Reproducibility of tract-based white matter microstructural measures using the ENIGMA-DTI protocol

Abstract

Background: In preparation for longitudinal analyses of white matter development in youths with family histories of substance use disorders (FH+) or without such histories (FH-), we examined the reproducibility and reliability of global and regional measures of fractional anisotropy (FA) values, measured using the Enhancing Neuro Imaging Genetics Through Meta Analysis (ENIGMA)-diffusion tensor imaging (DTI) protocol. Highly reliable measures are necessary to detect any subtle differences in brain development.

Methods: First, we analyzed reproducibility data in a sample of 12 healthy young adults (ages 20-28) imaged three times within a week. Next, we calculated the same metrics in data collected 1-year apart in the sample of 68 FH+ and 21 FH- adolescents. This is a timeframe where within subject changes in white matter microstructure are small compared to between subject variance. Reproducibility was estimated by examining mean coefficients of variation (MCV), mean absolute differences (MAD), and intraclass correlations (ICC) for global and tract-specific FA values.

Results: We found excellent reproducibility for whole-brain DTI-FA values and most of the white matter tracts, except for the corticospinal tract and the fornix in both adults and youths. There was no significant effect of FH-group on reproducibility (p = .4). Reproducibility metrics were not significantly different between adolescents and adults (all p > .2). In post hoc analyses, the reproducibility metrics for regional FA values showed a strong positive correlation (r = .6) with the regional FA heritability measures previously reported by ENIGMA-DTI.

Conclusion: Overall, this study demonstrated an excellent reproducibility of ENIGMA-DTI FA, positing it as viable analysis tools for longitudinal studies and other protocols that repeatedly assess white matter microstructure.

Keywords: ENIGMA‐DTI; diffusion tensor imaging; fractional anisotropy; reproducibility; white matter microstructure.

Figure 1

Three reproducibility measurements show the regional course of the major white matter tracts. Overall, reproducibility was excellent for most white matter regions with very low MCV and MAD values that indicate excellent reproducibility. MCV and MAD were higher as for the CST and fornix as indicated in a more intense/red color. For ICC where excellent reproducibility is reflected with a value close to 1, again reproducibility was excellent overall with ICCs close to 1. MCV, mean coefficients of variation; MAD, mean absolute difference, and ICC, intraclass correlation

Figure 2

Scatter plots of the three reproducibility parameters (MAD, MCV, and ICC) for each WM tract shown with linear regression fits ranging from r ² of 0.4–0.99. (top row). Each WM tract data point represents either MAD or MCV from both (FH+ and FH−) groups as summarized in the “All” section of Table 2. Scatter plots of the reproducibility parameters from each WM tract separated by group: youths with family histories of substance use disorders (FH+) vs. youths with no such histories (FH−; bottom row) shown with linear regression fits ranging from r ² of 0.34–0.80. Actual MCV and MAD values for each WM tract from each group (FH+ and FH−) are summarized in Table 2

Figure 3

Reproducibility parameters for adult controls (AC) were plotted vs. corresponding parameters in the adolescent cohort (FH). Linear regression analysis showed high correlation between MAD and MCV measurements for individual tracts (r = .82, p < .01). The correlation between ICC measurements was not significant (r = .33, p = .2)

Figure 4

Heritability (h ²) measurements for regional FA values calculated by ENIMGA‐DTI workgroup were plotted versus MCV and ICC reproducibility parameters in the adolescent cohort (FH). Linear regression analysis showed a significant and negative correlation between h ² and CV (r = −.53, p = .04) and a significant positive correlation with ICC (r = −.65, p = .01). The plot for h ² and MAD values was identical to that of MCV and therefore was omitted