Open-source magnetic resonance imaging acquisition: Data and documentation for two validated pulse sequences

Raw data, simulated and acquired phantom images, and quantitative longitudinal and transverse relaxation times (T1/T2) maps from two open-source Magnetic Resonance Imaging (MRI) pulse sequences are presented in this dataset along with corresponding “.seq” files, sequence implementation scripts, and reconstruction/analysis scripts [1]. Real MRI data were collected from a 3T Siemens Prisma Fit and a 1.5T Siemens Aera via the Pulseq open-source MR sequence platform, and corresponding in silico data were generated using the simulation module of Virtual Scanner [2]. This dataset and its associated code can be used to validate the pipeline for using the same pulse sequences at other research sites using Pulseq, to provide guidelines for documenting and sharing open-source pulse sequences in general, and to demonstrate practical, customizable acquisition scripts using the PyPulseq library.


a b s t r a c t
Raw data, simulated and acquired phantom images, and quantitative longitudinal and transverse relaxation times (T 1 /T 2 ) maps from two open-source Magnetic Resonance Imaging (MRI) pulse sequences are presented in this dataset along with corresponding ".seq" files, sequence implementation scripts, and reconstruction/analysis scripts [1] . Real MRI data were collected from a 3T Siemens Prisma Fit and a 1.5T Siemens Aera via the Pulseq open-source MR sequence platform, and corresponding in silico data were generated using the simulation module of Virtual Scanner [2] . This dataset and its associated code can be used to validate the pipeline for using the same pulse sequences at other research sites using Pulseq, to provide guidelines for documenting and sharing open-source pulse sequences in general, and to demonstrate practical, customizable acquisition scripts using the PyPulseq library.

Specifications Table
For all qualitative scans, the American College of Radiology (ACR) large MRI phantom [5] was acquired. Quantitative scans used the T 1 and T 2 planes of the International Society for Magnetic Resonance Medicine / National Institute of Standards and Technology (ISMRM/NIST) phantom [6] for T 1 and T 2 mapping, respectively.
Reconstruction scripts were provided in MATLAB (.m). The PDF forms were created using a combination of Adobe Indesign and Adobe Acrobat DC and filled by the two scanning sites.
Experimental parameters are shown in Table 1

Value of the Data
• The packaged and documented sequence waveforms, raw data, and reconstructed images can serve as a baseline for future effort s in standardizing open-source MRI acquisition, reconstruction and mapping pipelines. • Researchers working on repeatability and cross-site studies can use the data as a starting point for communicating and harmonizing sequences using the Pulseq platform. • Students working in MR can follow the framework to gain practical knowledge in designing MR sequences. • Further experiments can use the programs in vivo to evaluate clinical usefulness of opensource pulse sequences. • The sequence scripts can be used to assess the optimal degrees of freedom for sequence parameter customization. That is, how many changes and options should a sequence script allow before the more efficient choice is to branch out into multiple scripts. • Effort s in Accessible MRI can use the data, acquired using open-source sequences on major commercial scanners, as a reference point for experiments using the same sequences on new hardware.

Data Description
See Table 2 for the list of MRI terminology abbreviations for this dataset.

Documentation forms
The folder (documentation_templates) contains the empty developer and user forms (seq_validation_form_DEVELOPER.pdf, seq_validation_form_USER.pdf) for documenting test experiments in the proposed sequence validation framework [1] .

Sequence implementation
Two classic MRI sequences were implemented in the multi-vendor, open-source Pulseq format [3] in Python with the PyPulseq library [4] : Inversion Recovery Spin Echo (IRSE) and Turbo Spin Echo (TSE) [7] . The scripts were programmed in Google Colaboratory [8] as Jupyter Notebooks.

Simulation
Lower resolution numerical simulation based on the Bloch equations [9] was performed using in-house software [10] that directly converts a Pulseq sequence into a list of commands. These commands in turn are applied to individual isochromats, defined by their proton density, location in 3D space, and T 1 /T 2 relaxation times, that make up a numerical phantom. When exposed to temporally and spatially varying magnetic fields such as those defined by a pulse sequence program, the isochromat's magnetization vector evolves according to its initial condition, innate parameters, and the external driving fields. A numerical library is used to solve the differential equations. In the end, the detectable signals from the transverse magnetization are added up across all isochromats in a phantom to generate the raw MRI signal.
All simulations were performed on a Windows 10 operating system with an Intel(R) Core i7-8650 U CPU. Specific parameters used for the simulation were: FOV = 250 mm, slice thickness = 5 mm; TR = 4500 ms, TI = 200 ms, and TE = 10 ms for IRSE; TR = 4500 ms and TE = 10 ms for TSE.

Acquisition and reconstruction
Experiment parameters are shown in Table 1 . We reconstructed the raw data with simple 2D Inverse Fast Fourier Transform on each channel after correcting the ordering in the phase encoding dimension. The channels were combined using sum-of-squares. For the quantitative mapping experiments, each k-space was first multiplied with a 128-point 2D Hamming filter [11] before reconstruction to reduce ringing artifacts around and inside the small T 1 /T 2 spheres.

Quantitative mapping and image metrics
For each mapping experiment, the corresponding signal equation (T 1 : S = A (1 − 2 exp (−T I/ T 1 ) + exp ( −T R/ T 1 )) + B ); T 2 : S = A exp ( −T E/ T 2 ) + B ; A: signal scaling; B: signal offset value) was fitted to the multi-contrast images voxel-by-voxel using the MATLAB curve fitting toolbox (see scripts in the analysis data folders) within a manually selected overall ROI that covers the circular phantom. Following generation of T 1 /T 2 maps, small sphere-wise ROIs were manually selected to ensure only interior voxels are used. For each ROI, the mean and standard deviation of T 1 /T 2 values were computed.
For the qualitative images, Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index Measure (SSIM) [12] were computed against equivalent vendor images acquired in the same session. Normalization was performed across the 11 slices for each set before comparison. The seven standard ACR tests were performed on the corresponding slices according to the published guidance [5] .

Safety
The open-source sar4seq library was used to generate predicted time-averaged RF power and Specific Absorption Rate (SAR) for each sequence [ 13 , 14 ]. At acquisition, values were also read off from vendor software for measured SAR, time-averaged RF power, and percent Peripheral Nerve Stimulation (PNS) threshold. All evaluation assumed a 70 kg subject.