2182

Quantifying 3D-MRF Reproducibility Across Subjects, Sessions, and Scanners Automatically Using MNI Atlases
Andrew Dupuis1, Yong Chen1, Michael Hansen2, Kelvin Chow3, Dan Ma1, Mark Griswold1, and Rasim Boyacioglu1
1Biomedical Engineering, Case Western Reserve University, Cleveland, OH, United States, 2Microsoft Research, Redmond, WA, United States, 3Siemens Medical Solutions USA, Inc., Chicago, IL, United States

Synopsis

Keywords: MR Fingerprinting/Synthetic MR, Quantitative Imaging, Reproducibility, MNI, Automated, Gadgetron, FIRE

We evaluate the reproducibility of 3D-MRF versus clinical-standard MPRAGE and TSE acquisitions for ten subjects across three acquisition sets on two scanner via a fully-automated registration and regional analysis framework. T1 and T2 quantitative maps from 3D-MRF were found to be highly reproducible (T1+/-4.6%, T2+/-6.3%) across scanners and sessions, with no significant difference between repeating a scan immediately on the same scanner, repeating a scan after repositioning the subject and reshimming, and repeating a scan on a different scanner entirely. The same is not true for MPRAGE (+/-12.4%) or TSE (+/-28.8%) acquisitions.

Purpose

Magnetic Resonance Fingerprinting (MRF) has been shown to reliably and reproducibly quantify T1 and T2 in both phantom and in-vivo studies using a 2D acquisition with offline reconstruction and manually drawn ROIs. However, the intensive computation needed for MRF reconstruction has historically necessitated manually performing offline reconstructions, making comparison against clinical standard imaging difficult and limiting the utility of well-established post-processing tools like FSL.

In this work, we evaluate a fully-automatic online 3D-MRF reconstruction, as well as a cross-modality registration and analysis system allowing for diret comparison of 3D-MRF versus clinical standard MPRAGE and TSE acquisitions. We use this system to perform an intrascanner and interscanner reproducibility study to determine whether in-vivo 3D-MRF reproducibility meets or exceeds that of MPRAGE and TSE. Additionally, we evaluate whether the reproducibility of T1/T2 maps across sessions and scanners is stable enough to potentially eliminate scanner dependency as a differentiation between MRF datsets.

Methods

Ten healthy volunteers were imaged over two sessions on different days, with each session occuring on one of two 3T scanners on different software versions (MAGNETOM Vida, VA20 and VA31, Siemens Healthcare, Erlangen, Germany). Each session consisted of three “sets” of acquisitions , with each set consisting of three “series” of images: 3D-MRF FISP with a B1 mapping prescan, 3D-MPRAGE, and multislice 2D-TSE. All were acquired with a field of view of 250x250x150mm3 and a spatial resolution of 1x1x2.5mm3. First, an “original” image set was acquired for each subject, followed immediately by a “repetition” set. Subjects were removed from the scanner, asked to stand up, and then sent back into the bore. A new localizer was acquired, followed by the final “reposition” set. The procedure was then repeated on the second scanner (a maximum of 7 days apart).

TSE and MPRAGE datasets were processed online by each scanner’s ‘product’ reconstruction. 3D-MRF was reconstructed online via a custom Gadgetron Kubernetes cluster using the FIRE interface prototype. Quantitative maps were returned to the scanner via FIRE. All images were then exported to header-complete DICOM sets.

A custom fully-automated registration and region extraction pipeline was used for analysis. Figure 1 details the registration flow used in this study. Images were first converted to NIFTI using dcm2niix. Synthetic MPRAGE and TSE contrasts were generated for MRF acquisitions, then all images were linearly registered to the first scanner’s “original” MPRAGE image series via FLIRT. Finally, the “original” MPRAGE set was registered to MNI-152 via FNIRT. All transformation and warp matrices were saved.

