A New Automated AI-Assisted System to Assess Cervical Disc Herniation

Su Fu, MD; Chunlin Zhang, MD; Xu Yan, MD; Dongzhe Li, MD; Yongkui Wang, MD; Chao Dong, MD; Zhengming Cao, MD; Yongming Ning, MD; Chenglong Shao, MD; Tengyue Yang, MD


Spine. 2022;47(16):E536-E544. 

In This Article

Abstract and Introduction


Study Design: An algorithm was developed with MATLAB platform to automatically quantify the volume of cervical disc herniation (CDH) based on the sagittal magnetic resonance images. This automated program was used for CDH data set, and then compared with manual measurement results confirming its reliability.

Objective: The aim was to develop a new software for automated CDH volume measurement.

Summary of Background Data: CDH compresses the spinal cord, regarding as the leading cause of cervical myelopathy. However, the CDH volume, of great value to clinical symptoms, can be only manually measured with no-excellent but acceptable interobserver reliability. This was due to the manual error of outlining CDH area and inclusion of structure posterior vertebra. No studies has proposed such an automated algorithm of CDH volume quantification which is standardised to quantify the accurate volume of CDH thus helping doctors easily evaluate CDH progressing.

Methods: The algorithm of CDH volume measurement was proposed. This program was then tested for 490 CDHs data set, from 185 patients with two repeated magnetic resonance imaging detections. Three individual observers manually measured the volumes of these CDHs, to justify the accuracy of this software. CDH volume was either in the classic way or the revised way excluding the influence of structure posterior vertebra.

Results: The automated software was successfully developed on MATLAB platform, with no difference found with manual measurements (average level) in CDH volume measurement. The change ratios in CDH volumes were profoundly consistent with manual observation, showing the error of 5.8% in median. The revised method elevated the absolute value of ratio by amplifying the percentage change.

Conclusion: Our developed automated volumetry system was an standardized and accurate way, with selective removal module of structure posterior vertebra, replaceable for manual volume measurement of CDH, which was useful for spinal surgeons diagnosing and treating CDH disease.


Degenerative cervical myelopathy, as the very common form of spinal cord impairment,[1,2] gains in popularity with the aging of population world-wide.[3] Due to disc degeneration, the fragment of disc (mainly nucleus pulposus) is pushed into the spinal canal via a rupture in the annulus, referred to cervical disc herniation (CDH),[4] compressing the spinal cord/spinal nerve roots thus leading to clinical manifestations. Revealed by magnetic resonance imaging (MRI), CDH was the only finding significantly associated with clinical symptoms.[5] Due to the three-dimension (3D) character of CDH, the volume was a proper index,[6] commonly used for the prediction of disc herniation regression/determining the efficacy of treatment.[7,8]

As a kind of unregularly projection into spinal canal, CDH was in various shapes in 3D. Basically, the classic method/gold standard for volume calculation method of irregular objects was sum value of section area multiplying scan interval, showing the most the most accurate and reliable way.[9,10] For CDH volume calculation, the most widely applied quantification was calculated as the sum value of CDH on each sagittal plane which was manually measured, multiplying the sum of scan thickness and interval-slice gap.[6–8] However, several limits lied in this method. First, affected by the imaging process of MRI exam, not a single line but blurry edge was found between two distinct tissues. Meanwhile, unlike lumbar disc herniation with a clear boundary, CDH was associated with surrounding tissues, for example, the posterior longitudinal ligament (PLL) with similar grey-scale value, especially on top/bottom boundary. We herein defined them as structures posterior disc,[11] which was previously included when calculating thus over-estimating CDH volume. These unavoidably lead to manual error of outlining the CDH sagittal area.

Recently, we presented that CDH volumes decreased postoperatively in patients who accepted microendoscopic laminoplasty.[12] During our manually monitoring of CDH volume, the interobserver reliability was all 0.7–0.8 (no-excellent but acceptable, according to DeVellis study), consistent with the previous studies,[8] showing the rudeness of manual measuring the CDH volume. Therefore, the idea to design software automatically measuring CDH volume in a certain and standardised pattern was proposed. This could erase the basis of manual error. However, most of the commercially available software for CDH analysis have limited functionality and frequently lack this volume calculation part. This present study aimed to develop new algorithm/software for CDH volume measurement and then justifying its accuracy and usefulness with the extracted data set from our previous study.