Epilepsy & Behavior 37 (2014) 116–122

Contents lists available at ScienceDirect

Epilepsy & Behavior journal homepage: www.elsevier.com/locate/yebeh

Reliability and variability of diffusion tensor imaging (DTI) tractography in pediatric epilepsy Helen L. Carlson a,b,⁎, Christianne Laliberté a, Brian L. Brooks a,b,c, Jacquie Hodge a,c, Adam Kirton a,b,c, Luis Bello-Espinosa a,b,c, Walter Hader a,b,c,d, Elisabeth M.S. Sherman a,b,c,e a

Alberta Children's Hospital, 2888 Shaganappi Tr NW, Calgary, AB T3B 6A8, Canada Alberta Children's Hospital Research Institute (ACHRI), Room 293, Heritage Medical Research Building, 3330 Hospital Drive NW, Calgary, AB T2N 4N1, Canada University of Calgary, 2500 University Dr. NW, Calgary, AB T2N 1N4, Canada d Foothills Medical Centre, 1403 29 Street NW, Calgary, AB T2N 2T9, Canada e Copeman Healthcare Centre, 400-628 12 Avenue SW, Calgary, AB T2R 0H6, Canada b c

a r t i c l e

i n f o

Article history: Received 9 May 2014 Revised 10 June 2014 Accepted 12 June 2014 Available online xxxx Keywords: Tractography Diffusion tensor imaging DTI Reliability Pediatric Epilepsy

a b s t r a c t Background: Diffusion tensor imaging (DTI) tractography is useful for isolating white matter (WM) trajectories and exploring microstructural integrity. Tractography can be performed on atypical brain anatomy when landmarks are malformed or displaced but has been criticized for its subjectivity even when investigators have advanced anatomical knowledge. Also, little is known about the variability and reliability of tractography as a tool for assessing white matter damage in clinical populations such as children with pediatric epilepsy. Methods: Children diagnosed with epilepsy [N = 43, mean age = 11.7 years, standard deviation = 3.7 years, 53% male] underwent a DTI sequence (6 directions, 2 × 2 × 3 mm voxels). Tractography for six white matter tracts (anterior forceps, fornices, bilateral arcuate fasciculi, and bilateral anterior cingula) was conducted twice by two experienced tractographers. Percent coefficient of variation (CV; for measuring variability) and intraclass correlation coefficients (ICCs; for measuring reliability) were calculated for tract volume and diffusion variables (fractional anisotropy [FA], mean diffusivity [MD], axial diffusivity [AD] and radial diffusivity [RD]). Results: Diffusion variables showed low variability (CV = 2.7–8.8%) and very high reliability (ICC = .97–.99) except for limbic tracts [fornix (ICC = .75–.94); cingulum (ICC = .71–.98)]. Tract volume measurements showed high variability (CV = 21.9–62.0%) and moderate reliability (ICC = .54–.99). Overall, tract volume measurements were much more variable and less reliable than diffusion characteristics. Limbic structures showed more variability compared with others. Conclusions: This suggests that DTI tractography and resulting diffusivity variables can reliably inform on the integrity of WM structures in a clinical sample with pediatric epilepsy and highlights the importance of reporting reliability information in studies that aim to answer clinical questions about WM integrity. © 2014 Elsevier Inc. All rights reserved.

1. Introduction The development of diffusion tensor imaging (DTI) in the field of magnetic resonance imaging (MRI) has afforded an elegant, noninvasive way to segregate connective white matter (WM) from gray matter (GM) within the human brain [1–3]. By precisely modulating the Abbreviations: DTI, diffusion tensor imaging; WM, white matter; MRI, magnetic resonance imaging; EEG, electroencephalography; CV, coefficient of variation; ICC, intraclass correlation coefficient; FA, fractional anisotropy; MD, mean diffusivity; AD, axial diffusivity; RD, radial diffusivity; ROI, region of interest; SD, standard deviation; TE, echo time; TR, repetition time; FOV, field of view; FACT, fiber assignment by continuous tracking; AED, antiepileptic drug. ⁎ Corresponding author at: Neuropsychology, Neurosciences, Alberta Children's Hospital, 2888 Shaganappi Trail NW, Calgary, AB T3B 6A8, Canada. Tel.: +1 403 955 7360. E-mail address: [email protected] (H.L. Carlson).

http://dx.doi.org/10.1016/j.yebeh.2014.06.020 1525-5050/© 2014 Elsevier Inc. All rights reserved.

directional magnetic fields during a diffusion-weighted MRI sequence, the diffusion direction of water molecules through different tissues can be measured. Isotropic diffusion (i.e., motion equal in all directions) is indicative of a space with few anatomical barriers such as the ventricles. Anisotropic motion (i.e., motion in a primary axis) is indicative of constrained diffusion as would be seen in the axonal white matter. The preferential selection of anisotropic water diffusion results in an indirect representation of the microstructure and macrostructure of the underlying white matter neuroanatomy that can be mapped across space in three dimensions. Advanced tractography algorithms are used to isolate specific white matter tracts for study by connecting fibers that pass through anatomically placed regions of interest (ROIs). Once these tracts have been isolated, measurements of diffusivity and anisotropy can be taken to investigate tract integrity [2–7]. Given this rich white matter map, researchers can visualize in vivo the different

