Semi-automated detection of cervical spinal cord compression with the Spinal Cord Toolbox
| Authors | |
|---|---|
| Year of publication | 2022 |
| Type | Article in Periodical |
| Magazine / Source | QUANTITATIVE IMAGING IN MEDICINE AND SURGERY |
| MU Faculty or unit | |
| Citation | |
| Web | https://qims.amegroups.com/article/view/88416/html |
| Doi | http://dx.doi.org/10.21037/qims-21-782 |
| Keywords | Spinal cord compression (SCC); cervical spinal cord; myelopathy; magnetic resonance imaging (MRI); reproducibility |
| Description | Background: Degenerative cervical spinal cord compression is becoming increasingly prevalent, yet the MRI criteria that define compression are vague, and vary between studies. This contribution addresses the detection of compression by means of the Spinal Cord Toolbox and assesses the variability of the morphometric parameters extracted with it. Methods: Prospective cross-sectional study. Two types of MRI examination, 3 and 1.5 T, were performed on 66 healthy controls and 118 participants with cervical spinal cord compression. Morphometric parameters from 3T MRI obtained by Spinal Cord Toolbox (cross-sectional area, solidity, compressive ratio, torsion) were combined in multivariate logistic regression models with the outcome (binary dependent variable) being the presence of compression determined by two radiologists. Inter-trial (between 3 and 1.5 T) and inter-rater (three expert raters and SCT) variability of morphometric parameters were assessed in a subset of 35 controls and 30 participants with compression. Results: The logistic model combining compressive ratio, cross-sectional area, solidity, torsion and one binary indicator, whether or not the compression was set at level C6/7, demonstrated outstanding compression detection (area under curve =0.947). The single best cut-off for predicted probability calculated using a multiple regression equation was 0.451, with a sensitivity of 87.3% and a specificity of 90.2%. The inter-trial variability was better in Spinal Cord Toolbox (intraclass correlation coefficient was 0.858 for compressive ratio and 0.735 for cross-sectional area) compared to expert raters (mean coefficient for three expert raters was 0.722 for compressive ratio and 0.486 for cross-sectional area). The analysis of interrater variability demonstrated general agreement between SCT and three expert raters, as the correlations between SCT and raters were generally similar to those of the raters between one another. Conclusions: This study demonstrates successful semi-automated compression detection based on four parameters. The inter-trial variability of parameters established through two MRI examinations was conclusively better for Spinal Cord Toolbox compared with that of three experts' manual ratings. |
| Related projects: |
|