Accelerating High-Resolution MRSI Reconstruction: An ICE and FIRE Online Pipeline
Poster No:
1947
Submission Type:
Abstract Submission
Authors:
Korbinian Eckstein1, Bernhard Strasser1, Lukas Hingerl1, Barbara Dymerska2, Stanislav Motyka1, Aaron Osburg1, Amirmohammad Shamaei3, Steffen Bollmann4, Wolfgang Bogner1
Institutions:
1Department of Biomedical Imaging and Image-guided Therapy, Medical University of Vienna, Vienna, Austria, 2Department of Imaging Neuroscience, Queen Square Institute of Neurology, University College London, London, United Kingdom, 3Electrical and Software Engineering, University of Calgary, Calgary, Alberta, 4The University of Queensland, Brisbane, Queensland
First Author:
Korbinian Eckstein
Department of Biomedical Imaging and Image-guided Therapy, Medical University of Vienna
Vienna, Austria
Department of Biomedical Imaging and Image-guided Therapy, Medical University of Vienna
Vienna, Austria
Co-Author(s):
Bernhard Strasser
Department of Biomedical Imaging and Image-guided Therapy, Medical University of Vienna
Vienna, Austria
Department of Biomedical Imaging and Image-guided Therapy, Medical University of Vienna
Vienna, Austria
Lukas Hingerl
Department of Biomedical Imaging and Image-guided Therapy, Medical University of Vienna
Vienna, Austria
Department of Biomedical Imaging and Image-guided Therapy, Medical University of Vienna
Vienna, Austria
Barbara Dymerska
Department of Imaging Neuroscience, Queen Square Institute of Neurology, University College London
London, United Kingdom
Department of Imaging Neuroscience, Queen Square Institute of Neurology, University College London
London, United Kingdom
Stanislav Motyka
Department of Biomedical Imaging and Image-guided Therapy, Medical University of Vienna
Vienna, Austria
Department of Biomedical Imaging and Image-guided Therapy, Medical University of Vienna
Vienna, Austria
Aaron Osburg
Department of Biomedical Imaging and Image-guided Therapy, Medical University of Vienna
Vienna, Austria
Department of Biomedical Imaging and Image-guided Therapy, Medical University of Vienna
Vienna, Austria
Wolfgang Bogner
Department of Biomedical Imaging and Image-guided Therapy, Medical University of Vienna
Vienna, Austria
Department of Biomedical Imaging and Image-guided Therapy, Medical University of Vienna
Vienna, Austria
Introduction:
MR spectroscopic imaging (MRSI) provides localized information about metabolites, with the potential for clinical application in tumors (Bruhn et al, 1989), neurodegenerative diseases (Heckova et al, 2019), or to detect neurotransmitter networks (Veendaal et al, 2018). A barrier for clinical adoption is the complicated reconstruction of MRSI raw data, requiring algorithmically complex and computationally intense steps, which are usually implemented in high-level programming languages. For high-resolution MRSI, this requires the export of large raw data files (~100GB), a powerful machine with large amounts of memory (>100GB) and multiple hours of reconstruction time.
To overcome these barriers, we aimed to develop an online implementation that produces spectra and metabolic maps within minutes after the acquisition. The ability to automatically reconstruct and visualize MRSI data at the scanner console in near real-time and without expert knowledge is a big step towards clinical applicability.
To overcome these barriers, we aimed to develop an online implementation that produces spectra and metabolic maps within minutes after the acquisition. The ability to automatically reconstruct and visualize MRSI data at the scanner console in near real-time and without expert knowledge is a big step towards clinical applicability.
Methods:
We designed the reconstruction pipeline modularly in the Siemens frameworks ICE, which is a reconstruction pipeline in C++, and FIRE, which provides a containerized environment to run custom programs. It currently supports the 3D concentric ring trajectory (CRT) (Hingerl et al, 2020), which may contain in addition to the MRSI Data an interleaved iMUSICAL reference scan (Moser et al, 2019) and a short prescan for coil compression. The reconstruction steps and the transition from ICE to FIRE can be seen in Figure 1.
For the online reconstruction, we had to address the issues of reconstruction time and memory requirements, ensuring identical results to the offline pipeline, and making the reconstruction process robust and automatic.
The solutions required to reduce the reconstruction time from hours of the reference offline MATLAB and LCModel (Provencher, 1993) reconstruction to minutes at the scanner are: 1) The spectral fitting is performed by a fast deep learning model instead of LCModel 2) The reconstruction starts while the sequence is running and flexibly reconstructs the steps that become possible with the currently acquired subset of the data 3) Performance profiling was applied to detect and optimize bottlenecks, implementing algorithmic improvements and fast linear algebra functions (Intel math kernel library MKL) 4) Dynamic parallel execution was ensured of currently possible reconstruction steps via the ICE Functor framework 5) Early coil compression was integrated to reduce the amount of data and computation by an additional factor of ~4.
For steps that were too complicated to implement in the ICE framework, we used python in the containerized FIRE environment running on the scanner reconstruction hardware. These include the brain-mask and lipid-mask creation, which requires sophisticated image processing steps, and the deep learning based spectral fitting. The fitting was performed by a physics-informed deep auto-encoder which uses a model-based decoder that is based on linear combination modeling (Shamaei, 2023).
A pipeline prototype without lipid suppression has been tested on one healthy volunteer at a Siemens 7T scanner with a matrix size of 64x64x39, 840 spectral points and 32 channels.