H.L. Carlson et al. / Epilepsy & Behavior 37 (2014) 116–122

white matter tracts coursing through the brain. Studies investigating the efficacy of delineating structures through the use of tractography have demonstrated good correspondence with previous histological studies of white matter anatomy [8–11]. Diffusion tensor imaging tractography is particularly useful when isolating the trajectories of white matter pathways and measuring the integrity of those pathways in atypical brains as found in many clinical populations [see [12]]. A major advantage of manual tractography, compared with automated algorithms, is that it can still be performed on atypical brain anatomy using well-known landmarks to locate seed points even when such landmarks are displaced from their typical locations or are malformed. Manual tractography, however, has been criticized for its subjectivity, given that seed regions of interest are manually demarcated by tractographers [13], as well as its time-consuming nature and reliance on specialized anatomical knowledge [10]. An important step in investigating the efficacy of a tool is to systematically address the reliability of the measurements both within and between raters. Indeed, if a tool is not reliable, then it cannot be valid. Unfortunately, few studies have systematically analyzed rater reliability of tractography. While some post hoc reliability results are reported for tractography in healthy subjects, information is very limited about the within-rater and between-raters reliabilities of different white matter structures and diffusion variables in both healthy and clinical populations. Information on the extent of variability within each measure is also limited; the more variable a measure is, the less effective it is in quantifying underlying integrity. Such information is crucial in replicating studies and evaluating tractography as a potential clinical tool to assess white matter damage. Despite the general paucity of prospective, systematic reliability studies, a few have demonstrated excellent within-rater and betweenraters reliabilities in adult healthy populations [14–16], patients with epilepsy [17], depression [18], peripheral nerve tumors [19], Alzheimer's disease [20], and stroke [21,22]. Typically, in these adult studies, variability within a measure [as represented by coefficient of variance [CV]; [23]] is very low (CV ≤ 10%), and both within-rater and between-raters concordances [as measured by intraclass correlation coefficients [ICCs]; [24]] are very high (ICC ≥ 0.9) for most diffusivity and fractional anisotropy (FA) values. Measurements of tract volume, however, tend to have higher variability and lower reliability values compared with diffusivity and FA [14,18,25,26]. Taken together, these results would seem to suggest that diffusion variables resulting from DTI tractography are reliable measures for investigating the integrity of white matter tracts and may not be as subjective as previously mentioned [13]; however, volume measurements are more variable. Previous work investigating a healthy pediatric sample has also demonstrated low variability and excellent within-rater and betweenraters reliability for six WM structures [27]. Developmental studies encompassing large age ranges (4 months to 24 years) also demonstrate high reproducibility of some tracts but only “moderate” reproducibility of others [28]. Reliability of tractography has not been previously reported in a sample with pediatric epilepsy although good agreement has been reported for diffusion characteristics of epileptogenic normal-appearing WM adjacent to cortical tubers in a pediatric tuberous sclerosis sample [29]. Clearly, additional information is needed regarding measurement variability and both within-rater and betweenraters reliability in pediatric patient groups. The current prospective study aimed to systematically investigate measurement variability and within-rater and between-raters reliabilities among tractographers for WM structures typically investigated in pediatric epilepsy. To that end, diffusion variables (fractional anisotropy [FA], mean diffusivity [MD], axial diffusivity [AD], and radial diffusivity [RD] as well as tract volume) were measured in limbic system structures (fornices and anterior cingula), a frontal lobe structure (anterior forceps), and the frontoparietal language areas (arcuate fasciculi) in a sample with pediatric epilepsy.

117

2. Materials and methods 2.1. Participants Informed consent was obtained from a parent or guardian of all children before admission into the study. The research protocol was approved by the Conjoint Health Research Ethics Board of the Faculty of Medicine at the University of Calgary. Participants were recruited from the pediatric epilepsy clinic at the Alberta Children's Hospital in Calgary, Alberta. Participants were 4–18 years old with a primary diagnosis of epilepsy as determined by the treating neurologist. Seizure laterality and/or focus were determined by electroencephalography (EEG). No participants had any other progressive neurological condition, MRI contraindications (as determined by the radiologist), or prior history of epilepsy surgery or other neurosurgeries. 2.2. Image acquisition All patients underwent clinical MRI with an established clinical epilepsy protocol. This included a diffusion-weighted sequence in which 40–45 contiguous slices (affording full brain coverage) were acquired in the axial plane on a Siemens Avanto 1.5-T scanner (Siemens Medical Systems, Erlangen, Germany) using a 32-channel head coil. Six diffusion gradient directions were acquired using a b-value of 1000 s/mm2 and one acquisition using a b-value of 0 s/mm2. The voxel size was 2.0 × 2.0 mm, slices were 3.0 mm thick, and the sequence was 4.2 min in total duration which included 6 repetitions [TE = 90, TR = 6500, FOV = 256 × 256, matrix = 128 × 128, no interpolation]. From these images, diffusion tensors and eigenvalue and eigenvector maps were calculated using DTI Studio [www.mristudio.org; [30]]. 2.3. Tractography Tracts of interest were extracted by manually drawing multiple regions of interest (ROIs) at anatomically guided locations on the color-coded fractional anisotropy (FA) maps (color maps). The fiber assignment by continuous tracking (FACT) algorithm (a deterministic technique) was used to connect voxels containing fibers over a threshold FA value of 0.20 and those with similar orientation (b70° turning angle). A minimum FA value of 0.20 was used to maximize the inclusion of white matter fibers while excluding gray matter (FA ~ 0.15). This minimum FA criterion is reasonable for a pediatric population, given previous evidence that the average FA of these structures ranges from approximately 0.35 to 0.60 in healthy children aged between 5 and 18 [31]. Multiple ROIs were combined using a logical “OR” rule. This rule allowed the inclusion of all fibers that projected through any one of the ROIs to be included in the initial reconstructed tract. Subsequent “NOT” ROIs were drawn to exclude fibers that either were part of other known fiber bundles or were spurious fibers. Six white matter tracts (anterior forceps, fornices, anterior cingula [left and right], and arcuate fasciculi [left and right]) for each child were isolated twice by two experienced tractographers on different occasions. Tractography was conducted such that the time interval between measurements 1 and 2 for a given tract was at least 24 h to minimize the chances of biasing region of interest placement [mean time interval between measurements 1 and 2 = 51.23 days, SD = 71.55 days]. The locations of ROI placements in relation to anatomical structures were agreed upon a priori by the two tractographers based on anatomical landmarks, previously published techniques [5,18,32], and white matter atlases [33,34]. This was done to approximate an independent experimenter attempting to replicate a published study by using the description of ROI placement in the Methods section.

