AI Promising for Axial Spondyloarthritis Diagnosis

Dr. Denis Poddubnyy

Artificial intelligence (AI) detected axial spondyloarthritis (axSpA) in radiographic scans of sacroiliac joints as accurately as human experts, shows a new study.

The findings have the potential to greatly expand access to accurate axSpA diagnoses for patients in areas that lack trained experts on the condition, said Denis Poddubnyy, M.D., head of rheumatology at Charité University in Germany. The findings will be presented on Monday at the annual meeting of the American College of Rheumatology.

AxSpA is typically diagnosed by radiograph, but successfully identifying signs of axSpA in a scan can be difficult. Readings by specialists with little experience in rheumatology can be inaccurate, but then, not all healthcare sites have easy access to physicians with rheumatology expertise.

“We see a big discrepancy between the local and central assessment of sacroiliitis reaching sometimes half of the cases,” Dr. Poddubnyy said.

Dr. Poddubnyy’s team wanted to see if they could develop an artificial neural network capable of identifying signs of AxSpA in radiographic scans of patients’ joints, so they trained a machine learning program using scans of 1,669 radiographs of sarolitic joints.

The results were evaluated by human experts who used the modified New York criteria to detect radiographic sacroiliitis. A local clinician and an expert reader evaluated each radiograph. If the local clinician and expert disagreed, a second expert evaluated the radiograph. This ensured that the data was evaluated by at least two humans.

​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​

Once the neural network had been trained, researchers tested it on 100 more scans and compared its results with those of human experts evaluating the same set of scans.

The software produced results that were overwhelmingly accurate: 90 percent of the neural network’s judgments agreed with those of the expert readers.

The researchers are hopeful that their neural network will be helpful not only to clinicians seeking to diagnose and treat patients with axSpA, but also other researchers looking to sign patients up for clinical trials.

“We think that such an approach could be a useful tool for clinical practice and also for clinical research,” Dr. Poddubnyy said. “The next step, would be to develop a similar network for MRIs. This would be especially relevant for the diagnosis and differential diagnosis of AxSpA at the early stage.”

_________________

REFERENCE

ABSTRACT NUMBER: 2018. "Development and Validation of an Artificial Intelligence Approach for the Detection of Radiographic Sacroiliitis." Monday, November 9, 2020