Identify myelopathic cervical spinal cord using diffusion tensor image: A data-driven approach

Abstract

Diffusion tensor image (DTI) of the cervical spinal cord has been proposed to be used to identify the myelopathic level in the cervical spinal cord. Fractional anisotropy (FA) from DTI is usually used to diagnose the level of cervical spondylotic myelopathy (CSM). However, the solely use of FA value does not consider a full information of 3D multiple indices of diffusion from DTI. This study proposed to use a classification based on machine learning to extract and determine the myelopathic cord in CSM. A classification based on support tensor machine (STM) was applied on eigenvalues extracted from DTI at compressive levels of the cervical spinal cord. This is a validation study to apply STM classification in 30 patients with CSM. The benchmark of classification was the clinical level diagnosis with consensus of senior spine surgeons. The accuracy, sensitivity and specificity of the classification were evaluated in the study. Results showed the use of STM classification provided diagnostic accuracy of 89.2%, sensitivity of 71.8% and specificity of 90.1%. Using the classification based on STM, eigenvalues of DTI can be detected by computational intelligence to provide level diagnosis of CSM, which could help the surgeons to select the most appropriate surgical plan to treat CSM.