In Spine ; h5-index 57.0
STUDY DESIGN : Cross-sectional study.
OBJECTIVE : Validate the diagnostic accuracy of a deep learning algorithm for cervical cord compression due to degenerative canal stenosis on radiography.
SUMMARY OF BACKGROUND DATA : The diagnosis of degenerative cervical myelopathy (DCM) is often delayed, resulting in improper management. Screening tools for suspected DCM would help identify patients who require detailed physical evaluation.
METHODS : Data from 240 patients (120 with cervical stenosis on magnetic resonance imaging [MRI] and 120 age- and sex-matched controls) were randomly divided into training (n=198) and test (n=42) datasets. The deep learning algorithm, designed to identify the suspected stenosis level on radiography, was constructed using a convolutional neural network model called EfficientNetB2, and radiography and MRI data from the training dataset. The accuracy and area under the curve (AUC) of the receiver operating characteristic curve were calculated for the independent test dataset. Finally, the number of correct diagnoses was compared between the algorithm and 10 physicians using the test cohort.
RESULTS : The diagnostic accuracy and AUC of the deep learning algorithm were 0.81 and 0.81, respectively, in the independent test dataset. The rate of correct responses in the test dataset was significantly higher for the algorithm than for physicians' consensus (81.0% vs. 66.2%; P=0.034). Furthermore, the accuracy of the algorithm was greater than that of each individual physician.
CONCLUSION : We developed a deep learning algorithm capable of suggesting the presence of cervical spinal cord compression on cervical radiography and highlighting the suspected levels on radiographic imaging when cord compression is identified. The diagnostic accuracy of the algorithm was greater than that of spine physicians.
LEVEL OF EVIDENCE : IV.
Tamai Koji, Terai Hidetomi, Hoshino Masatoshi, Tabuchi Hitoshi, Kato Minori, Toyoda Hiromitsu, Suzuki Akinobu, Takahashi Shinji, Yabu Akito, Sawada Yuta, Iwamae Masayoshi, Oka Makoto, Nakaniwa Kazunori, Okada Mitsuhiro, Nakamura Hiroaki
2023-Feb-10
