Spinal cord metrics derived from diffusion MRI: improvement in prognostication in cervical spondylotic myelopathy compared with conventional MRI
Justin K Zhang,Salim Yakdan,Muhammad I Kaleem,Saad Javeed,Jacob K Greenberg,Kathleen S Botterbush,Braeden Benedict,Martin Reis,Natasha Hongsermeier-Graves,Spencer Twitchell,Brandon Sherrod,Marcus S Mazur,Mark A Mahan,Andrew T Dailey,Erica F Bisson,Sheng-Kwei Song,Wilson Z Ray
DOI: https://doi.org/10.3171/2024.4.SPINE24107
2024-07-26
Abstract:Objective: A major shortcoming in optimizing care for patients with cervical spondylotic myelopathy (CSM) is the lack of robust quantitative imaging tools offered by conventional MRI. Advanced MRI modalities, such as diffusion MRI (dMRI), including diffusion tensor imaging (DTI) and diffusion basis spectrum imaging (DBSI), may help address this limitation by providing granular evaluations of spinal cord microstructure. Methods: Forty-seven patients with CSM underwent comprehensive clinical assessments and dMRI, followed by DTI and DBSI modeling. Conventional MRI metrics included 10 total qualitative and quantitative assessments of spinal cord compression in both the sagittal and axial planes. The dMRI metrics included 12 unique measures including anisotropic tensors, reflecting axonal diffusion, and isotropic tensors, describing extraaxonal diffusion. The primary outcome was the modified Japanese Orthopaedic Association (mJOA) score measured at 2 years postoperatively. Extreme gradient boosting-supervised classification algorithms were used to classify patients into disease groups and to prognosticate surgical outcomes at 2-year follow-up. Results: Forty-seven patients with CSM, including 24 (51%) with a mild mJOA score, 12 (26%) with a moderate mJOA score, and 11 (23%) with a severe mJOA score, as well as 21 control subjects were included. In the classification task, the traditional MRI metrics correctly assigned patients to healthy control versus mild CSM versus moderate/severe CSM cohorts, with an accuracy of 0.647 (95% CI 0.64-0.65). In comparison, the DTI model performed with an accuracy of 0.52 (95% CI 0.51-0.52) and the DBSI model's accuracy was 0.81 (95% CI 0.808-0.814). In the prognostication task, the traditional MRI metrics correctly predicted patients with CSM who improved at 2-year follow-up on the basis of change in mJOA, with an accuracy of 0.58 (95% CI 0.57-0.58). In comparison, the DTI model performed with an accuracy of 0.62 (95% CI 0.61-0.62) and the DBSI model had an accuracy of 0.72 (95% CI 0.718-0.73). Conclusions: Conventional MRI is a powerful tool to assess structural abnormality in CSM but is inherently limited in its ability to characterize spinal cord tissue injury. The results of this study demonstrate that advanced imaging techniques, namely DBSI-derived metrics from dMRI, provide granular assessments of spinal cord microstructure that can offer better diagnostic and prognostic utility.