118

H.L. Carlson et al. / Epilepsy & Behavior 37 (2014) 116–122

2.3.1. Anterior forceps To isolate the anterior forceps, one ROI was drawn on the sagittal slice corresponding to the midline of the patient (Fig. 1A). The ROI enclosed the entire anterior portion of the genu when seen in full profile and included fibers projecting into the frontal white matter in both hemispheres. The posterior edge of the ROI was determined by drawing a vertical line from the posterior edge of the genu to ensure that the fibers selected were from the medial prefrontal regions [18,32,33].

2.3.2. Fornix Bilateral fornices were isolated by drawing five ROIs (Fig. 1B) combined by logical “OR” operators on three axial slices of the color map [5]. The first ROI was drawn at the level of the superior portion of the body of the fornix before it divides into the two hemispheres. The next two ROIs were drawn around the crura of the fornices as they project in the vertical plane just anterior to the splenium of the corpus callosum. The remaining two ROIs were drawn on the superior surface of each hippocampus to capture the inferior portions of the fornices.

2.3.3. Arcuate fasciculus The left and right arcuate fasciculi were individually isolated using two ROIs each (Fig. 1C). The first ROI was delineated using a halfmoon-shaped region on the axial slice at the most dorsal portion of the arcuate [33]. The second ROI was outlined on the axial slice around the small, blue, triangular-shaped region lateral to the splenium of the corpus callosum. This ROI marked the posterior portion of each arcuate where it projects into the temporal lobe.

A.

B.

C.

D.

Fig. 1. Region of interest placement and representative tracts. A. Anterior forceps were selected using one ROI on the sagittal slice at the midline of the patient. B. Bilateral fornices were selected by using five ROIs on three axial slices. C. Left and right arcuate fasciculi were selected by using two ROIs on axial slices. D. Anterior cingula were selected using a 3-dimensional ROI drawn on successive axial slices. For more information on the placement of ROIs, refer to the Materials and methods section.

2.3.4. Anterior cingulum The left and right cingula were individually isolated using a 3dimensional ROI approach (Fig. 1D). The first ROI was drawn on the most inferior axial slice where the cingulum is visible immediately anterior to the genu of the corpus callosum. The tract was then followed dorsally through the axial slices where one ROI was drawn per slice as the tract changed from blue (dorsal–ventral) to green (anterior–posterior). The final ROI was placed on the most dorsal portion of the cingulum, as indicated by Catani and Thiebault de Schotten [33] as “a single cigarshaped region” which runs in the anterior–posterior direction. A limit plane was placed to delineate the anterior portion of the cingula, and only the portion of the tracts lying anterior to this plane was included in the analysis. This vertical plane was placed on the coronal slice at the posterior edge of the genu at midline. The positioning of the limit plane is illustrated by the white line just posterior to the genu on an axial slice in Fig. 1D [18]. 2.4. Calculation of diffusion variables and tract volume Once the tracts were individually isolated, customized MATLAB scripts were used to calculate FA, MD, RD, and AD, ensuring that voxels containing more than one streamline were only used once [5]. Tract volume was approximated as the number of voxels that have at least one tract passing through them multiplied by the volume of a single voxel (12 mm3). 2.5. Statistical analyses Microsoft Excel and the Statistical Package for the Social Sciences (IBM SPSS; version 19 for Windows) were used for statistical calculations. Means and standard deviations were calculated for each structure and variable. Normality assumptions were evaluated using the Kolmogorov–Smirnov test of normality. Duration of epilepsy was compared with white matter integrity variables using Pearson's partial correlation controlling for age at testing. To illustrate the relative variability of each measurement, percent coefficients of variation [CV; [23]] were calculated for each tract variable and structure. This technique allows the comparison of overall variability within a measure to be compared with the variability of another measure that does not necessarily share the same mean. Coefficient of variation was calculated by taking the standard deviation of the group of scores for a given variable (e.g., FA) and dividing it by the mean of that same group of scores. Results were then multiplied by 100 to give a percentage coefficient of variation (CV = [SD / mean] ∗ 100). Coefficients of variation of ≤10% were considered acceptable and indicated that the dependent variable had a relatively small amount of variability [35]. Coefficients of variation between 11% and 20% were considered adequate and indicated a moderate amount of variability. Coefficients of variation ≥ 21% were considered highly variable. Intraclass correlation coefficients [ICCs; [24,35]] were also calculated for each dependent variable to measure within-rater and betweenraters reliabilities. Intraclass correlation coefficient was calculated between measurements 1 and 2 for a tractographer's results (within-rater reliability) and for measurement 1 between the two tractographers' results (between-raters reliability). In both cases, a two-way mixed model was used to determine absolute agreement between the two measurements. Average measures of ICCs were reported and values of ≥0.80 were taken to indicate reproducible results. Intraclass correlation coefficient values between 0.79 and 0.70 were considered adequate for an in vivo imaging study [35]. Values less than 0.69 were considered to be of limited reproducibility. Intraclass correlation coefficient is similar to the Pearson r correlation; however, ICC has the advantage of comparing the homogeneity and the variance between large sets of measurements rather than just their linear relationships [24].

