A Korean research team said it has found that a deep learning artificial intelligence (AI) model showed high accuracy to detect and classify humerus fractures.
The team, led by Chung Seok-won, a professor in the orthopedic surgery department of Konkuk University Hospital, evaluated the ability of the AI model to diagnose proximal humerus fractures using X-ray images of 1,891 patients. The machine achieved 96 percent accuracy, with 0.99 of sensitivity and 0.97 of specificity.
|Chung Seok-won, a professor of the orthopedic surgery department at Konkuk University Hospital|
The AI model was also more accurate than orthopedists to classify types of fractures. Chung’s team divided proximal humerus fracture types into four – greater tuberosity, surgical neck, three-part fracture, and four-part fractures. The AI showed similar or higher accuracy than its human counterparts.
The research on the AI’s diagnosis ability using deep learning algorithm in orthopedic surgery is the second in the world, after Sweden’s report in December 2017.
“In orthopedic surgery, it was the first in the world to prove an AI’s disease-specific diagnostic ability with high accuracy,” Chung said.
In classifying fracture types, the AI showed similar or higher capacity than orthopedists specialized in the shoulder, according to Chung.
“When fracture was complex, the AI showed superior performance,” he said.
Chung noted that the study was particularly meaningful because it confirmed the potential of using an AI model in orthopedic surgery where X-ray films are the basis of diagnosis.
“The AI will make a quick and accurate diagnosis possible for trauma patients,” he said.
Chung’s team includes Kim Young-joon of Center for Bionics at Korea Institute of Science and Technology, Kyungpook National University Hospital, Kangwon National University Hospital, National Police Hospital, Seoul St. Mary’s Hospital, and Dong-A University Hospital. Han Seung-seok of I-dermatology Clinic helped develop the AI model.
The study results were published in the international journal Acta Orthopaedica in March.
<© Korea Biomedical Review, All rights reserved.>