1272

Standardized Parameterization of Echo-Planar Compressed Sensing MRSI Acquisition and Reconstruction

Jason C. Crane¹, Marram P Olson¹, Yan Li¹, Maryam Vareth¹, Hsin-Yu Chen¹, Zihan Zhu¹, Sukumar Subramaniam¹, Peder E.Z. Larson¹, Duan Xu¹, Daniel B. Vigneron¹, and Sarah J. Nelson¹

¹Department of Radiology and Biomedical Imaging, UCSF, San Francisco, CA, United States

Synopsis

Advanced MRSI acquisition strategies can be complex to implement and require customized reconstruction software, typically designed for a specific raw file format and that relies upon a priori knowledge of the specific implementation of the pulse sequence being applied. The ISMRMRD format¹ has begun to address standardization in describing data acquisition parameters for different types of imaging data, but further development is needed. Here we build on this strategy by demonstrating XML encoding of parameters that describe flyback echo-planar, compressed sensing MRSI acquisitions being implemented on scanners from multiple vendors at UCSF that can be supported with generalized reconstruction software.

Purpose

The purpose of this work is to define a set of standardized parameters that describe echo-planar, compressed sensing MRSI k-space trajectories^2-3 and can be used for cross-platform implementation and reconstruction. In order to acquire and reconstruct data with randomly sampled k-space trajectories it is necessary for both the pulse sequences and reconstruction software to know the specifics of the trajectory and data ordering. At the current time this is addressed on a sequence specific basis for spectroscopic acquisitions and the software used to interpret the raw data or file formats for recording the parameters has to be tailored to individual uses. This can be resolved by defining a flexible set of parameters to describe a randomly sampled echo-planar acquisition for compressed sensing reconstruction that can be implemented on multiple vendor platforms and developing standardized software modules that are capable of reordering such acquisitions onto a zero-padded Cartesian k_x,k_y,k_z,t grid suitable for reconstruction (Figure 1).

Implementation

The k-space trajectory of the flyback echo-planar compressed sensing MRSI acquisition was defined by 3 parameter groups that describe 1) the 2D phase encoding trajectory, 2) the flyback waveform parameters for encoding the 3rd spatial and spectroscopic dimensions and 3) the blipped phase encoding pattern that defines the specific k_x,k_y points sampled along the flyback trajectories. This information was encoded in an XML Data Acquisition Descriptor (DAD) file⁴ with specific XML elements for each of these 3 parameter groups. Software libraries for reading and writing DAD XML files as well as data ordering modules used in reconstruction pipelines were implemented in C++ using the open-source SIVIC⁵ software package. Separate modules were developed for interpreting the blipped compressed sensing phase encoding scheme and for interpreting the flyback encoding (Figure 1). This enables the latter module to also be utilized for fully sampled EPSI acquisitions.

2D phase encoding is defined by a trajectory and a sampling mask element in the DAD file (Figure 2). The trajectory is defined by the linear indices corresponding to specific k_x,k_y pairs for each point in the trajectory, here shown for centric encoding. The sampling mask indicates which k-space indices are actually sampled. The EPSI waveform is defined by 6 parameters (Figure 3). Each EPSI waveform follows either a fully sampled trajectory along k_x,k_y,k_z or a randomly undersampled path as prescribed by the DAD file <blip> elements (Figure 4) which define the blipped phase encoding gradient changes for each lobe along the flyback waveform. For example the EPSI readout starting at index 29 (Figure 2) fully samples the trajectory k_x,k_y,{k_z}, whereas the readout starting at index 1 meanders randomly through a 2x2 block of k_x,k_y for each lobe of the waveform as indicated by the gradient blips for <blip index=1>.

Results

Software modules for reading and reordering ¹³C echo-planar compressed sensing data² were implemented on a 14T Varian (Agilent) scanner in SIVIC. This initially used hardcoded trajectories and blip patterns to test the implementation. The reconstruction software was then refactored and generalized to read the parameters dynamically from a DAD file. This simplified the software modules and enabled them to be generalized to independently support arbitrary random sampling and echo-planar encoded acquisitions. The reordered, zero-padded data are shown in Figure 5. The EPSI reordering module, together with DAD file EPSI descriptors was able to support interpretation of fully sampled EPSI acquisitions from a GE scanner with DAD writing libraries linked into the custom PSD.

