Automated detection of cervical ossification of the posterior longitudinal ligament in plain lateral radiographs of the cervical spine using a convolutional neural network

Abstract

Cervical ossification of the posterior longitudinal ligament (OPLL) is a contributing factor to spinal cord injury or trauma-induced myelopathy in the elderly. To reduce the incidence of these traumas, it is essential to diagnose OPLL at an early stage and to educate patients how to prevent falls. We thus evaluated the ability of our convolutional neural network (CNN) to differentially diagnose cervical spondylosis and cervical OPLL. We enrolled 250 patients with cervical spondylosis, 250 patients with cervical OPLL, and 180 radiographically normal controls. We evaluated the ability of our CNN model to distinguish cervical spondylosis, cervical OPLL, and controls, and the diagnostic accuracy was compared to that of 5 board-certified spine surgeons. The accuracy, average recall, precision, and F1 score of the CNN for classification of lateral cervical spine radiographs were 0.86, 0.86, 0.87, and 0.87, respectively. The accuracy was higher for CNN compared to any expert spine surgeon, and was statistically equal to 4 of the 5 experts and significantly higher than that of 1 expert. We demonstrated that the performance of the CNN was equal or superior to that of spine surgeons.