H.L. Carlson et al. / Epilepsy & Behavior 37 (2014) 116–122

3. Results The final sample consisted of 43 children diagnosed with epilepsy (53% male). Children had a mean age of 11.7 years (standard deviation [SD] = 3.7 years). Additional demographic and clinical variables of the sample are summarized in Table 1. Means, SDs, and the resulting percent coefficients of variance (%CV) as well as within-rater and between-raters reliabilities (ICC) are reported in Table 2. Boxplots illustrating the variance of each variable and structure are plotted in Fig. 2. All distributions were normally distributed with the exception of right cingulum volume [D(43) = 0.16, p = 0.01]. Axial diffusivity (AD) was predictably higher than radial diffusivity (RD) for all structures as would be expected in myelinated white matter tracts [4]. Fractional anisotropy and diffusivity values varied across white matter tracts (Fig. 2). Generally, diffusivity variables (FA, MD, RD, and AD) showed very low variability for all structures (Table 2, columns 4 and 7) and fell within the acceptable range [CV b 10%; [35]]. Tract volume, however, showed a high degree of variability for all tracts falling into the highly variable range (CV ≥ 21%). Of note, despite generally high within-rater reliability for most structures, within-rater ICCs were more variable for the fornix and the cingulum (ICC range = .71–.99). Between-raters reliability showed the same pattern for the fornix and the cingulum (ICC range = .49–.99). Intraclass correlation coefficients both within-rater and between-raters were also generally lower for tract volume (ICC range = .48–.99). Duration of epilepsy (mean [SD] = 5.0[4.2] years) was not significantly related to FA or any diffusivity variables (p N .05) when controlling for age.

4. Discussion The goal of this study was to systematically investigate measurement variability and within-rater and between-raters reliability of five different DTI tractography variables that are typically used to

Table 1 Patient demographics and clinical descriptors. Category

Mean (SD) [%]

Age Sex

11.7 years (3.7) Male, n = 23 [53.4%] Female, n = 20 [46.6%] 1.5 (0.7) 1.1 (1.48) 6.8 years (4.3) 5.0 years (4.2)

Number of current AEDs Number of failed AEDs Age at epilepsy onset Duration of epilepsy Side of seizure focus (EEG) [%] Right Left Bilateral/generalized EEG normal Not specified MRI normal [%] MRI abnormal [%] Mesial temporal sclerosis (MTS) Focal cortical dysplasia (FCD) Lesion(s) Tuberous sclerosis (TS) Atrophy Nonspecific WM changes Encephalomalacia Dominant hand [%] Right Left Not specified

13 [30.2%] 15 [34.9%] 10 [23.3%] 3 [7.0%] 2 [4.7%] 25 [58.1%] 18 [41.9%] 6 [14.0%] 4 [9.3%] 4 [9.3%] 1 [2.3%] 1 [2.3%] 1 [2.3%] 1 [2.3%] 32 [74.4%] 3 [7.0%] 8 [18.6%]

SD = standard deviation; AED = antiepileptic drug; EEG = electroencephalography; MRI = magnetic resonance imaging; WM = white matter.