Atlas label images were generated for each atlas/subject/scanner/set combination and uploaded to Azure. Twelve well-defined, relatively homogenous regions from the Harvard-Oxford subcortical atlas were selected for the purpose of comparison. Bland-Altman plots were then generated comparing the reproducibility of regional means across subjects, scanners, and repetitions.

Results

60 imaging sets were acquired for 10 healthy subjects, constituting ~1.4TB of raw data. Data were converted to ISMRMD format for future analysis. All raw data, DICOMs, NIFTI-converted images, linear and nonlinear transformation matrices, warped label maps, and atlas-label voxel dictionaries were uploaded to Azure Blob storage for validation and retrospective analysis. Access via a Python API is available upon request to encourage further investigation and statistical analysis using these data.

Figures 2, 3, and 4 compare the reproducibility of 3D-MRF, TSE, and MPRAGE acquisitions using regional means compared between immediate repetitions, subject repositions on the same scanner, and multiple repetitions on two scanners respectively. Voxelwise T1/T2 and T1w/T2w ratios for regions are included since they are a common pseudoquantitative biomarker reliant on inter/intrasession repeatability. Figure 5 summarizes the mean percent differences and standard deviations of 3D-MRF, TSE, and MPRAGE across all three comparisons. Note that the standard deviation of the MPRAGE and TSE data sets was substantially larger than either of the MRF maps.

Discussion

The consistent biases and standard deviations of T1 and T2 values in Figure 5 indicate that in-vivo 3D-MRF is reproducible whether a scan is repeated immediately (T1: 1.25±4.33%, T2: 1.88±6.18%), with a reposition of the subject on the same scanner (T1: 0.43±4.49%, T2: 1.31±4.88%), or on a different scanner on a different day (T1: -0.57±4.90%, T2: -3.28±5.34%). Of note, the standard deviations observed are the same magnitude as one “step” within the MRF dictionary used during reconstruction. The T1/T2 ratio calculations showed similar stability.

MPRAGE, TSE , and the resulting T1w/T2w ratios do not exhibit the same reproducibility as 3D-MRF. Although MPRAGE and TSE are qualitative acquisitions and therefore not expected to produce reliable quantitative output, the highly unpredictable differences observed across repetitions (MPRAGE: 6.56±14.21%, TSE: -0.43±23.11%), sessions (MPRAGE: -0.98±12.67%, TSE: -2.25±26.87%) and scanners (MPRAGE: 1.10±13.75%, TSE: 3.74±32.30%) casts doubt on the reliability of metrics like T1w/T2w ratio, even for contrasts acquired within the same scan session.

The apparent reproducibility of in-vivo 3D-MRF offers multiple opportunities: intrascanner and interscanner variations are negligible such that data from many sessions, scanners, and sites can potentially be treated as a single dataset for inference or other deep learning processing . Similarly, intrasubject cross-scanner comparisons are likely valid. Finally, the reliability of the resulting maps suggests that synthetic contrast generation based on MRF maps may allow for system- and session-agnostic T1/T2 weighted contrasts with reproducibility exceeding the current clinical standard.

Acknowledgements

This work was supported by Siemens Healthineers and Microsoft Corporation

References

Lo WC, Bittencourt LK, Panda A, Jiang Y, Tokuda J, Seethamraju R, Tempany-Afdhal C, Obmann V, Wright K, Griswold M, Seiberlich N, Gulani V. Multicenter Repeatability and Reproducibility of MR Fingerprinting in Phantoms and in Prostatic Tissue. Magn Reson Med. 2022 Oct;88(4):1818-1827. doi: 10.1002/mrm.29264. Epub 2022 Jun 17. PMID: 35713379; PMCID: PMC9469467.

