Deep learning detects common shoulder pain on x-rays—a potential safeguard for busy physicians
Deep learning can detect underlying causes of common shoulder pain on radiography images, potentially helping busy physicians manage their worklists and safeguard against errors.
That’s according to a group of German radiologists who tested a convolutional neural network trained on more than 2,700 shoulder x-rays from multiple institutions. It proved highly accurate at detecting a number of injuries, including osteoarthritis, joint dislocations, and fractures, researchers reported Saturday in Skeletal Radiology.
They also noted the network may be particularly valuable for referring physicians, who often read shoulder radiographs before radiologists and are forced to make treatment decisions under pressure.
“We present a robust CNN with the ability to detect common causes of shoulder pain on plain radiographs, even when faced with impaired image quality as often seen in clinical practice, especially in emergency settings,” Nils F. Grauhan, a radiologist at the Universitätsmedizin Berlin, and colleagues explained in the study.
Despite an ever-growing amount of research investigating AI in radiology, few, focus on shoulder radiographs and corresponding causes of pain, Grauhan et al. noted.
So they decided to explore the issue using a ResNet-50 network trained on thousands of clinical radiographs. Radiologists hand-labeled the images for six findings: proximal humeral fractures, joint dislocation, periarticular calcification, osteoarthritis, osteosynthesis, and joint endoprosthesis
Overall, the tool performed well, notching near perfect accuracy for detecting osteosynthesis and endoprosthesis. It produced relatively high area under the curve scores, hitting the 0.945 mark for detecting osteoarthritis, 0.896 for joint dislocation, 0.871 for fractures, and 0.800 for periarticular calcifications.
And in terms of sensitivity and specificity, the CNN reached 0.75 and 0.86 for fractures, 0.95 and 0.65 for joint dislocation, 0.90 and 0.86 for osteoarthrosis, and 0.60 and 0.89 for calcification, respectively.
“We conclude that plain radiographs of the shoulder may pose a considerable challenge to CNNs, especially when they are expected to detect more subtle findings such as periarticular calcification,” the group explained. “Overall, however, our results are promising and strongly encourage further trials ultimately leading to clinical implementation.”
Matt joined Chicago’s TriMed team in 2018 covering all areas of health imaging after two years reporting on the hospital field. He holds a bachelor’s in English from UIC, and enjoys a good cup of coffee and an interesting documentary.