119

measure white matter integrity in pediatric epilepsy. This study indicates that diffusion variables (FA, MD, AD, and RD) appear to have low measurement variability. Indeed, all diffusivity variables for all structures investigated had measurement variability (CV) well within the acceptable range [b10%; [35]]. This pattern also held true for within-rater reliabilities of diffusion variables, indicating that all structures had an ICC within the adequate range and many showing ICCs in the excellent range [35]. Reliability between raters was generally found to be high, given that most ICC values were in the acceptable range; however, an exception to this trend was for the fornix and the cingulum, which displayed lower ICCs than for other structures. The results indicate that diffusivity variables are reliable estimates of white matter integrity. In addition, more than one diffusivity measurement of white matter integrity should be considered since they represent different information about the health of the underlying axonal bundles and, taken together, give a more complete picture [4]. We previously noted that reliability appeared to be structurespecific. Limbic tracts had poorer reliability and higher variability compared with other structures examined. Adult patients with epilepsy have been shown to have reduced FA and increased mean diffusivity in the limbic circuits, a finding demonstrated specifically in the fornix and the cingulum [6,36]. It has been proposed that these changes are caused by damage to the white matter resulting from excessive electrical activity and Wallerian degeneration extending from the epileptogenic site that is often found in the temporal lobe [36]. Similar changes in WM have been found in samples with pediatric epilepsy [37,38]. Whether the process of Wallerian degeneration may explain the more variable tractography results for the fornix and the cingulum in our sample with pediatric epilepsy remains to be studied in future research. The absence of a healthy control group in this study precludes conclusions about specific effects of disease. Longitudinal studies examining white matter changes in children with epilepsy (not eligible for epilepsy surgery) would be of benefit to investigate whether epileptic activity in pediatric brains affects WM over time compared with healthy children. Interestingly, structures removed from the limbic system such as the arcuate fasciculus and the anterior forceps showed quite low variability in our sample. It is also possible that the variability differences across WM structures are due to proximity to cerebral spinal fluid (CSF) and the partial volume effect [3,39]. For example, the anterior forceps and arcuate fasciculi may be less variable than the fornices because the fornix–fimbria are located in close proximity to the CSF in the ventricles. The increased variability may, therefore, be due to the partial volume effect [40] in which a given voxel near the edge of a structure contains both CSF (FA ~ 0) and white matter (FA ~ .40) and the resulting fractional anisotropy is an average of these values. The partial volume effect essentially causes an underestimation of the signal within these structures [3,39] and can also occur with smaller structures surrounded by gray matter as may be the case for the cingulum [39,41]. Using a fluid-attenuated inversion recovery (FLAIR) sequence during image acquisition has been shown to significantly decrease the signal intensity contamination in limbic structures from the CSF and, in turn, may reduce the partial volume effect [5]. Moreover, using a higher resolution scan with smaller voxels would also reduce the partial volume effect. Given that tract volume measurements for all structures were found to be much more variable and, therefore, less reliable than other measures, it seems that it is of less utility. This pattern is similar across the six structures investigated here and is consistent with other reliability studies in healthy adults [14–16,18,25,26,41–43] and in young adults with major depressive disorder [18]. Vos and colleagues [39] have even suggested that tract volume should be included as a nuisance covariate in models addressing WM diffusivity, especially in pediatric samples investigating age-related changes in microstructure. Our results are consistent with this in light of the large amount of variability in tract volume within our sample.

120

H.L. Carlson et al. / Epilepsy & Behavior 37 (2014) 116–122

Table 2 Tract integrity and volume measurements for six white matter structures and associated measurement variability, within-rater and between-raters reliability. Structures

Rater 1

Measurements

Mean

(SD)

[%CV]

Mean

(SD)

[%CV]

Rater 1

Rater 2

Rater 1 vs rater 2

.51 8.5 13.7 5.9 19,185

(.02) (.45) (.43) (.49) (5214)

[4.7] [5.3] [3.2] [8.3] [27.2]*

.50 8.5 13.7 5.9 20,756

(.02) (.34) (.37) (.36) (5450)

[4.4] [4.0] [2.7] [6.1] [26.3]*

.99 .99 .99 .99 .96

.99 .99 .99 .99 .99

.98 .90 .89 .91 .94

.42 11.5 16.8 8.8 7557

(.02) (.75) (.97) (.68) (1804)

[4.8] [6.6] [5.8] [7.8] [23.9]*

.41 11.1 16.2 8.6 8931

(.02) (.64) (.81) (.58) (1954)

[3.9] [5.8] [5.0] [6.8] [21.9]*

.94 .82 .75 .87 .93

.93 .91 .86 .94 .69*

.66* .78 .69* .83 .64*

.44 8.3 12.5 6.2 26,391

(.02) (.34) (.38) (.36) (7020)

[5.0] [4.1] [3.0] [5.8] [26.6]*

.44 8.3 12.4 6.2 20,680

(.02) (.36) (.37) (.38) (6123)

[5.2] [4.3] [3.0] [6.1] [29.6]*

.99 .99 .98 .99 .69*

.99 .99 .97 .99 .92

.98 .99 .96 .99 .66*

.45 8.2 12.4 6.1 26,281

(.02) (.32) (.34) (.33) (7819)

[4.3] [3.9] [2.8] [5.4] [29.8]*

.45 8.2 12.4 6.1 22,892

(.02) (.32) (.34) (.34) (6525)

[4.9] [3.9] [2.8] [5.6] [28.5]*

.98 .99 .98 .99 .74

.99 .99 .99 .99 .92

.97 .99 .96 .99 .83

.44 8.5 12.9 6.3 883

(.03) (.34) (.52) (.38) (29)

[6.8] [4.0] [4.1] [6.0] [32.7]*

.42 8.6 12.7 6.6 1236

(.04) (.35) (.56) (.42) (734)

