Gustavo Rohde, professor of biomedical engineering and electrical and computer engineering at the University of Virginia, is senior author of a collaborative research study in machine learning to identify patients vulnerable to osteoarthritis.
Rohde’s Ph.D. advisee Shinjini Kundu, M.D., a resident physician in radiology and computer scientist at the Johns Hopkins Hospital, co-led the team with Rohde, joined by researchers from the University of Pittsburgh and Carnegie Mellon University, the University of Copenhagen, and the U.S. National Institutes of Health. M. Shifat-E-Rabbi, a Ph.D. student of biomedical engineering, also contributed to the study as a member of Rohde’s Imaging Data and Science Lab.
Osteoarthritis is the most common type of arthritis, in which the cushion that ensures smooth joint movement is damaged or breaks down. Osteoarthritis reveals itself in joint pain confirmed with a traditional x-ray. At this point, the loss of the cushion, or bone cartilage, is irreversible. Doctors today can only treat symptoms, typically resulting in joint replacement surgery.
The research team investigated whether machine learning applied to MRI images of knee cartilage can point to early signs of osteoarthritis and predict who will develop the disease. The team used a customized machine learning technique invented in Rohde’s lab called transport-based morphometry, an approach that quantifies information about the shape and texture of biological forms seamlessly from image data.
The team conducted their research on knee scans of 86 individuals enrolled in the National Institutes of Health Osteoarthritis Initiative, a longitudinal, mutli-center, prospective, observational knee osteoarthritis study. None of the patients had symptoms when the study began; about half developed osteoarthritis after three years.
Using machine-learning algorithms, the team trained the system to automatically differentiate between people who would and would not progress to osteoarthritis. The technique detected specific biochemical changes in the center of the knee’s cartilage of those who were pre-symptomatic at the time of the baseline imaging, including decreases in water concentration. The system accurately detected 78% of future osteoarthritis cases.
More studies are needed to determine whether the team’s machine learning method could be useful as a clinical tool, to predict who may develop osteoarthritis and benefit from early interventions.
The team published their paper, Enabling Early Detection of Osteoarthritis from Presymptomatic Cartilage Texture Maps via Transport-based Learning, in the Proceedings of the National Academy of Sciences.
A number of news outlets highlighed their work, including CBS News and the National Institutes of Health: