Systematic Comparison of Reproducibility of Diffusion MRI‐Derived Free‐Water Measures Across Two Multi‐Shell Protocols
Virendra R Mishra, Karthik Sreenivasan, Dietmar Cordes- Psychiatry and Mental health
- Cellular and Molecular Neuroscience
- Geriatrics and Gerontology
- Neurology (clinical)
- Developmental Neuroscience
- Health Policy
- Epidemiology
Abstract
Background
Free‐water (FW) measures have been widely used to understand and predict the progression of various neurodegenerative disorders1,2. However, the estimation of FW in a brain voxel needs a multi‐shell acquisition to ensure reproducible fit3,4. With the availability of multi‐shell diffusion MRI (dMRI) protocol using simultaneous multislice technique from the Human Connectome Project (HCP) and Alzheimer’s Disease Neuroimaging Initiative (ADNI‐3), such protocols can now be harmonized across various centers and MRI vendors. However, limited knowledge exists on whether the results obtained for FW measures across HCP protocols are generalizable across ADNI‐3 protocols along with which protocol is more reproducible that will assist us in understanding the variability of FW measure in an individual participant longitudinally. Hence, in this study, we compared the reproducibility of FW across major white matter (WM) tracts across both HCP and ADNI‐3 protocols.
Method
A 32‐year‐old healthy male participant was scanned over five weeks, both with Basic (2mm3, b‐value = 1000s/mm2) and Advanced ADNI‐3 protocol and 2 variants of the HCP sequence (1.5mm3 and 2mm3) both with 3 b‐values (500s/mm2, 1000s/mm2, and 2500s/mm2) and 213 directions on a clinical 3T Skyra scanner. FW dMRI measures were obtained using the original Matlab implementation (Matlab)5 and DiPY implementation6 of multi‐shell dMRI acquisition (DiPY). WM atlas from Johns Hopkins University (JHU) of 20 major WM tracts pre‐registered to MNI152 was then used as a mask to obtain FW at each WM tract, and the coefficient of variance (CoV) was computed across each protocol. The total acquisition scan time was 75 minutes.
Result
FW estimated using Matlab had a mean CoV = 2.78±1.09 across all the protocols with the least CoV for Basic ADNI‐3 (CoV = 1.31±0.54) and the maximum CoV for 2mm3 HCP sequence (CoV = 4.98±1.83) (Fig.1A) while FW estimated using DiPY had a mean CoV = 6.71±2.86 across all the protocols with the least CoV for Advanced ADNI‐3 (CoV = 4.31±1.85) and the maximum CoV for 2mm3 HCP sequence (CoV = 9.66±4.15) (Fig.1A).
Conclusion
Both acquisition parameters and processing techniques significantly affect the absolute quantification of FW measures. Furthermore, our data suggest that the pooling of dMRI data across various protocols should be performed with caution.