The efficiency of retrospective artifact correction methods in improving the statistical power of between-group differences in spinal cord DTI

Gergely David,Patrick Freund,Siawoosh Mohammadi

doi:10.1016/j.neuroimage.2017.06.051

Abstract

Diffusion tensor imaging (DTI) is a promising approach for investigating the white matter microstructure of the spinal cord. However, it suffers from severe susceptibility, physiological, and instrumental artifacts present in the cord. Retrospective correction techniques are popular approaches to reduce these artifacts, because they are widely applicable and do not increase scan time.In this paper, we present a novel outlier rejection approach (reliability masking) which is designed to supplement existing correction approaches by excluding irreversibly corrupted and thus unreliable data points from the DTI index maps. Then, we investigate how chains of retrospective correction techniques including (i) registration, (ii) registration and robust fitting, and (iii) registration, robust fitting, and reliability masking affect the statistical power of a previously reported finding of lower fractional anisotropy values in the posterior column and lateral corticospinal tracts in cervical spondylotic myelopathy (CSM) patients.While established post-processing steps had small effect on the statistical power of the clinical finding (slice-wise registration: −0.5%, robust fitting: +0.6%), adding reliability masking to the post-processing chain increased it by 4.7%. Interestingly, reliability masking and registration affected the t-score metric differently: while the gain in statistical power due to reliability masking was mainly driven by decreased variability in both groups, registration slightly increased variability. In conclusion, reliability masking is particularly attractive for neuroscience and clinical research studies, as it increases statistical power by reducing group variability and thus provides a cost-efficient alternative to increasing the group size.

Highlights

Diffusion tensor imaging (DTI) is based on the acquisition of diffusion-weighted MR images (Le Bihan and Breton, 1985; Merboldt et al, 1985; Le Bihan et al, 1986) and provides information about the tissue microstructure of the central nervous system
Determining the optimal threshold for reliability masking Both factors composing the standard error of the mean of the fractional anisotropy (FA) sampling distribution in white matter (WM) decreased continuously with decreasing threshold for reliability masking (Fig. 3)
The mean of the distribution was increased for FA 340 and decreased for mean diffusivity (MD), AD, and MD, these changes were considerably smaller compared to the standard deviation

Summary

Introduction

Diffusion tensor imaging (DTI) is based on the acquisition of diffusion-weighted MR images (Le Bihan and Breton, 1985; Merboldt et al, 1985; Le Bihan et al, 1986) and provides information about the tissue microstructure of the central nervous system. DTI characterizes the magnitude, anisotropy, and orientation of the water diffusion in each voxel using a diffusion tensor model (Basser et al, 1994a,b; Pierpaoli et al, 1996). DTI indices in the spinal cord can be more readily associated with the spinal cord microstructure. Radial diffusivity in the spinal cord has been shown to most closely correlate with myelin content, while fractional anisotropy and axial diffusivity have proved to be more indicative of axonal integrity and axonal degeneration (Budde et al, 2007, 2008; Zhang, 2010; Brennan et al, 2013). Spinal cord DTI has been successfully related to various disorders with spinal cord involvement (Cohen-Adad et al, 2011; Freund et al, 2012; Grabher et al, 2016, Wheeler-Kingshott et al, 2013)

Methods

Results

Discussion

Conclusion