Discussion

Raw data formats do not currently include sufficient information to reconstruct MRSI data that are acquired with advanced sampling strategies. Standardized parameterization of MRSI acquisition strategies such as parallel imaging, compressed sensing, and EPSI will streamline translation of sequences between vendors and enable utilization of these strategies using standardized software modules. Next steps in this work will include modifying echo-planar compressed sensing sequences, including for dynamic acquisitions, on GE and Bruker scanners to write DAD files describing these acquisition parameters and validating the reordering of raw data to a Cartesian k-space grid using the software modules used for the Varian implementation. Once tested on multiple scanner platforms, these descriptors will be proposed as additions to the ISMRMRD¹ working group.

Conclusion

Standardization of parameters required to describe common classes of complex MRSI sampling strategies will permit the development of software packages capable of interpreting the results and streamline development and translation of MRSI methods. These parameters are currently being encoded via a DAD/XML file, but will be proposed for incorporation into broader standards such as the ISMRMRD or DICOM. This will be critical for streamlining MRSI development and practical implementation.

Acknowledgements

This work was supported by NIH grant P41 EB013598.

References

1. Inati SJ, Naegele JD, Zwart NR, et al. ISMRM Raw data format: A proposed standard for MRI raw datasets. Magn. Reson. Med. 2016;0(August 2015):n/a-n/a. Available at: http://doi.wiley.com/10.1002/mrm.26089.

2. Hu S, Lustig M, Balakrishnan A, et al. 3D compressed sensing for highly accelerated hyperpolarized (13)C MRSI with in vivo applications to transgenic mouse models of cancer. Magn. Reson. Med. 2010;63(2):312–21. Available at: http://www.pubmedcentral.nih.gov/articlerender.fcgi?artid=2829256&tool=pmcentrez&rendertype=abstract. Accessed August 25, 2014.

3. Larson PEZ, Hu S, Lustig M, et al. Fast dynamic 3D MR spectroscopic imaging with compressed sensing and multiband excitation pulses for hyperpolarized 13C studies. Magn. Reson. Med. 2011;65(3):610–619.

4. Olson MP, Crane JC, Larson P, et al. A Vendor-Agnostic MRSI Acquisition and Reconstruction XML Descriptor Format. In: ISMRM 25th annual meeting.; 2017. Available at: http://dev.ismrm.org/2017/5533.html.

5. Crane JC, Olson MP, Nelson SJ. SIVIC: Open-Source, Standards-Based Software for DICOM MR Spectroscopy Workflows. Int. J. Biomed. Imaging. 2013;2013:169526. Available at: http://www.pubmedcentral.nih.gov/articlerender.fcgi?artid=3732592&tool=pmcentrez&rendertype=abstract.

Figures

Figure 1. Vendor neutral DAD files are used by scanners to define echo-planar compressed sensing acquisition trajectory and also by generalized software modules that are used to reorder data from raw file onto a zero-padded Cartesian grid that is suitable for reconstruction and analysis by generalized software packages.

Figure 2. Representation of phase encoding (PE) trajectory (A) by XML <phase_encoding> element indicating order of acquired k_x,k_y points in index space. A sampling mask indicates which k-space elements are encoded (grey). Element-wise multiplication of the PE trajectory (A) by the sampling mask (B) indicates the sampling order. Pink boxes indicate different PE random sampling blocks (Figure 4). The center of k-space (grey) is fully sampled along the k-space trajectory. The periphery of k-space is divided in this example into undersampled 2x2 k-space blocks where each flyback lobe is sampled at 1 of the four k_x,k_y pairs (see Figure 4).

Figure 3. XML elements for parameters representing of EPSI flyback waveform.

Figure 4. Representation of random compressed sensing blipped phase encoding of k_x,k_y along flyback trajectory. k_f is the flyback direction and each flyback lobe (k_z,t) is acquired at a randomly located k_x,k_y location (pink). The trajectory in k_x,k_y is defined by the blipped phase encoding scheme encoded in the corresponding DAD file <blips> element shown here for the first 11 lobes of the 2x2 block starting at index 1 (Figure 2). Fully sampled flyback readouts in k_x,k_y are indicated by 0 valued blips elements indicating no change in G_x, G_y during the flyback readout (e.g. index 29, Figure 2).

Figure 5. 3D ¹³C Flyback EPSI compressed sensing data from 14 T Varian scanner reordered onto zero-padded Cartesian k-space grid (top right). The center of k-space is fully sampled (pink square, bottom right). A 2x2 undersampled block of k-space is shown at left. The time points corresponding to flyback lobes 29-33 from this undersampled region are shown bottom left. Note that for this 2x2 block only 1 k-space point (k_x,k_y) is sampled at each time point (flyback lobe) corresponding to the k_x,k_y pair defined by the blipped trajectory for that k_x,k_y,lobe (see Figure 4).

Proc. Intl. Soc. Mag. Reson. Med. 26 (2018)

1272