Affecting millions of people worldwide, osteoarthritis (OA) is the most common form of arthritis. It occurs when the protective cartilage that cushions the ends of your bones wears down over time. X-rays of the affected joints are the main way osteoarthritis is identified. Now researchers at the University of Pittsburgh School of Medicine and Carnegie Mellon University College of Engineering have created a machine-learning algorithm that can detect subtle signs of osteoarthritis on an MRI scan taken years before symptoms even begin.
Their study, “Enabling early detection of osteoarthritis from presymptomatic cartilage texture maps via transport-based learning,” is published in PNAS.
“The gold standard for diagnosing arthritis is x-ray. As the cartilage deteriorates, the space between the bones decreases,” explained coauthor Kenneth Urish, MD, PhD, associate professor of orthopedic surgery at Pittsburgh School of Medicine and associate medical director of the bone and joint center at UPMC Magee-Womens Hospital. “The problem is, when you see arthritis on x-rays, the damage has already been done. It’s much easier to prevent cartilage from falling apart than trying to get it to grow again.”
“Today, OA is detected after bone damage has occurred, at an irreversible stage of the disease. Currently, no reliable method exists for OA detection at a reversible stage. We present an approach that enables sensitive OA detection in presymptomatic individuals. Our approach combines optimal mass transport theory with statistical pattern recognition,” noted the researchers.
OA may develop in any joint, but most commonly affects the knees, hips, hands, facet joints, and feet. Although OA primarily affects the elderly, sports-related traumatic injuries at all ages can lead to post-traumatic OA.
“When doctors look at these images of the cartilage, there isn’t a pattern that jumps out to the naked eye, but that doesn’t mean there’s not a pattern there. It just means you can’t see it using conventional tools,” added Shinjini Kundu, MD, PhD, who is currently a resident physician and medical researcher, department of radiology at Johns Hopkins.
Kundu trained the model on a subset of the knee MRI data and then tested it on patients it had never seen before. Kundu repeated the process dozens of times, with different participants withheld each time, to test the algorithm on all the data.
The researchers observed 86 healthy individuals selected from the Osteoarthritis Initiative, with no symptoms or visual signs of disease on imaging. The algorithm predicted osteoarthritis with 78% accuracy from MRIs performed three years before symptom onset.
The primary driver behind the emergence of AI in medical imaging has been the desire for greater efficacy and efficiency in clinical care. These findings can one day lead to patients being treated with preventative drugs rather than undergoing joint replacement surgery, and a greater adoption of AI in imaging.
The researchers believe their work demonstrates that OA detection may be possible at a potentially reversible stage. “In the future, coupling presymptomatic OA detection with emergent clinical therapies could modify the outcome of a disease that costs the United States healthcare system $16.5 billion annually. Furthermore, our technique is broadly applicable to earlier image-based detection of many diseases currently diagnosed at advanced stages today.”