John Zech, MD, Third-Year Radiology Resident

It was a very comfortable life, but I wanted to do something more constructive.

As John Zech trains to become a radiologist, he's already thinking about how to improve the way we diagnose disease. A previous career in finance, along with a masters in statistics, have left him with a habit of thinking quantitatively. "My big interest is in how we can use machine learning to improve the quality of our diagnoses," he says.

When Zech graduated from Harvard University in 2006, in the midst of a booming economy, medicine was not on his mind. "Half my class went into finance or consulting, so I sort of rode that wave," he explains. He moved to New York City to pursue a career in investment management.

Before long, he found himself building quantitative models for his investment management firm, to help guide them as they selected stocks and companies to analyze. He learned more about machine learning while pursuing his masters in statistics at Columbia. By the time graduated, his interests had shifted away from finance.

"It was a very comfortable life, but I wanted to do something more constructive," he says. "I was looking for a challenge."

The son of a physician, Zech remembers hanging out in his father's office when the babysitting fell through, or tagging along on hospital rounds. "As a kid, I was mostly concerned with the magazine selection in the waiting room," he says. "I'd hope they'd have Car and Driver." Looking back, he realizes how much he absorbed on those visits. "You get exposed to what a professional privilege it is to have a job where just by showing up to work and doing what's in front of you, you have a chance to really help somebody out."

In the meantime, he'd also gotten excited about new deep learning tools and noticed how limited their application was in medicine.

Zech went back to school to complete his pre-med requirements and attended medical school at Mt. Sinai, where he discovered that on all his rotations, his favorite part was making a diagnosis. He also liked the fast pace of radiology. "If I'm on call overnight, I might see 100 different patients, which you could never do if you were a primary clinician."

He pursued a residency at Columbia after noticing during medical school that many of the most complex patients were referred to NewYork-Presbyterian. "As a trainee, it's really instructive to see these kinds of cases," he says.

Zech laughs as he remembers his first days as a resident. "You don't know how to do anything! You've spent years in medical school and applying to radiology residency, and then you show up and don't know how to use a dictaphone."

Now in his third year, Zech is very comfortable with a dictaphone and still thinking about where machine learning fits into a field where there is a lot of data and unique challenges. "A lot of what we do is look at prior studies and try to infer what the course has been and what new information has been added by this new study," he explains. "That's really hard for computers." He is working on projects with Drs. Diego Jaramillo and Tony Wong, both musculoskeletal radiologists who are investigating the use of deep learning techniques to improve diagnoses or extract more information from standard X-rays or MRI's.

"How do we take a cool idea and learn to deploy it in a useful way where it actually answers questions and doesn't annoy people," Zech says. "That's what I'm excited about."