3006

Multi-site Scanner Characterisation for Global Ultra-low Field MR Imaging

Emil Ljungberg^1,2, Francesco Padormo³, John Evans⁴, Petter Clemensson¹, Shannon Kolind⁵, Samson Lecurieux Lafayette⁶, Carly Bennallick², Layla E. Bradford⁷, Kirsten A. Donald^7,8, Able Khosa⁹, Maclean Vokhiwa⁹, Talat Shama¹⁰, Michael S. Pepper¹¹, Alexica De Canha¹¹, Lydia Sekoli¹¹, Jeanne Van Rensburg¹¹, William J. Hollander¹², Todor Karaulanov¹², Steve Williams², and Sean Deoni¹³
¹Medical Radiation Physics, Lund University, Lund, Sweden, ²Neuroimaging, King's College London, London, United Kingdom, ³Hyperfine Inc., Guilford, CT, United States, ⁴Cardiff University Brain Research Imaging Centre (CUBRIC), Cardiff University, Cardiff, United Kingdom, ⁵University of British Columbia, Vancouver, BC, Canada, ⁶Perinatal Imaging & Health, King's College London, London, United Kingdom, ⁷Department of Paediatrics and Child Health, Red Cross War Memorial Children’s Hospital, University of Cape Town, Cape Town, South Africa, ⁸The Neuroscience Institute, University of Cape Town, Cape Town, South Africa, ⁹Training & Research Unit of Excellence (TRUE), Zomba, Malawi, ¹⁰Infectious Diseases Division, icddr,b, Dahka, Bangladesh, ¹¹Institute for Cellular and Molecular Medicine, Department of Medical Immunology, Faculty of Health Sciences, University of Pretoria, Pretoria, South Africa, ¹²CaliberMRI, Boulder, CO, United States, ¹³MNCH D&T, Bill & Melinda Gates Foundation, Seattle, WA, United States

Synopsis

Keywords: Phantoms, MR Value, Quality Assurance

Motivation: Combining data from multiple sites is necessary in order to reach large sample sizes. This can be challenging, especially with portable ultra-low field (ULF) systems where conditions can vary.

Goal(s): To characterise image quality across multiple sites using ULF-MRI for future data harmonization.

Approach: Acquisition of QA scans across sites with centralised analysis for cross-sectional and longitudinal assessment. SNR and T2w contrast were calculated

Results: SNR was consistent within site, but some variation between sites. Image contrast was consistent, but extra care should be taken with software updates.

Impact: Quality assurance is essential for portable ultra-low field MRI. Our results demonstrate that SNR and contrast are useful metrics for image quality characterisation. This paves the way for harmonization of in vivo data in global studies using ultra-low field MRI.

Introduction

Combining MRI data in multi-site studies can prove challenging, but it is evident that pooling data from multiple sites is the only way to achieve large sample sizes needed to map, for instance, normal brain development.¹ Ultra-low field MRI scanners lower the threshold for accessing MR imaging through reduced cost and siting requirements, which will facilitate inclusive global imaging studies. However, there is limited knowledge about harmonization methods for ultra-low field MRI systems in multi-site studies.

In this work we describe our efforts towards harmonization of ultra-low field MRI systems in a multi-site, global neuroimaging study of the developing brain, using a commercially available phantom. Image contrast and signal-to-noise ratios are reported, with cross-sectional results from 11 sites and longitudinal results from 3 sites.

Phantom description

The phantom (CaliberMRI Mini-Hybrid model 137) is a smaller version (170 mm in diameter) of the ISMRM/NIST phantom² with T1, T2, and ADC tissue mimics, as well as a set of solid fiducial markers to map geometrical distortions (Figure 1). An MR-readable liquid crystal (LC) thermometer is also embedded.³ The T1 and T2 mimics are filled with NiCl₂ and MnCl₂, with concentrations ranging between 0.3-65.3 mM and 0.01-1.6 mM respectively, with relaxivity at 64 mT measured by Martin et al. to be NiCl₂: (r₁,r₂)=(0.59, 0.57), MnCl₂: (20.46, 39.28) mM^-1s^-1.⁴ The ADC vials are filled with 0-50% of polyvinylpyrrolidone (PVP). The fill solution is CuSO₄.

