Toward Open MRI Consistency Data for fMRI and dMRI scans
Poster No:
2218
Submission Type:
Abstract Submission
Authors:
Jaemin Shin1, Brice Fernandez2, Baolian Yang3, Ethan Steingraber3, David Shin4, Jerome Maller5, Suchandrima Banerjee4, Scott Peltier6, Chris Rorden7, Flavio Dell’Acqua8
Institutions:
1GE HealthCare, New York, USA, 2GE HealthCare, Buc, France, 3GE HealthCare, Waukesha, WI, 4GE HealthCare, Menlo Park, CA, 5GE HealthCare, Hawthorn, Australia, 6University of Michigan, Ann Arbor, MI, 7University of South Carolina, Columbia, SC, 8King's College London, London, United Kingdom
First Author:
Co-Author(s):
Introduction:
With a shift towards quantitative MRI, neuroimaging initiatives, clinical trials, and longitudinal studies are increasingly employing advanced MRI techniques like functional or diffusion MRI with quantitative analysis. Concerns about data consistency between software versions often discourages sites running longitudinal studies from availing of new features and enhancements. We investigated data consistency in fMRI and dMRI scans from version 29.1 to 30.1 software upgrades on two GEHC MR systems, as a needed and essential first step in ensuring that MR systems are able to keep up-to-date with new capabilities without compromising data consistency. We introduce the Open MRI consistency data publicly available on the OSF (https://osf.io/uh2jx/), as a way to enable researchers to check data consistency between different software upgrades (e.g. from 28 to 30.1) or between different hardware systems and to contribute to the database.
Methods:
FUNSTAR fBIRN phantom (Gold Standard, Sheffield, UK) and five healthy subjects (age 19-71 years) were scanned under an IRB-approved protocol on a 3T Premier with 48 CH head coil (GE HealthCare, Waukesha, WI) and a 3T MR750 with Nova 32 CH head coil. For each software version and system, the phantom and three subjects were scanned, with test-retest performed on the phantom and 1 subject. Acquisitions include: Standard fBIRN QA protocol on the phantom [1], structural 3D T1/T2 images, rsfMRI, dMRI, B0 field map, two SE-EPI with opposite PE polarity (see details MRI protocol on the OSF page).
Resting-state fMRI data were processed using SPM12 and homemade Matlab code. EPI data was slice-time corrected, motion corrected, registered to MNI via T1 and then scaled to the mean, detrended (2nd order) and nuisances were regressed-out (aCompCor) and smoothed (6 mm). The tSNR was evaluated to compare the data quality for 11 fMRI scans after removing 5 scans due to excessive motion.
Diffusion MRI data were processed via MP-PCA denoising, removal of B1 inhomogeneities and Gibbs ringing using Mrtrix3, and then processed for eddy current and movement correction using FSL. FODs estimation, whole brain tractography and FA maps were then calculated using Mrtrix3. NODDI maps were generated using the NODDI Matlab toolbox.
Resting-state fMRI data were processed using SPM12 and homemade Matlab code. EPI data was slice-time corrected, motion corrected, registered to MNI via T1 and then scaled to the mean, detrended (2nd order) and nuisances were regressed-out (aCompCor) and smoothed (6 mm). The tSNR was evaluated to compare the data quality for 11 fMRI scans after removing 5 scans due to excessive motion.
Diffusion MRI data were processed via MP-PCA denoising, removal of B1 inhomogeneities and Gibbs ringing using Mrtrix3, and then processed for eddy current and movement correction using FSL. FODs estimation, whole brain tractography and FA maps were then calculated using Mrtrix3. NODDI maps were generated using the NODDI Matlab toolbox.
Results:
Open Data Sharing: The open MRI consistency data consists of detailed MRI protocol data, the raw DICOM data, and the BIDS data converted by dcm2niix (v1.0.20230807) and dcm2bids (v2.1.7) as well as the configuration file. We validated that DICOM to NIfTI/BIDS conversion were consistent across software versions with no issue.
Data Consistency for fMRI scan: Phantom fMRI datasets were evaluated using GE fMRI QA tool [1] available on console. The QA metrics (RMS, SFNR, SNR, RDC, Mean Ghost) were very similar between the two software releases (Fig 1a). For volunteer fMRI data, the tSNR in grey matter voxels was consistent across software release and no significant difference was found in the tSNR (Fig 1).
Data Consistency for dMRI scan: No significant difference in b0 SNR was observed between two software versions for phantom data (paired t-test: t=0.856, df=3, p=0.45) and volunteer dMRI data (N=16)(Fig 2b,c). Qualitative comparison of noise maps, FA/FOD/CNR/NODDI maps (Fig 2a), tractography for all 16 scans was performed and results were consistent across software versions. An ANOVA analysis didn't show significant differences in any metrics considered: MP-PCA noise (mean, SD, max), tractography FA (mean, median, SD), whole brain FA (mean, SD), FSL eddy QC metrics (Fig 2d).
Data Consistency for fMRI scan: Phantom fMRI datasets were evaluated using GE fMRI QA tool [1] available on console. The QA metrics (RMS, SFNR, SNR, RDC, Mean Ghost) were very similar between the two software releases (Fig 1a). For volunteer fMRI data, the tSNR in grey matter voxels was consistent across software release and no significant difference was found in the tSNR (Fig 1).
Data Consistency for dMRI scan: No significant difference in b0 SNR was observed between two software versions for phantom data (paired t-test: t=0.856, df=3, p=0.45) and volunteer dMRI data (N=16)(Fig 2b,c). Qualitative comparison of noise maps, FA/FOD/CNR/NODDI maps (Fig 2a), tractography for all 16 scans was performed and results were consistent across software versions. An ANOVA analysis didn't show significant differences in any metrics considered: MP-PCA noise (mean, SD, max), tractography FA (mean, median, SD), whole brain FA (mean, SD), FSL eddy QC metrics (Fig 2d).
·Figure 1. Consistency check for functional MRI scans
·Figure 2. Consistency check for diffusion MRI scans
Conclusions:
We present the Open MRI Consistency Data. While done in a small subject cohort, this is the first necessary step towards addressing concerns in discrepancies in fMRI and dMRI measurements between software versions. No statistically significant differences were observed in the comparisons between the software versions. We invite our collaborators to participate in this ongoing initiative by contributing data and results.
Neuroinformatics and Data Sharing:
Databasing and Data Sharing 1
Workflows 2
Keywords:
FUNCTIONAL MRI
MRI
Open Data
Systems
WHITE MATTER IMAGING - DTI, HARDI, DSI, ETC
1|2Indicates the priority used for review