Harmonization of an SMS-EPI fMRI Protocol Using Pulseq: Sequence Implementation in Human Subjects
Poster No:
1937
Submission Type:
Abstract Submission
Authors:
Maximillian Egan1, Scott Peltier1, Qingping Chen2, Maxim Zaitsev2, Stephen LaConte3, Jeff Soldate3, Jonathan Lisinski3, Aaron Anderson4, Brad Sutton4, Jon-Fredrik Nielsen1
Institutions:
1University of Michigan, Ann Arbor, MI, 2University Medical Center Freiburg, Freiburg, Germany, 3Virginia Tech, Roanoke, VA, 4University of Illinois at Urbana-Champaign, Urbana, IL
First Author:
Co-Author(s):
Introduction:
In principle, pooling functional MRI data across multiple sites and timepoints allows for increased statistical power, provided that the experiment can be conducted in a known and reproducible way. While post-processing attempts to correct for 'site' variance have been somewhat successful, an additional approach would be to control for potential differences (e.g., vendor, scanner model/software) during the acquisition itself. Historically this has been a challenge as the details of the vendor-provided fMRI acquisition and reconstruction software are generally not known. One could in principle implement the imaging protocol using each vendor's pulse sequence programming environment, but this is a technically difficult and time-consuming task that does not guarantee identical sequences.
Pulseq is an open, vendor-independent MR pulse programming platform that allows an MRI sequence to be created and analyzed in interactive programming environments such as MATLAB or Python, saved to a simple text file, and executed on hardware using sequence-agnostic interpreters (Layton, 2017). Pulseq guarantees exact, low-level sequence harmonization across vendor platform. Here, we implement a Pulseq scanning sequence and reconstruction in human subjects to demonstrate the applicability of vendor-agnostic sequences in human fMRI research.
Pulseq is an open, vendor-independent MR pulse programming platform that allows an MRI sequence to be created and analyzed in interactive programming environments such as MATLAB or Python, saved to a simple text file, and executed on hardware using sequence-agnostic interpreters (Layton, 2017). Pulseq guarantees exact, low-level sequence harmonization across vendor platform. Here, we implement a Pulseq scanning sequence and reconstruction in human subjects to demonstrate the applicability of vendor-agnostic sequences in human fMRI research.
Methods:
Subjects: 2 human subjects underwent fMRI recording using both vendor-specific (product) and Pulseq sequences at multiple scanners. 5 minutes of resting state data were recorded, followed by four 3.5 minute task runs consisting of 20s alternating blocks of bilateral finger tapping and resting while focused on a fixation cross.
Sequence design and scanner implementation: Using the Pulseq MATLAB toolbox we designed the SMS-EPI sequence, which consists of a fat saturation pulse, SMS excitation (multiband factor 6), and CAIPI EPI readout (2.4mm isotropic resolution; TR=0.8s). Data were acquired on two scanners, a GE MR750 using a 32 channel Nova Medical head array, and a Siemens Vida using the Siemens head and neck 64 channel array.
Analyses: Resting state functional connectivity (rsFC) matrices were generated from the subjects' resting state runs for each scanner/sequence combination. These rsFC matrices were then pooled for each sequence (i.e., collapsed across scanners), the correlations of the matrices within each pool were calculated, and the average correlations of the sequences were compared. To assess ability to reproduce task activation maps, the finger tapping vs. fixation cross blocks were contrasted for each of the scanner and sequence types, and peak activations, locations, and cluster sizes were compared.
Sequence design and scanner implementation: Using the Pulseq MATLAB toolbox we designed the SMS-EPI sequence, which consists of a fat saturation pulse, SMS excitation (multiband factor 6), and CAIPI EPI readout (2.4mm isotropic resolution; TR=0.8s). Data were acquired on two scanners, a GE MR750 using a 32 channel Nova Medical head array, and a Siemens Vida using the Siemens head and neck 64 channel array.
Analyses: Resting state functional connectivity (rsFC) matrices were generated from the subjects' resting state runs for each scanner/sequence combination. These rsFC matrices were then pooled for each sequence (i.e., collapsed across scanners), the correlations of the matrices within each pool were calculated, and the average correlations of the sequences were compared. To assess ability to reproduce task activation maps, the finger tapping vs. fixation cross blocks were contrasted for each of the scanner and sequence types, and peak activations, locations, and cluster sizes were compared.
Results:
rsFC: The rsFC matrices showed higher correlation within-sequence using the Pulseq sequence as compared to the product sequences (rPulseq = 0.45 vs. rproduct = 0.34), though this difference was not statistically significant. Figures 1a and 1b visualize these rsFC matrices, with Figure 1c showing the difference between the two averaged rsFC matrices of the sequences.
Task activations: Activations in primary motor cortex were similar for both product and Pulseq sequences during the finger tapping vs. fixation cross contrast. Peak activations, cluster sizes, and locations are visualized for one subject for each scanner and sequence in Figure 2. While peak activations and locations varied slightly between each sequence/scanner, robust M1 activation was comparable across all sequences.
Task activations: Activations in primary motor cortex were similar for both product and Pulseq sequences during the finger tapping vs. fixation cross contrast. Peak activations, cluster sizes, and locations are visualized for one subject for each scanner and sequence in Figure 2. While peak activations and locations varied slightly between each sequence/scanner, robust M1 activation was comparable across all sequences.
Conclusions:
Pulseq is an open-source, vendor agnostic tool that aims to harmonize image sequencing and reconstruction. Here we demonstrate in human subjects that the Pulseq platform produces similar results in rsFC and task activation measures as compared to multiple product scanners and vendors. Upcoming work will increase the number of both traveling subjects and unique scanner/vendors in this study.
Novel Imaging Acquisition Methods:
BOLD fMRI 2
Imaging Methods Other 1
Keywords:
FUNCTIONAL MRI
Open-Source Software
Other - Harmonization
1|2Indicates the priority used for review