Methods

Parameter mapping
Relaxation properties of the phantom mimics were measured at 64 mT using the Hyperfine Swoop (software version 8.7beta1). T1 was quantified using a IR-FSE sequence with 20 TIs logarithmically spaced TIs (25-4000 ms), and TR=5 s. Images were reconstructed offline and T1 maps were calculated using QUIT.^5,6,7 T2 maps were acquired with a multi-echo FSE sequence with first TE at 9 ms, echo spacing of 27 ms and TR=3 s. Image reconstruction and parameter map estimation were performed using scanner online methods.

QA Protocol
A QA protocol was developed including five scans for the following purposes:

Rapid gradient echo - visualisation of the LC thermometer
2x axial T2w FSE - Signal-to-noise ratio (SNR), contrast, and geometric distortions
Cor and Sag T2w FSE - geometric distortions

Each site was provided with a QA manual outlining how to position the phantom in the head coil and the QA protocol was installed on each scanner.

QA image analysis
All compartments of the phantom (T1, T2, ADC mimics and CuSO4 fill) were automatically segmented. SNR was calculated using the NEMA subtraction method⁷ from voxels in the CuSO_4, as this gives a large number of voxels for robust SNR estimates. Image contrast was calculated from the T1 and T2 mimics by normalizing to the respective mimic with lowest salt concentration. In this study, we only investigated SNR and image contrast; geometric distortions will be reported in future work.

Pilot QA data collection
Data was collected from 11 sites, all using Hyperfine Swoop systems, running different hardware and software versions, and with scanners located in different environments. Three of these sites also provided longitudinal data.

Results

Parameter mapping
The T1 and T2 relaxation measurements are shown in Figure 2. The vials with the highest doping were excluded due to too short T1/T2 for accurate parameter estimation. We found a linear relationship between PVP concentration and T1 and T2 (Figure 2), higher PVP concentration yields lower ADC. Mimics 37,36 and 32,36 were identified as being close to adult and neonatal WM/GM T2 contrast (Table 1).^9,10

Pilot multi-site QA
Analysis of image contrast showed consistent results between sites (Figure 3A,B). Longitudinal proxy WM/GM contrast (Figure 3C,D) was consistent within sites, but some effect of the software update which included a known change in sequence parameters. SNR analysis showed high consistency within sites while there were some differences between sites (Figure 4 A,B), as expected when scanners are situated in different environments, evident from differences in operation temperature of 6°C (Figure 4C). We observe a linear relationship between scanner center frequency and temperature, but no clear correlation with SNR.

Discussion and conclusion

We demonstrate that a standardized phantom can be used across a large ultra-low field imaging consortium for scanner characterisation, as a first step towards harmonization. Through this process we identify participation by local team members for QA data collection and management as a critical part for success. Continuous assessment of metrics such as SNR and contrast across the consortium makes it possible to identify factors affecting image quality, e.g., scanner environment or software updates. The next step is to validate these methods in the context of harmonizing in vivo data from multiple sites.

Acknowledgements

This work was funded by the Bill and Melinda Gates Foundation (INV-041062), and part-funded by The Crafoord Foundation (#20220898).

References

1. Bethlehem RAI, Seidlitz J, White SR, Vogel JW, Anderson KM, Adamson C, et al. Brain charts for the human lifespan. Nature. 2022 Apr 21;604(7906):525–33.

2. Stupic KF, Ainslie M, Boss MA, Charles C, Dienstfrey AM, Evelhoch JL, et al. A standard system phantom for magnetic resonance imaging. Magnetic Resonance in Med. 2021 Sep;86(3):1194–211.

