Texture Analysis of Magnetic Resonance Images Enables Phenotyping of Potentially Painful Annular Fissures

Abstract

Study design: Retrospective analysis of prospectively collected data.

Objective: To investigate whether intervertebral disc (IVD) image features, extracted from magnetic resonance (MR) images, can depict the extension and width of annular fissures and associate them to pain.

Summary of background data: Annular fissures are suggested to be associated with low back pain (LBP). Magnetic resonance imaging (MRI) is a sensitive method, yet fissures are sometimes unobservable in T2-weighted MR-images, even though fissure information is present in the image. Image features can mathematically be calculated from MR-images and might reveal fissure characteristics.

Methods: Forty four LBP patients who underwent MRI, low-pressure discography (<50 psi), and computed tomography (CT) sequentially in 1 day, were reviewed. After semi-automated segmentation of 126 discs, image features were extracted from the T2-weighted images. The number of image features was reduced with principle component analysis (PCA). CT-discograms were graded and dichotomized regarding extension and width of fissures. IVDs were divided into fissures extending to outer annulus versus short/no fissures. Fissure width was dichotomized into narrow (<10%) versus broad fissures (>10%), and into moderately broad (10%-50%) versus very broad fissures (>50%). Logistic regression was performed to investigate if image features could depict fissure extension to outer annulus and fissure width. As a sub-analysis, the association between image features used to depict fissure characteristics and discography-provoked pain-response were investigated.

Results: Fissure extension could be depicted with sensitivity/specificity = 0.97/0.77 and area under curve (AUC) = 0.97. Corresponding results for width depiction were sensitivity/specificity = 0.94/0.39 and 0.85/0.62, and AUC = 0.86 and 0.81 for narrow versus broad and moderately broad versus very broad fissures respectively. Pain prediction with image features used for depicting fissure characteristics showed sensitivity/specificity = 0.90/0.36, 0.88/0.4, 0.93/0.33; AUC = 0.69, 0.75, and 0.73 respectively.

Conclusion: Standard MR-images contains fissure information associated to pain that can be depicted with image features, enabling non-invasive phenotyping of potentially painful annular fissures.Level of Evidence: 2.