Reproducibility and Repeatability of MR Fingerprinting Relaxometry in the Human Brain Gregor Körzdörfer, Rainer Kirsch, Kecheng Liu, Josef Pfeuffer, Bernhard Hensel, Yun Jiang, Dan Ma, Marcel Gratz, Peter Bär, Wolfgang Bogner, Elisabeth Springer, Pedro Lima Cardoso, Lale Umutlu, Siegfried Trattnig, Mark Griswold, Vikas Gulani, and Mathias Nittka Radiology 2019 292:2, 429-437

Makris N, Goldstein JM, Kennedy D, Hodge SM, Caviness VS, Faraone SV, Tsuang MT, Seidman LJ. Decreased volume of left and total anterior insular lobule in schizophrenia. Schizophr Res. 2006 Apr;83(2-3):155-71

Jenkinson, M., Bannister, P., Brady, J. M. and Smith, S. M. Improved Optimisation for the Robust and Accurate Linear Registration and Motion Correction of Brain Images. NeuroImage, 17(2), 825-841, 2002

Andersson JLR, Jenkinson M, Smith S (2010) Non-linear registration, aka spatial normalisation. FMRIB technical report TR07JA2

M.W. Woolrich, S. Jbabdi, B. Patenaude, M. Chappell, S. Makni, T. Behrens, C. Beckmann, M. Jenkinson, S.M. Smith. Bayesian analysis of neuroimaging data in FSL. NeuroImage, 45:S173-86, 2009

Chen Y, et al. Rapid 3D MR Fingerprinting Reconstruction using a GPU-Based Framework. ISMRM 2020.

Hansen MS, Sørensen TS. Gadgetron: An Open Source Framework for Medical Image Reconstruction. Magn Reson Med. 2013 Jun;69(6):1768-76.

Xue H, Inati S, Sørensen TS, Kellman P, Hansen MS. Distributed MRI reconstruction using Gadgetron-based cloud computing. Magn Reson Med. 2015 Mar;73(3):1015-25. doi: 10.1002/mrm.25213. Epub 2014 Mar 31. PMID: 24687458; PMCID: PMC6328377.

Kubernetes (K8s): Production-Grade Container Scheduling and Management. Kubernetes Github Repository, 2021 Nov; https://github.com/kubernetes/kubernetes

Gadgetron-Azure: Sample code for deploying Gadgetron image reconstruction in Azure. Azure-Gadgetron Github Repository, 2021 Nov; https://github.com/microsoft/gadgetron-azure

Chow K, Kellman P, Xue H. Prototyping Image Reconstruction and Analysis with FIRE. SCMR 2021.

Figures

Figure 1: 7-parameter registration via FLIRT was used within subjects. For each imaging set, TSE images were registered to MRF maps via synthetic TSE . MRF was then registered to MPRAGE via synthetic MPRAGE. MPRAGE images from each imaging sets and scanner were then registered to the “original” MPRAGE series, which was then nonlinearly warped via FNIRT to MNI-152 space. The linear and nonlinear transformations saved for each scanner/set/series combination were then inverted and used to generate set-specific atlas label maps used to sample and save voxel buffers for all atlas regions.

Figure 2: First, the reproducibility between the "original" and "repetition" sets on each scanner were compared for all subjects. This represents a same-session test-retest on the same scanner hardware, since the subject remained in the bore and the scanner was not reshimmed between acquisitions.

Figure 3: Second, the reproducibility between the "original" and "reposition" sets on each scanner were compared for all subjects. Between acquisitions, the subject was removed from the bore then placed back inside, triggering a reshimming of the system. This represents a cross-session test-retest on the same scanner hardware.

Figure 4: Finally, the reproducibility across all combinations of subjects, scanners, and sets was compared. The data included is the full matrix of 9 combinations of original, repetition, and reposition sets across both scanners so that no single set is used as a "master" comparison point.

Figure 5: The above table summarizes the three tests performed in order to determine whether the mean (bias) or standard deviation vary substantially depending on immediate, repositioned, or cross-scanner repetitions. Standard deviation, which also drives the confidence intervals seen in Figures 2/3/4, is the most important metric to compare, and is displayed in bold. No significant difference in performance is seen for the 3D-MRF derived T1, T2, or T1/T2 ratio across tests. The same is true for MPRAGE, but the reproducability of TSE struggles across sessions and scanners.

Proc. Intl. Soc. Mag. Reson. Med. 31 (2023)
2182