[8.4] [4.1] [4.4] [6.4] [59.4]*

.90 .98 .96 .96 .89

.89 .96 .94 .93 .94

.49* .88 .82 .71 .54*

.41 8.6 12.5 6.6 910

(.04) (.37) (.54) (.44) (398)

[8.8] [4.3] [4.3] [6.7] [43.7]*

.38 8.7 12.3 6.8 1312

(.03) (.33) (.46) (.35) (814)

[6.6] [3.8] [3.7] [5.1] [62.0]*

.94 .92 .95 .91 .89

.71 .91 .85 .88 .90

.48* .88 .68* .80 .74

Anterior forceps FA MD AD RD Volume Fornix FA MD AD RD Volume Arcuate fasciculus, left FA MD AD RD Volume Arcuate fasciculus, right FA MD AD RD Volume Cingulum, left FA MD AD RD Volume Cingulum, right FA MD AD RD Volume

Rater 2

Within (ICC)

Between (ICC)

FA = fractional anisotropy; MD = mean diffusivity (× 10−4 mm2/s); AD = axial diffusivity (× 10−4 mm2/s); RD = radial diffusivity (× 10−4 mm2/s); volume units = mm3; %CV = percent coefficient of variation. %CVs of ≤10% of the mean were considered acceptable; %CVs between 11% and 20% were considered adequate; %CVs ≥21% were considered highly variable (denoted with *). ICC = intraclass correlation coefficient. Within-rater reliability was calculated using measurement 1 vs measurement 2 by a tractographer. Betweenraters reliability was calculated using measurement 1 between two raters. ICCs ≥ 0.80 were considered to indicate reproducible results; ICCs between 0.79 and 0.70 were considered adequate; ICCs ≤ 0.70 were considered to be of limited reproducibility (denoted with boldface and *).

Another possibility that could explain the lower reliability within the fornix and the cingulum is the fact that the DTI sequence used just six orthogonal diffusion directions. Although six diffusion directions are enough to calculate the tensor, and FA and MD values have been shown to be very similar to a 30-direction sequence [44], the fornix and the cingulum are curvilinear, and the results of the tractography algorithms may benefit from having more directions and, therefore, higher angular resolution [25,39,45]. That the current results are primarily applicable to deterministic tractography algorithms is also worthy of note. Given the fundamental underlying differences among algorithms, tractography results often differ; therefore, further systematic investigation is warranted to address reliability and variability for other methods (such as probabilistic tractography), especially in clinical populations [46,47]. There were some limitations to the current study. First, our sample included a heterogeneous group of children seen at a tertiary care center, and so this sample was not representative of patients with epilepsy in the general population. The diversity of this sample has the advantage of including both MRI-positive and MRI-negative cases as well as a range of epilepsy severity and underlying etiologies but necessarily introduces variability and, in some cases, made it challenging to map structure in the context of atypical brain morphology. Despite this, high reliability was achieved for most of the structures and variables tested. This supports the idea that DTI tractography and the resulting diffusivity variables can inform on the integrity of white matter

structures in a clinical sample with pediatric epilepsy. Volume measurements would appear to be of less utility, given their higher variability. Further, measures of variability and reliability should be reported in studies investigating white matter integrity. Second, patterns of variation across structures and among diffusion variables may be different in healthy controls than in our clinical population with epilepsy and in children with less severe epilepsy. Interestingly, our pattern of results is similar to that of existing literature [14–16,18,25,26,42]. Subsequent investigations should address variability and reliability differences over time in these populations and their relationships with disease duration and severity. 5. Conclusions We have demonstrated in a sample with pediatric epilepsy that white matter diffusivity variables (MD, AD, and RD) and fractional anisotropy (FA) measures resulting from DTI tractography have very low measurement variability and high within-rater and between-raters reliability. Tract volume measurements, however, are much more variable and, therefore, are of less utility. This study supports the idea that DTI tractography and the resulting diffusivity variables can inform on the integrity of white matter structures in a clinical sample with pediatric epilepsy. It also highlights the importance of reporting reliability information in studies that aim to answer clinical questions about white matter integrity in pediatric epilepsy using DTI.

H.L. Carlson et al. / Epilepsy & Behavior 37 (2014) 116–122

121

Fig. 2. Boxplots illustrating variability across WM structures and measurements. A. Fractional anisotropy. B. Mean diffusivity. C. Axial diffusivity. D. Radial diffusivity. E. Tract volume (in mm3). All diffusivity values are ×10−4 mm2/s. Each shaded box illustrates the range of scores falling within the interquartile range (IQR), horizontal bars represent the median, and error bars represent upper and lower 25% of scores. Outliers are indicated by circles (outlier) or stars (extreme value).

Acknowledgments This research was supported by a grant to EMS from the Alberta Children's Hospital Research Institute for Child and Maternal Health (ACHRI) (RT734407) at the University of Calgary, Alberta, Canada.

[9] [10]

[11]

Conflict of interest statement [12]

Dr. Sherman receives funding from a test publisher (Psychological Assessment Resources, Inc.) and book royalties from Oxford University Press. Dr. Brooks receives in-kind test credits from a computerized test publisher (CNS Vital Signs), funding from a test publisher (Psychological Assessment Resources, Inc.), and book royalties from Oxford University Press. All other authors have no conflict of interest to disclose.

