Keywords:
Artificial Intelligence, Musculoskeletal spine, Digital radiography, CAD, CT, Computer Applications-Detection, diagnosis, Segmentation, Calcifications / Calculi
Authors:
H.-D. Chae, S. H. Hong, H. E. Shin, Y. Lee, M. Park, J. Bae, J.-Y. Choi, H. J. Yoo; Seoul/KR
DOI:
10.26044/ecr2019/C-3569
Aims and objectives
Ossification of the posterior longitudinal ligament (OPLL) of the cervical spine is one of the most common causes of cervical myelopathy in Eastern Asia including Korea and Japan (1).
Cervical radiography is usually performed as an initial diagnostic test for the assessment of symptoms from OPLL.
On the conventional lateral radiography,
OPLL appears as continuous or segmental ossifications posterior to the vertebral bodies along the course of the posterior longitudinal ligament (PLL).
However,
many instances of cervical OPLL can be missed on lateral plain radiographs due to the overlap of facet joints and bony spurs with ossification area.
The overall false negative rate of cervical OPLL in lateral radiograph was reported to be about 48% (2),
and the interobserver variability showed only fair agreement in the lateral radiograph-based classification of cervical OPLL (3).
Computed tomography (CT) can most accurately delineate OPLL and is the diagnostic modality of choice for preoperative planning.
CT can detect small lesions that are indistinct on simple radiographs and can precisely assess the thickness and length of the OPLL.
However,
CT is not routinely performed as an initial evaluation in daily practice,
and there is a limitation to use as a follow-up test repeatedly due to the high radiation dose.
Therefore,
radiographs of the cervical spine play an essential role in the evaluation and diagnosis of cervical OPLL.
Thus,
even in the era of cross-sectional imaging,
plain radiography plays a fundamental role in the diagnosis and assessment of cervical OPLL.
Recent advances in artificial intelligence have shown promising results in various fields of radiology for the detection and characterization of lesions (4).
Several early results showed that deep-learning models outperform radiologists in the detection of pulmonary nodules on chest radiographs (5),
and improve the diagnostic performance of radiologists in detecting breast cancer at mammography (6).
Therefore,
the purpose of this study was to evaluate the diagnostic performance of the deep-learning model for the diagnosis of cervical OPLL using plain radiographs.