Healthwatch 16: Machine Learning Aids Diagnosis

MAHONING TOWNSHIP, Pa. -- Technology has come a long way when it comes to how we connect with people, and now at Geisinger, it's helping patients who need care first be treated first.

Intracranial hemorrhaging, or a brain bleed, is a fairly common problem, according to Dr. Brandon Fornwalt, from the Department of Imaging Science & Innovation at Geisinger Medical Center near Danville.

Scans show the brain of an elderly patient whose only symptom was confusion.

"Lots of things can cause elderly patients to be confused. It's a diagnostic dilemma, and it's an acute finding that needs to be treated very quickly," said Dr. Fornwalt.

That patient is 89-year-old Marion Strausser from Berwick. Her daughter Camille Bartlett, who now lives in Lancaster County, says her mom likes the outdoors and loves naps. They talk every day. One day, about a year and a half ago, Camille noticed something odd.

"There were a couple things she described in what she had for dinner that made no sense, so I immediately called the nursing house staff and said, 'You've got to check mom out.'"

But Marion looked and felt fine, and her vital signs were good. She was sent for a CT scan, but there's a chance her case would not have been deemed critical, according to Dr. Aalpen Patel, if not for machine learning.

"Like in Ms. Strausser's case, we had scans which showed bleeding," Dr. Patel said. "We were able to easily interpret that she needed to come to the ED and be acted on right away."

Dr. Patel is the chairman of radiology at Geisinger Health System. He and Dr. Fornwalt explained using some 37,000 CT scans, labeled "bleed" or "no bleed," they were able to teach the machine which is which, learning to spot the most critical cases.

"You may have heard Google does 'x, y, z,' or Facebook can do 'x, y, z.' We are using similar technology to teach machines for patient care," said Dr. Patel.

Left undetected, the brain bleed could have been fatal. Marion needed only a tweak in medication.

Dr. Patel points out that machine learning is not meant to replace human workers. It's meant to help human workers triage important cases. He says Geisinger will look into machine learning in other areas of medicine as well.