Introduction

Neuroimaging research relies heavily on the consistent image quality. This concerns both structural anatomic imaging and functional MR imaging (fMRI). Especially the fMRI, that associates blood oxygenation level dependent (BOLD) signal changes to brain function, depends critically on the system stability. Furthermore, intricate details of data acquisition, image reconstruction and integrated post-processing also have a great effect on the results of fMRI studies. Quality assurance (QA) protocols can test the stability of MR scanners and help researches to detect errors during longitudinal MRI studies, thereby greatly enhancing their success rate. In multicenter settings, QA protocols help establishing consistent data standards across scanners and ensuring data comparability, which is essential for data pooling. However, heterogeneity of the imaging hardware may present substantial challenges for cross-scanner and cross-center quality comparisons due to the necessity of implementing consistent protocols in varying vendor-specific environments.
In this work we use Pulseq to implement an open-source transparent quality assessment MR protocol based on recommendations of the American Congress of Radiology (ACR) and the Function Biomedical Informatics Research Network (fBIRN), that is accompanied by an open-source data processing pipeline. The proposed protocol has been thoroughly tested on three 3T Siemens scanners with the manufacturing date separated by over 17 years and also successfully executed on one from 3T GE scanner.

Methods

Pulse sequences
A multi-slice spin echo (SE) sequence with two Shinnar-Le Roux radiofrequency pulses for excitation and refocusing was implemented with the Pulseq framework to measure the structural quality. The sequence parameters are based on those of the ACR Axial T1 series [1]: TR/TE = 500/20; field of view (FOV) = 250 * 250 mm2; number of slices = 11; slice thickness/gap = 5/5 mm; repetition = 2; matrix size = 256 * 256; total scan time = 2:16. A 2D slice-selective echo-planar imaging (EPI) sequence was also implemented using Pulseq for fMRI scan. The fMRI acquisition parameters follow Friedman et al [2]: TR/TE=2000/22 ms; FOV=220*220 mm2; number of slices = 27; slice thickness/gap = 4/1 mm; flip angle=90°; number of volumes = 200; scan time = 6:40 min. Siemens product SE and EPI protocols were configured to closely match the corresponding Pulseq protocols.
Automatic processing
An iterative circle detection algorithm was used to automatically detect the position of phantom. This position information was used to automatically calculate the structural quality and functional quality. The structural quality metrics are[1]: signal-to-noise ratio (SNR), percent signal ghosting (PSG), Percent Image Uniformity (PIU). The functional quality metrics are [3]: signal-to-fluctuation-noise ratio (SFNR), temporal SNR (tSNR), temporal signal-to-background noise ratio (tSBNR), and functional signal-to-noise ratio (SNR-functional).
Measurement
A water-filled phantom doped with a NiCl/NaCl mixture was used as an alternative of the fBIRN phantom. A first fMRI acquisition was performed to let the system reach a temperature equilibrium. And the second fMRI scan and the SE sequence from Siemens and Pulseq were executed to produce data for QA metrics calculation.

Results and Discussion

Figure 1 presents example phantom images acquired with the spin-echo protocol. As seen, a consistent image quality can be observed both for vendor sequences on three scanners and the Pulseq counterparts, although image homogeneity varies, which can be attributed to differences in receiver coil sensitivities, which were not normalized here. Figure 2 shows EPI results for both vendor and Pulseq sequences.
Detailed numerical analysis can be seen in Table 1, which shows a general tendency for the image quality to improve with new scanner generations, with a slight exception for EPI performance on CimaX, which needs to be investigated further.
The same protocol has been tested on MR750 3T GE scanner, but the automated data processing pipeline was not finalized at the time of this writing. The cross-platform capability of the framework has however been verified.

Acknowledgements

This work was supported by research grants NIH R01 EB032378 and NIH U24 NS120056.