[13]

[14]

[15]

References [16] [1] Basser PJ, Mattiello J, LeBihan D. MR diffusion tensor spectroscopy and imaging. Biophys J 1994;66:259–67. [2] Basser PJ, Pajevic S, Pierpaoli C, Duda J, Aldroubi A. In vivo fiber tractography using DT-MRI data. Magn Reson Med 2000;44:625–32. [3] Mori S, Zhang J. Principles of diffusion tensor imaging and its applications to basic neuroscience research. Neuron 2006;51:527–39. [4] Beaulieu C. The basis of anisotropic water diffusion in the nervous system — a technical review. NMR Biomed 2002;15:435–55. [5] Concha L, Gross DW, Beaulieu C. Diffusion tensor tractography of the limbic system. AJNR Am J Neuroradiol 2005;26:2267–74. [6] Concha L, Beaulieu C, Collins DL, Gross DW. White-matter diffusion abnormalities in temporal-lobe epilepsy with and without mesial temporal sclerosis. J Neurol Neurosurg Psychiatry 2009;80:312–9. [7] Basser PJ, Mattiello J, LeBihan D. Estimation of the effective self-diffusion tensor from the NMR spin echo. J Magn Reson B 1994;103:247–54. [8] Dini LI, Vedolin LM, Bertholdo D, Grando RD, Mazzola A, Dini SA, et al. Reproducibility of quantitative fiber tracking measurements in diffusion tensor imaging of frontal

[17]

[18]

[19]

[20]

[21]

lobe tracts: a protocol based on the fiber dissection technique. Surg Neurol Int 2013;4:51. Mandelstam SA. Challenges of the anatomy and diffusion tensor tractography of the Meyer loop. AJNR Am J Neuroradiol 2012;33:1204–10. Lawes IN, Barrick TR, Murugam V, Spierings N, Evans DR, Song M, et al. Atlas-based segmentation of white matter tracts of the human brain using diffusion tensor tractography and comparison with classical dissection. Neuroimage 2008;39:62–79. Concha L, Livy DJ, Beaulieu C, Wheatley BM, Gross DW. In vivo diffusion tensor imaging and histopathology of the fimbria–fornix in temporal lobe epilepsy. J Neurosci 2010;30:996–1002. Ciccarelli O, Catani M, Johansen-Berg H, Clark C, Thompson A. Diffusion-based tractography in neurological disorders: concepts, applications, and future developments. Lancet Neurol 2008;7:715–27. Zhang Y, Zhang J, Oishi K, Faria AV, Jiang H, Li X, et al. Atlas-guided tract reconstruction for automated and comprehensive examination of the white matter anatomy. Neuroimage 2010;52:1289–301. Kristo G, Leemans A, de Gelder B, Raemaekers M, Rutten GJ, Ramsey N. Reliability of the corticospinal tract and arcuate fasciculus reconstructed with DTI-based tractography: implications for clinical practice. Eur Radiol 2013;23:28–36. Vollmar C, O'Muircheartaigh J, Barker GJ, Symms MR, Thompson P, Kumari V, et al. Identical, but not the same: intra-site and inter-site reproducibility of fractional anisotropy measures on two 3.0 T scanners. Neuroimage 2010;51:1384–94. Danielian LE, Iwata NK, Thomasson DM, Floeter MK. Reliability of fiber tracking measurements in diffusion tensor imaging for longitudinal study. Neuroimage 2010;49: 1572–80. Concha L, Gross DW, Wheatley BM, Beaulieu C. Diffusion tensor imaging of timedependent axonal and myelin degradation after corpus callosotomy in epilepsy patients. Neuroimage 2006;32:1090–9. Malykhin N, Concha L, Seres P, Beaulieu C, Coupland NJ. Diffusion tensor imaging tractography and reliability analysis for limbic and paralimbic white matter tracts. Psychiatry Res 2008;164:132–42. Chhabra A, Thakkar RS, Andreisek G, Chalian M, Belzberg AJ, Blakeley J, et al. Anatomic MR imaging and functional diffusion tensor imaging of peripheral nerve tumors and tumorlike conditions. AJNR Am J Neuroradiol 2013;34:802–7. Yasmin H, Nakata Y, Aoki S, Abe O, Sato N, Nemoto K, et al. Diffusion abnormalities of the uncinate fasciculus in Alzheimer's disease: diffusion tensor tract-specific analysis using a new method to measure the core of the tract. Neuroradiology 2008;50: 293–9. Borich MR, Wadden KP, Boyd LA. Establishing the reproducibility of two approaches to quantify white matter tract integrity in stroke. Neuroimage 2012;59:2393–400.