For the online reconstruction, we had to address the issues of reconstruction time and memory requirements, ensuring identical results to the offline pipeline, and making the reconstruction process robust and automatic.
The solutions required to reduce the reconstruction time from hours of the reference offline MATLAB and LCModel (Provencher, 1993) reconstruction to minutes at the scanner are: 1) The spectral fitting is performed by a fast deep learning model instead of LCModel 2) The reconstruction starts while the sequence is running and flexibly reconstructs the steps that become possible with the currently acquired subset of the data 3) Performance profiling was applied to detect and optimize bottlenecks, implementing algorithmic improvements and fast linear algebra functions (Intel math kernel library MKL) 4) Dynamic parallel execution was ensured of currently possible reconstruction steps via the ICE Functor framework 5) Early coil compression was integrated to reduce the amount of data and computation by an additional factor of ~4.
For steps that were too complicated to implement in the ICE framework, we used python in the containerized FIRE environment running on the scanner reconstruction hardware. These include the brain-mask and lipid-mask creation, which requires sophisticated image processing steps, and the deep learning based spectral fitting. The fitting was performed by a physics-informed deep auto-encoder which uses a model-based decoder that is based on linear combination modeling (Shamaei, 2023).
A pipeline prototype without lipid suppression has been tested on one healthy volunteer at a Siemens 7T scanner with a matrix size of 64x64x39, 840 spectral points and 32 channels.
·Figure 1: Online Reconstruction Pipeline
Results:
The resulting metabolic maps and corresponding CRLB maps could be directly viewed on the scanner console (Figure 2). All the online reconstruction steps, with the exclusion of metabolic fitting, were tested to yield identical results to the offline MATLAB pipeline within numerical tolerances.
·Figure 2: On-Console Output Images
Conclusions:
A further verification with five subjects is planned to ensure robustness and equivalence in quality.
The online implementation makes it straightforward to distribute the reconstruction together with the sequence, allowing easier sharing with other sites and removing the need for manual reconstruction. Combined with the possibility of live quality control at the scanner, this can aid a future transition of MRSI into clinical practice.
The online implementation makes it straightforward to distribute the reconstruction together with the sequence, allowing easier sharing with other sites and removing the need for manual reconstruction. Combined with the possibility of live quality control at the scanner, this can aid a future transition of MRSI into clinical practice.
Disorders of the Nervous System:
Neurodegenerative/ Late Life (eg. Parkinson’s, Alzheimer’s) 2
Neuroanatomy, Physiology, Metabolism and Neurotransmission:
Transmitter Systems
Novel Imaging Acquisition Methods:
MR Spectroscopy 1
Keywords:
Acquisition
CHEMOARCHITECTURE
Demyelinating
GABA
Glutamate
HIGH FIELD MR
Magnetic Resonance Spectroscopy (MRS)
Neurotransmitter
Workflows
Other - MRSI
1|2Indicates the priority used for review
Abstract Information
By submitting your proposal, you grant permission for the Organization for Human Brain Mapping (OHBM) to distribute your work in any format, including video, audio print and electronic text through OHBM OnDemand, social media channels, the OHBM website, or other electronic publications and media.
The Open Science Special Interest Group (OSSIG) is introducing a reproducibility challenge for OHBM 2025. This new initiative aims to enhance the reproducibility of scientific results and foster collaborations between labs. Teams will consist of a “source” party and a “reproducing” party, and will be evaluated on the success of their replication, the openness of the source work, and additional deliverables. Click here for more information. Propose your OHBM abstract(s) as source work for future OHBM meetings by selecting one of the following options:
Please indicate below if your study was a "resting state" or "task-activation” study.
Healthy subjects only or patients (note that patient studies may also involve healthy subjects):
Was this research conducted in the United States?
Were any human subjects research approved by the relevant Institutional Review Board or ethics panel? NOTE: Any human subjects studies without IRB approval will be automatically rejected.
Were any animal research approved by the relevant IACUC or other animal research panel? NOTE: Any animal studies without IACUC approval will be automatically rejected.
Please indicate which methods were used in your research:
For human MRI, what field strength scanner do you use?
Provide references using APA citation style.
Heckova, E., Strasser, B., Hangel, G. J., et al. (2019). 7T magnetic resonance spectroscopic imaging in multiple sclerosis: How does spatial resolution affect the detectability of metabolic changes in brain lesions? Investigative Radiology, 54(4), 247–254.
Hingerl, L., Strasser, B., Moser, P., et al. (2020). Clinical high-resolution 3D-MR spectroscopic imaging of the human brain at 7 T. Investigative Radiology.
Moser, P., Bogner, W., Hingerl, L., et al. (2019). Non-Cartesian GRAPPA and coil combination using interleaved calibration data: Application to concentric-ring MRSI of the human brain at 7T. Magnetic Resonance in Medicine, 82(5), 1587–1603.
Provencher, S. W. (1993). Estimation of metabolite concentrations from localized in vivo proton NMR spectra. Magnetic Resonance in Medicine, 30(6), 672–679.
Shamaei, A., Starcukova, J., & Starcuk, Z. Jr. (2023). Physics-informed deep learning approach to quantification of human brain metabolites from magnetic resonance spectroscopy data. Computers in Biology and Medicine, 158, 106837.
Veenendaal, T. M., Backes, W. H., Tse, D. H. Y., et al. (2018). High field imaging of large-scale neurotransmitter networks: Proof of concept and initial application to epilepsy. NeuroImage: Clinical, 19, 47–55.