MRI scans can be nerve-racking for patients and expensive to the health care system. They’re also one of the best ways to determine if patients with multiple sclerosis have developed new brain lesions — spots that indicate inflammation or damage to the coating of certain nerves — to see who might benefit from treatment interventions. It’s common for a neurologist to order an MRI to monitor their patients at most visits.

But what if there were a way to reduce unnecessary MRIs for patients who are unlikely to have new lesions? That’s the idea behind a precision medicine project at Johns Hopkins.

Precision Medicine at Work

Neurologists, including Ellen Mowry, M.D., director of the Multiple Sclerosis Experimental Therapeutics Program, working with data scientists Michael Robinette and William Gray-Roncal at the Johns Hopkins University Applied Physics Laboratory, built a machine-learning program to help predict the likelihood of a patient developing new lesions and needing an MRI.

The program, called LESION (Longitudinal, Equitable, Systemized Imagine Operations for Neuroimmunology), was designed using deidentified data from over 900 patients with multiple sclerosis treated at Johns Hopkins. It incorporates information such as medication use, whether a patient had new lesions during their previous MRI scan, and responses to an iPad clinical assessment that patients complete at the time they come to clinic. The assessment includes cognitive and physical performance testing and patient-reported symptoms related to anxiety, depression and fatigue.

In a simulated trial using existing data from patients, LESION predicted whether or not a physician should order an MRI for patients with MS. “We found that we could get about a 60% reduction in MRIs based on the tool’s recommendations,” Robinette says. “Patients who at the last MRI had new lesions, tended to continue having new lesions — that’s the kind of ongoing risk you would expect. Patients who are on aggressive medications specifically designed to cut down lesion formation, and older patients, had fewer new lesions, and therefore may not need an MRI at every clinical visit.”

“The program helps clinicians make decisions from a very broad perspective of patient data compared with what a clinician can see with only a single person’s clinical history,” he adds. “To make decisions based on 12, 15, 20 variables at once is something that machine learning can do really easily.”

The team is planning a prospective trial of the tool to investigate whether the program is equally accurate using data from other sites participating in the Multiple Sclerosis Partners Advancing Technology and Health Solutions (MS PATHS) study.

Find treatment options at the Johns Hopkins Multiple Sclerosis Center, and learn about the research underway at the Multiple Sclerosis Precision Medicine Center of Excellence.