122

H.L. Carlson et al. / Epilepsy & Behavior 37 (2014) 116–122

[22] Tang PF, Ko YH, Luo ZA, Yeh FC, Chen SH, Tseng WY. Tract-specific and region of interest analysis of corticospinal tract integrity in subcortical ischemic stroke: reliability and correlation with motor function of affected lower extremity. AJNR Am J Neuroradiol 2010;31:1023–30. [23] Bland JM, Altman DG. Statistical methods for assessing agreement between two methods of clinical measurement. Lancet 1986;1:307–10. [24] McGraw KO, Wong SP. Forming inferences about some intraclass correlation coefficients. Psychol Methods 1996;1:30–46. [25] Tensaouti F, Lahlou I, Clarisse P, Lotterie JA, Berry I. Quantitative and reproducibility study of four tractography algorithms used in clinical routine. J Magn Reson Imaging 2011;34:165–72. [26] Wang JY, Abdi H, Bakhadirov K, Diaz-Arrastia R, Devous Sr MD. A comprehensive reliability assessment of quantitative diffusion tensor tractography. Neuroimage 2012;60:1127–38. [27] Bonekamp D, Nagae LM, Degaonkar M, Matson M, Abdalla WM, Barker PB, et al. Diffusion tensor imaging in children and adolescents: reproducibility, hemispheric, and age-related differences. Neuroimage 2007;34:733–42. [28] Verhoeven JS, Sage CA, Leemans A, Van Hecke W, Callaert D, Peeters R, et al. Construction of a stereotaxic DTI atlas with full diffusion tensor information for studying white matter maturation from childhood to adolescence using tractography-based segmentations. Hum Brain Mapp 2010;31:470–86. [29] Widjaja E, Simao G, Mahmoodabadi SZ, Ochi A, Snead OC, Rutka J, et al. Diffusion tensor imaging identifies changes in normal-appearing white matter within the epileptogenic zone in tuberous sclerosis complex. Epilepsy Res 2010;89:246–53. [30] Jiang H, van Zijl PC, Kim J, Pearlson GD, Mori S. DtiStudio: resource program for diffusion tensor computation and fiber bundle tracking. Comput Methods Programs Biomed 2006;81:106–16. [31] Lebel C, Walker L, Leemans A, Phillips L, Beaulieu C. Microstructural maturation of the human brain from childhood to adulthood. Neuroimage 2008;40:1044–55. [32] Hofer S, Frahm J. Topography of the human corpus callosum revisited—comprehensive fiber tractography using diffusion tensor magnetic resonance imaging. Neuroimage 2006;32:989–94. [33] Catani M, Thiebaut de Schotten M. A diffusor tensor imaging tractography atlas for virtual in vivo dissections. Cortex 2008;44:1105–32. [34] Wakana S, Jiang H, Nagae-Poetscher LM, van Zijl PC, Mori S. Fiber tract-based atlas of human white matter anatomy. Radiology 2004;230:77–87.

[35] Marenco S, Rawlings R, Rohde GK, Barnett AS, Honea RA, Pierpaoli C, et al. Regional distribution of measurement error in diffusion tensor imaging. Psychiatry Res 2006;147:69–78. [36] Concha L, Beaulieu C, Gross DW. Bilateral limbic diffusion abnormalities in unilateral temporal lobe epilepsy. Ann Neurol 2005;57:188–96. [37] Nilsson D, Go C, Rutka JT, Rydenhag B, Mabbott DJ, Snead III OC, et al. Bilateral diffusion tensor abnormalities of temporal lobe and cingulate gyrus white matter in children with temporal lobe epilepsy. Epilepsy Res 2008;81:128–35. [38] Kimiwada T, Juhasz C, Makki M, Muzik O, Chugani DC, Asano E, et al. Hippocampal and thalamic diffusion abnormalities in children with temporal lobe epilepsy. Epilepsia 2006;47:167–75. [39] Vos SB, Jones DK, Viergever MA, Leemans A. Partial volume effect as a hidden covariate in DTI analyses. Neuroimage 2011;55:1566–76. [40] Alexander AL, Hasan KM, Lazar M, Tsuruda JS, Parker DL. Analysis of partial volume effects in diffusion-tensor MRI. Magn Reson Med 2001;45:770–80. [41] Heiervang E, Behrens TE, Mackay CE, Robson MD, Johansen-Berg H. Between session reproducibility and between subject variability of diffusion MR and tractography measures. Neuroimage 2006;33:867–77. [42] Ciccarelli O, Parker GJ, Toosy AT, Wheeler-Kingshott CA, Barker GJ, Boulby PA, et al. From diffusion tractography to quantitative white matter tract measures: a reproducibility study. Neuroimage 2003;18:348–59. [43] Wakana S, Caprihan A, Panzenboeck MM, Fallon JH, Perry M, Gollub RL, et al. Reproducibility of quantitative tractography methods applied to cerebral white matter. Neuroimage 2007;36:630–44. [44] Lebel C, Benner T, Beaulieu C. Six is enough? Comparison of diffusion parameters measured using six or more diffusion-encoding gradient directions with deterministic tractography. Magn Reson Med 2012;68:474–83. [45] Frank LR. Characterization of anisotropy in high angular resolution diffusionweighted MRI. Magn Reson Med 2002;47:1083–99. [46] Lilja Y, Ljungberg M, Starck G, Malmgren K, Rydenhag B, Nilsson DT. Visualizing Meyer's loop: a comparison of deterministic and probabilistic tractography. Epilepsy Res 2014;108:481–90. [47] Anastasopoulos C, Reisert M, Kiselev VG, Nguyen-Thanh T, Schulze-Bonhage A, Zentner J, et al. Local and global fiber tractography in patients with epilepsy. AJNR Am J Neuroradiol 2014;35:291–6.