3. Keenan KE, Stupic KF, Russek SE, Mirowski E. MRI‐visible liquid crystal thermometer. Magn Reson Med. 2020 Sep;84(3):1552–63.

4. Martin MN, Jordanova KV, Kos AB, Russek SE, Keenan KE, Stupic KF. Relaxation measurements of an MRI system phantom at low magnetic field strengths. Magn Reson Mater Phy [Internet]. 2023 May 20 [cited 2023 May 25]; Available from: https://link.springer.com/10.1007/s10334-023-01086-y

5. Wood TC, Ljungberg E, Wiesinger F. Radial Interstices Enable Speedy Low-volume Imaging. Journal of Open Source Software. 2021;6(66):3500–3500.

6. Wood TC. QUIT: QUantitative Imaging Tools. Journal of Open Source Software. 2017;3(26):656–656.

7. Padormo F, Cawley P, Dillon L, Hughes E, Almalbis J, Robinson J, et al. In vivo T1 mapping of neonatal brain tissue at 64 mT. Magnetic Resonance in Med. 2023;89(3):1016–25.

8. National Electrical Manufacturers Association. NEMA standards publication MS 1‐2008 (R2014): determination of signal‐to‐noise ratio (SNR) in diagnostic magnetic resonance imaging [Internet]. 2014. Available from: https://www.nema.org/docs/default-source/standards-document-library/ms1-2008-r2014-watermarked.pdf?sfvrsn=2101f7b9_2

9. Artz, Nathan S, Sien, S Maura, Robinson, Amie S, Hu, Houchun H, O’Halloran, Rafael, Poorman, Megan, et al. T1 and T2 Mapping of the Infant Brain with a Low-Field, Portable MRI System. In: Proc Intl Soc Mag Reson Med 30 (2022). London, UK; p. 1811.

10. Jordanova KV, Martin MN, Ogier SE, Poorman ME, Keenan KE. In vivo quantitative MRI: T1 and T2 measurements of the human brain at 0.064 T. Magn Reson Mater Phy [Internet]. 2023 May 20 [cited 2023 May 25]; Available from: https://link.springer.com/10.1007/s10334-023-01095-x

Figures

Figure 1: (A) Photo and (B) 3D model of the phantom showing the three quantitative arrays: T1, T2, and ADC, as well as fiducial markers and temperature array. (C) T2w images acquired at 64 mT showing the internal layout of the T1, T2, and ADC mimics on the three different layers. (D) Examples of the liquid crystal thermometer showing the phantom temperature through the number of bright/dark vials.

Figure 2: Results from quantitative T1 and T2 mapping in the phantom with relaxivity values in (A) T1 mimics, (B) T2 mimics, (C-D) diffusion mimics and the CuSO4 fill liquid. A comparison of the range of T1 and T2 values in the T1 and T2 mimics at 64 mT relative to 3T are shown in (E) together with typical values for T1 and T2 in white and gray matter in healthy adult and neonate brain. Room temperature was 20°C as measured by the embedded LC thermometer.

Figure 3: Assessment of T2w image contrast across sites shown in (A) and (B). Subfigures (C) and (D) shows contrast ratio longitudinally for three sites between tissue mimics with representative T1 and T2 values for adult white and gray matter (T2 mimics 37 and 36), and neonatal white matter and thalamus (T2 mimics 32 and 36).

Figure 4: Assessment of signal-to-noise ratio (SNR) between sites (A) and over time for three sites (B). Subfigure B also indicates when a software update was performed to “London 1, UK”. Representative temperature measurements from each site (C) demonstrates the varying operational environment. (D) The Larmour frequency, from DICOM header, is linearly correlated with phantom temperature, while no clear correlation appears with SNR.

Table 1: T1 and T2 values in the NiCl2 and MnCl2 T1 and T2 mimics, shown as mean ± standard deviation.

Proc. Intl. Soc. Mag. Reson. Med. 32 (2024)

3006

DOI: https://doi.org/10.58530/2024/3006