A machine learning based prognostic prediction of cervical myelopathy using diffusion tensor imaging

Abstract

Diffusion Tensor imaging (DTI), composing of various metrics, including fractional anisotropy (FA), axial diffusivity (AD), mean diffusivity (MD) and radial diffusivity (RD) has been considered as a useful clinical tool to reveal microstructure of spinal cord. Previous studies have intensively applied DTI in investigating the pathology of cervical spondylotic myelopathy (CSM), as well the symptomatic level diagnosis of CSM. However, it still remains unclear whether the DTI metric could be used in the prognosis of CSM, which is of great significance for selection of the best treatment strategy. Thus, the present study attempted to establish a prognosis model of CSM based on DTI metrics using machine learning methods. Particularly, three conventional machine learning algorithms, Naive Bayesian, Least Squares Support Vector Machine (LS-SVM), and Multi-label K-nearest Neighbour (ML-KNN) were tested on DTI data from 35 CSM patients accepting surgery treatments with post-operative outcomes followed. The results showed that prognosis of CSM with DTI metrics using LS-SVM algorithms could achieve higher prediction performance, with accuracy of 88.62%, and the learning curve of LS-SVM showed that the performance would be significantly improved if the sample size is greater than 202, indicating the potential application of the prognosis prediction of CSM from DTI metrics using machine learning algorithms.