Reliability of fiber tracking measurements in diffusion tensor imaging for longitudinal study

Abstract

The statistical reliability of diffusion property measurements was evaluated in ten healthy subjects using deterministic fiber tracking to localize tracts affected in motor neuron disease: corticospinal tract (CST), uncinate fasciculus (UNC), and the corpus callosum in its entirety (CC), and its genu (GE), motor (CCM), and splenium (SP) fibers separately. Measurements of fractional anisotropy (FA), mean diffusivity (MD), axial diffusivity (λ₁), transverse diffusivity (λ_⊥), and volume of voxels containing fibers (VV) were obtained within each tract. To assess intra-rater and inter-rater reliability, two raters carried out fiber tracking five times on each scan. Scan–rescan and longitudinal reliability were assessed in a subset of four subjects who had six scans, with two sets of three scans separated by 1 year. The statistical reliability of repeated measurements was evaluated using intraclass correlation coefficients (ICC) and coefficients of variation (CV). Spatial agreement of tract shape was assessed using the kappa (κ) statistic. Results: Repeated same-scan fiber tracking evaluations showed good geometric alignment (intra-rater κ > 0.90, inter-rater κ > 0.76) and reliable diffusion property measurements (intra-rater ICC > 0.92, inter-rater ICC > 0.77). FA, MD, and λ_⊥ were highly reliable with repeated scans on different days, up to a year apart (ICC > 0.8). VV also exhibited good reliability, but with higher CVs. We were unable to demonstrate reproducibility of λ₁. Longitudinal reliability after one year was improved by averaging measurements from multiple scans at each time point. Fiber tracking provides a reliable tool for the longitudinal evaluation of white matter diffusion properties.

Introduction

Diffusion tensor imaging (DTI) has been widely used to study the integrity of white matter tracts in a variety of neurological diseases since its introduction in 1994 (Basser et al., 1994). Most studies have assessed fractional anisotropy (FA) and mean diffusivity (MD) in regions of interest (ROIs) within white matter tracts that were identified by anatomical landmarks. More recently, deterministic fiber tracking algorithms have been used to reconstruct white matter tracts (Catani et al., 2002, Conturo et al., 1999, Wakana et al., 2004) by mapping tensors with a common direction (Mori et al., 1999) on the assumption that the primary diffusion direction is aligned parallel to fibers within the tracts (Le Bihan et al., 1986, Pierpaoli et al., 1996). For each tract that is localized by fiber tracking algorithms, values of anisotropy, diffusivity, and volume of “fibers” can be calculated over its tracked extent.

Normative data for FA and MD measurements have been established for a number of tracts and anatomical structures (Huisman et al., 2006, Hunsche et al., 2001, Lee et al., 2008). Most studies using DTI in neurological disorders have measured FA and MD. In a number of diseases, such as amyotrophic lateral sclerosis (ALS) (Ciccarelli et al., 2006, Ellis et al., 1999, Iwata et al., 2008, Sach et al., 2004, Sage et al., 2007, Yin et al., 2008), primary lateral sclerosis (PLS) (Ulug et al., 2004), frontotemporal dementia (FTD) (Matsuo et al., 2008), multiple sclerosis (Cassol et al., 2004), and stroke (Gupta et al., 2006, Moller et al., 2007), DTI has demonstrated quantifiable changes in these water diffusion properties compared to healthy controls. The decreases in FA and increases in MD that are measured in these disorders provide evidence for disruption of tissue microstructure, including axons and myelin in various white matter tracts (Basser and Pierpaoli, 1996, Pierpaoli et al., 2001). There are a relatively small number of reports on changes in other quantitative fiber tracking measures in neurological disease (Thomas et al., 2005, Wang et al., 2006, Yin et al., 2008).

In the field of motor neuron disease (MND) research, there is a need for quantitative objective markers to assess corticospinal or upper motor neuron (UMN) dysfunction and disease progression (Dengler et al., 2005, Floyd et al., 2009, Mitsumoto et al., 2007, Wang and Melhem, 2005). DTI may be a promising marker to follow longitudinal changes in white matter tracts in neurodegenerative diseases. Correlations between clinical ratings and imaging findings of FA and MD in the corticospinal tract were seen in ALS using both 2D-ROI and fiber tracking techniques (Ellis et al., 1999, Sage et al., 2007, Schimrigk et al., 2007). Additionally, in ALS, changes in FA are measurable prior to the observation of clinical signs (Sach et al., 2004). DTI has been successful in following longitudinal changes in various white matter tracts with disease progression or therapeutic interventions in multiple sclerosis and stoke (Cassol et al., 2004, Gupta et al., 2006, Liang et al., 2008, Moller et al., 2007, Reich et al., 2006). However, for longitudinal studies to be feasible in motor neuron disorders, it is imperative to establish that DTI measurements can be made reliably on individual subjects.

There are several sources of variability in DTI measures including noise, partial-volume effects, and variations in manual ROI placement techniques (Farrell et al., 2007, Ozturk et al., 2008, Schimrigk et al., 2007). In fiber tracking, in particular, sources of variability include localization of white matter tracts by human raters, as well as changes in MRI scanner characteristics and subject alignment from scan to scan (Holodny et al., 2005, Huang et al., 2004, Pierpaoli et al., 2001, Reich et al., 2006, Wakana et al., 2007). Previous studies using 2D-ROI anatomical analysis methods showed high reproducibility of evaluations of FA and MD from same-subject scans performed on separate days (Bonekamp et al., 2007, Pfefferbaum et al., 2003, Schimrigk et al., 2007). Recent studies have indicated that fiber tracking may provide more reproducible fiber tract measures (Partridge et al., 2005). Studies that performed fiber tracking on a single set of data using anatomically based ROI algorithms showed high reproducibility of repeated measurements of FA and MD both by a single rater and between multiple raters (Huang et al., 2004, Stieltjes et al., 2001, Wakana et al., 2007). To our knowledge, only two previous studies, both using probabilistic tracking methods, investigated the scan–rescan reliability of fiber tracking in various white matter tracts (Ciccarelli et al., 2003, Heiervang et al., 2006). However, the scan–rescan reliability and longitudinal variability of numerous fiber tracking measurements, particularly over longer time frames, is not well established.

The goals of this study were to assess the intra-rater and inter-rater reliability of measurements of FA and MD in several white matter tracts that are affected in MND and related disorders, such as FTD (Geser et al., 2009, Matsuo et al., 2008), using fiber tracking methods to define the tracts of interest. Intraclass correlation coefficients (ICCs) and coefficients of variation (CVs) were calculated to assess statistical reliability of repeated measurements. Spatial agreement of tract shape was evaluated with Cohen's kappa (κ). Specifically, we evaluated the corticospinal tract (CST), uncinate fasciculus (UNC), and the corpus callosum (CC). Because the CC is made up of fibers arising from several areas that may be differentially affected in MNDs, we assessed the CC in its entirety and separately assessed its genu (GE), motor fibers (CCM), and splenium (SP). We also evaluated the reliability of several diffusion measurements that have received less mention in past publications: axial diffusivity (λ₁) and transverse diffusivity (λ_⊥), which, when used cautiously, may be useful in classifying white matter degeneration (Pierpaoli et al., 2001, Wheeler-Kingshott and Cercignani, 2009), as well as the volume of voxels containing fibers (VV). Additionally, we quantified the scan–rescan and longitudinal reliability of repeated diffusion property measures for a clinically salient interval of one year.