Scientific Reports (Feb 2022)

A deep learning algorithm to identify cervical ossification of posterior longitudinal ligaments on radiography

  • Koji Tamai,
  • Hidetomi Terai,
  • Masatoshi Hoshino,
  • Akito Yabu,
  • Hitoshi Tabuchi,
  • Ryo Sasaki,
  • Hiroaki Nakamura

DOI
https://doi.org/10.1038/s41598-022-06140-8
Journal volume & issue
Vol. 12, no. 1
pp. 1 – 9

Abstract

Read online

Abstract The cervical ossification of the posterior longitudinal ligament (cOPLL) is sometimes misdiagnosed or overlooked on radiography. Thus, this study aimed to validate the diagnostic yield of our deep learning algorithm which diagnose the presence/absence of cOPLL on cervical radiography and highlighted areas of ossification in positive cases and compare its diagnostic accuracy with that of experienced spine physicians. Firstly, the radiographic data of 486 patients (243 patients with cOPLL and 243 age and sex matched controls) who received cervical radiography and a computer tomography were used to create the deep learning algorithm. The diagnostic accuracy of our algorithm was 0.88 (area under curve, 0.94). Secondly, the numbers of correct diagnoses were compared between the algorithm and consensus of four spine physicians using 50 independent samples. The algorithm had significantly more correct diagnoses than spine physicians (47/50 versus 39/50, respectively; p = 0.041). In conclusion, the accuracy of our deep learning algorithm for cOPLL diagnosis was significantly higher than that of experienced spine physicians. We believe our algorithm, which uses different diagnostic criteria than humans, can significantly improve the diagnostic accuracy of cOPLL when radiography is used.