Detection of Shoulder Rotator Cuff Tears from X-Ray Image by Using Convolutional Neural Network

Abstract

Shoulder rotator cuff tears are common disabilities of the shoulder joint caused by daily activities, sports injuries, and aging. Magnetic resonance imaging is widely used to diagnose rotator cuff tears, but radiography is a more rapid and popular imaging technique. In this study, we proposed a detection method using convolutional neural networks (CNN) on X-ray images of the shoulder for simpler diagnosis of rotator cuff tears. 139 subjects, classified into two classes according to the severity of the tear, were compared in terms of detection accuracy between five regions of interest on the shoulder and two combinations of CNN models, with a maximum accuracy of 79.3% accuracy.