Elsevier

The Spine Journal

Volume 22, Issue 6, June 2022, Pages 934-940
The Spine Journal

Clinical Study
Detecting ossification of the posterior longitudinal ligament on plain radiographs using a deep convolutional neural network: a pilot study

https://doi.org/10.1016/j.spinee.2022.01.004Get rights and content

Abstract

BACKGROUND CONTEXT

Its rare prevalence and subtle radiological changes often lead to difficulties in diagnosing cervical ossification of the posterior longitudinal ligament (OPLL) on plain radiographs. However, OPLL progression may lead to trauma-induced spinal cord injury, resulting in severe paralysis. To address the difficulties in diagnosis, a deep learning approach using a convolutional neural network (CNN) was applied.

PURPOSE

The aim of our research was to evaluate the performance of a CNN model for diagnosing cervical OPLL.

STUDY DESIGN AND SETTING

Diagnostic image study.

PATIENT SAMPLE

This study included 50 patients with cervical OPLL, and 50 control patients with plain radiographs.

OUTCOME MEASURES

For the CNN model performance evaluation, we calculated the area under the receiver operating characteristic curve (AUC). We also compared the sensitivity, specificity, and accuracy of the diagnosis by the CNN with those of general orthopedic surgeons and spine specialists.

METHODS

Computed tomography was used as the gold standard for diagnosis. Radiographs of the cervical spine in neutral, flexion, and extension positions were used for training and validation of the CNN model. We used the deep learning PyTorch framework to construct the CNN architecture.

RESULTS

The accuracy of the CNN model was 90% (18/20), with a sensitivity and specificity of 80% and 100%, respectively. In contrast, the mean accuracy of orthopedic surgeons was 70%, with a sensitivity and specificity of 73% (SD: 0.12) and 67% (SD: 0.17), respectively. The mean accuracy of the spine surgeons was 75%, with a sensitivity and specificity of 80% (SD: 0.08) and 70% (SD: 0.08), respectively. The AUC of the CNN model based on the radiographs was 0.924.

CONCLUSIONS

The CNN model had successful diagnostic accuracy and sufficient specificity in the diagnosis of OPLL.

Introduction

Ossification of the posterior longitudinal ligament (OPLL) is a disease of cervical origin causing neurological impairments manifesting as sensorimotor symptoms [1]. Plain radiographs are initially used to diagnose OPLL, including functional radiographs. However, this method alone may be insufficient to achieve at a definitive diagnosis due to subtle radiological changes on plain cervical spine radiographs [2]. Thus, it is not rare to take computed tomography (CT) and magnetic resonance imaging (MRI) images to confirm the radiological diagnosis. However, delays in diagnosis of OPLL are often made by non-spine physicians when only relying on plain radiographic images. Even spine experts experience difficulties in arriving at an accurate diagnosis when using only plain radiographs. Therefore, increasing the diagnostic accuracy of plain radiographs will reduce delays in the diagnosis of OPLL. Additionally, improving the diagnostic accuracy of OPLL may reduce the risk of deterioration of the neurologic symptoms of patients with OPLL due to minor trauma, such as falls. Furthermore, increasing the diagnostic accuracy of radiographic images will lead to a decreased usage of CT or MRI, leading to reduced time, cost, and radiation exposure. Thus, an improved accuracy of the diagnosis by plain radiographs will be beneficial to institutions that have limited medical resources.

Convolutional neural networks (CNNs) are a method of machine learning recently applied in the medical field [3,4], including research using MRI images in spinal surgery field [5], [6], [7]. CNNs demonstrate a high capability for the diagnostic image analysis of plain radiographs, and can be applied in various medical fields. In patients with lung cancer, a 20% increase in accuracy was found when screening with plain chest radiographs using machine learning as compared to that with the current diagnostic standard [8]. Additionally, improvements in the diagnostic accuracy of upper and lower extremity fractures have been observed when machine learning was used [9,10].

Despite its advantages, research regarding the application of CNNs in the diagnosis of cervical spinal diseases is under explored. Therefore, this study aimed to evaluate the diagnostic performance of the CNN model based on plain radiographic images and compare it with that of general orthopedic surgeons and spine experts.

Section snippets

Study participants and input image

We retrospectively examined the medical records of patients who underwent surgical treatment of the cervical or lumbar spine at our institution between January 2016 and December 2017. This study included the records of 50 patients with symptomatic cervical OPLL who underwent surgery with thickness of ossification more than 3 mm, and 50 control patients who underwent surgery for lumbar spine diseases that manifested asymptomatic cervical spondylotic changes. Patients with the following

Patient demographics

The patient characteristics are shown in Table 1A and B. The mean age was 67.1 years (standard deviation, 10.0 years) and 61% of the patients were men.

Performance evaluation for the diagnosis by the CNN model and surgeons

The percentage of correct diagnoses made by the general orthopedic surgeons, spine specialists, and the CNN model based on neutral/flexion/extension radiographs for the 20 cases (10 OPLL and 10 control) in the test dataset are shown in Table 2. Fig. 2 shows the ROC curve of the CNN model, which had an AUC of 0.924. The CNN model had the highest

Discussion

This study aimed to evaluate the diagnostic accuracy of plain radiographs with machine learning in the diagnosis of OPLL, particularly in comparison to general orthopedic and spine surgeons. We found an excellent AUC of the diagnostic accuracy of plain radiographs, and its superior accuracy compared to that of general orthopedic and spine surgeons.

An increasing trend has been recently observed in the application of image recognition with machine learning [3,4]. These technologies have been

Declaration of competing interests

Support for this study was provided by Japan Agency for Medical Research and Development (JP21ek1126633).

Acknowledgment

Support for this study was provided by Japan Agency for Medical Research and Development (JP21ek1126633).

References (21)

There are more references available in the full text version of this article.

Cited by (4)

  • Radiomics in Spine Surgery

    2023, International Journal of Spine Surgery

FDA device/drug status: Not applicable.

Author disclosures: TO: Nothing to disclose. TY: Grant: Japan Agency for Medical Research and Development (C, Paid directly to institution). JO: Nothing to disclose. NS: Nothing to disclose. TA: Nothing to disclose. TS: Nothing to disclose. MH: Nothing to disclose. SM: Nothing to disclose. TT: Nothing to disclose. TM: Nothing to disclose. TO: Nothing to disclose. TY: Nothing to disclose. HO: Nothing to disclose. TH: Nothing to disclose. HI: Nothing to disclose. YN: Nothing to disclose. AO: Grant: Japan Agency for Medical Research and Development (C, Paid directly to institution).

Ethical review committee statement: The study protocol was approved by the institutional review boards of Tokyo Medical and Dental University. (M2020-065)

A statement of the location where the work was performed: All research work has been performed in Tokyo Medical and Dental University.

View full text