An aspiring radiologist designs an award-winning algorithm to detect pneumonia.
The third year of medical school is stressful for any student. But Ian Pan ’16 AM’16 MD’20 didn’t want to pass up the opportunity to compete in the 2018 Radiological Society of North America (RSNA) Pneumonia Detection Challenge.
So last fall, he arranged his clinical rotations so he could devote more than 20 hours a week to his project. Competition was stiff: more than 1,400 teams from across the globe, including industry professionals, physicians, and students, vied to design an algorithm to recognize pneumonia, using more than 30,000 X-ray images provided by the RSNA.
“There are two elements to the task,” Pan says. “The first is to figure out if there is pneumonia in the image. If there is, you have to determine where it is and draw a box around it with the algorithm.”
His hard work and sleepless nights paid off: in December, Pan and his collaborator, Alexandre Cadrin-Chênevert, MD, a radiologist and computer engineer at CISSS de Lanaudière in Québec, won first place. During one of Pan’s presentations, an RSNA committee member praised him as the “future of [the]field.”
“I was interested in radiology when I first went to medical school and was trying to figure out ways to incorporate machine learning,” Pan says. “I stumbled onto deep learning and realized it was a very powerful tool to build models that could recognize images, which is what a radiologist does.”
In the past three years, Pan has created algorithms to detect lung disease on chest X-rays, head bleeds on CT scans, and thyroid cancer on ultrasounds. The experiences were instrumental in helping him compete in the RSNA challenge.
He also credits his success to the mentorship of Professors Lisa and Derek Merck. Under their guidance, he completed several projects using machine learning, including examining ultrasounds for cancer and X-rays for bone age.
“When he first came in, he had momentum,” says Derek Merck, PhD, assistant professor of diagnostic imaging, of engineering, and of radiation oncology. “He had been doing a master’s in statistics, so he already had a background in machine learning and had been working with big data sets. Basically when he showed up, he just started looking for data.”
“Ian is a hard worker and takes advantage of the resources that are available to him from the data we curate in the lab,” adds Lisa Merck, MD, MPH, associate professor of emergency medicine, of diagnostic imaging, and of neurosurgery.
Pan wants to pursue radiology and continue to integrate machine learning and medicine. “Having people who speak both languages [medicine and AI]is really important. AI is going to influence medicine and radiology greatly over the next few decades,” he says. “It’s important for doctors to figure out how this will happen. Because ultimately, our hope is to use this technology in a way that benefits our patients.”