Abstract
PURPOSE: To evaluate the reliability of magnetic resonance imaging (MRI) diffusion kurtosis imaging (DKI) derived metrics from the human brain as well as the effect of spatial smoothing preprocess on measurement reliability through test-retest measurements. MATERIALS AND METHODS: A total of 23 healthy volunteers who underwent MRI DKI scanning during two sessions performed at an interval of two weeks were enrolled in the study. For the eight DKI-derived metrics, intraclass correlation coefficient (ICC) and coefficient of variation (CV) were used to assess regional and global reliability and reproducibility of based gray matter (GM) atlas in the automated anatomical labeling (AAL) and white matter (WM) atlas in JHU-White Matter-Labels. Additional comparisons were made between the reliabilities of images with and without smoothing preprocessing. RESULTS: The average ICCs of the eight DKI metrics varied from 0.45 to 0.91, CVs ranged from 1.48% to 4.60% in global GM and WM regions, and the reliability degree ranged from moderate to almost perfect, with regional variations. Kurtosis-related metrics presented inferior reliability to the tensor-related metrics. Among the eight DKI metrics, RD had the best reliability, while KFA ranked the last. The reliability of FA in WM was superior to that in GM; however, FA did not result in a good index for GM reliability assessment. Although the DKI reliability of GM regions varied, the default mode and visual areas consistently demonstrated the highest ICC reliability for a series of kurtosis-related metrics. Additionally, smoothing preprocessing increased reliability for a few metrics only in WM compared to preprocessing without smoothing (P < 0.05, FDR corrected). CONCLUSION: The DKI metrics can be reliably used as biomarkers for diffusion property measurements. The kurtosis-related metrics presented lower reliability compared to the tensor-related metrics from DKI data. Default mode-related areas exhibited higher reliability than other brain cortices in terms of implicated microstructural DKI property. Additionally, an appropriate smoothing preprocessing could improve the reliability of a few DKI metrics only in WM regions.