place for health medicine in state, along with, I believe, third in engineering from the Southern Arizona Science, Research and Engineering Fair,” he said. “That qualified me for the Thermo Fisher Junior Innovators Challenge, in which I got top 300 in the country.” What Chandra enjoyed about the project was the way it combined his passion for medicine and artificial intelligence. “I feel happy because my interest in artificial intelligence and my interest in health and medicine can come together for this project, and I honestly feel very happy because artificial intelligence is a growing field and I thought it could really help with health and medicine, but I never really found out until this point how I can combine the two to make a project that actually is for the people,” said Chandra. “But I came up with this project, I did it, and at least to my terms it came out successfully, and I’m really happy that this project is getting recognition.” Chandra’s mother, Preeti Chandra, said that she was impressed by her son’s ability to do the research and work out the technical issues, particularly as he largely worked on his own through the project. “I’m feeling very proud because he put so much work into it,” she said. “I didn’t know what was to come of it because we don’t have any technology background at home to guide him, so he himself learned and came up with his own ideas and troubleshooted and sometimes he would get frustrated. Sometimes he would get encouraged. Ultimately a project came out of it and I’m so happy that the results were good and that he got recognized here at Yuma County.” But even more than the intellectual achievement, she was proud of his heart and his goals with the project. “More than the achievement aspect, it’s about the intention to do good for public benefit,” she said. Chandra has since released his model for free online as a resource for others to study and use. Of the initial 100,000 cases he used to train the model, he reserved 20,000 for testing, where he was able to reach an overall accuracy of 88%.