Distributed Memory-Efficient Physics-Guided Deep Learning Reconstruction for Large-Scale 3d Non-Cartesian MRI

Chi Zhang; Davide Piccini; Omer Burak Demirel; Gabriele Bonanno; Burhaneddin Yaman; Matthias Stuber; Steen Moeller; Mehmet Akcakaya

doi:10.1109/ISBI52829.2022.9761485

Distributed Memory-Efficient Physics-Guided Deep Learning Reconstruction for Large-Scale 3d Non-Cartesian MRI

Chi Zhang
, Davide Piccini
, Omer Burak Demirel
, Gabriele Bonanno
, Burhaneddin Yaman
, Matthias Stuber
, Steen Moeller
, Mehmet Akcakaya

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution

8 Scopus citations

Abstract

Physics-guided deep learning (PG-DL) reconstruction has emerged as a powerful strategy for accelerated MRI. However, adopting PG-DL on 3D non-Cartesian MRI remains a challenge due to GPU hardware limitations. In this paper, we utilize multiple memory-efficient techniques to accomplish PG-DL on large-scale 3D kooshball coronary MRI. We first leverage a recently proposed approach to keep only one unrolled step on GPUs. We then utilize a Toeplitz approach to represent the multi-coil encoding operator. Subsequently, we distribute the most memory-consuming data consistency operations into multiple GPUs, enabling conjugate gradient iterations without necessitating coil compression. Finally, we employ mixed-precision training to further reduce memory consumption.The combination of these methods enable training of high-quality PG-DL reconstruction for 3D kooshball trajectories, and our results show reconstruction improvement compared to existing strategies.

Original language	English (US)
Title of host publication	IEEE ISBI 2022 Proceedings - 2022 IEEE International Symposium on Biomedical Imaging
Publisher	IEEE Computer Society
ISBN (Electronic)	9781665429238
DOIs	https://doi.org/10.1109/ISBI52829.2022.9761485
State	Published - 2022
Event	19th IEEE International Symposium on Biomedical Imaging, ISBI 2022 - Hybrid, Kolkata, India Duration: Mar 28 2022 → Mar 31 2022

Publication series

Name	2022 IEEE 19th International Symposium on Biomedical Imaging (ISBI)

Conference

Conference	19th IEEE International Symposium on Biomedical Imaging, ISBI 2022
Country/Territory	India
City	Hybrid, Kolkata
Period	3/28/22 → 3/31/22

Bibliographical note

Funding Information:
This work was partially supported by NIH R01HL153146, NIH P41EB027061, NIH R21EB028369, NSF CAREER CCF-1651825.

Publisher Copyright:
© 2022 IEEE.

Keywords

GPU
accelerated imaging
deep learning
implementation
non-Cartesian MRI

Center for Magnetic Resonance Research (CMRR) tags

IRP

Publisher link

10.1109/ISBI52829.2022.9761485

Find It

Find It at U of M Libraries

Cite this

Zhang, C., Piccini, D., Demirel, O. B., Bonanno, G., Yaman, B., Stuber, M., Moeller, S., & Akcakaya, M. (2022). Distributed Memory-Efficient Physics-Guided Deep Learning Reconstruction for Large-Scale 3d Non-Cartesian MRI. In IEEE ISBI 2022 Proceedings - 2022 IEEE International Symposium on Biomedical Imaging (2022 IEEE 19th International Symposium on Biomedical Imaging (ISBI)). IEEE Computer Society. https://doi.org/10.1109/ISBI52829.2022.9761485

Distributed Memory-Efficient Physics-Guided Deep Learning Reconstruction for Large-Scale 3d Non-Cartesian MRI. / Zhang, Chi; Piccini, Davide; Demirel, Omer Burak et al.
IEEE ISBI 2022 Proceedings - 2022 IEEE International Symposium on Biomedical Imaging. IEEE Computer Society, 2022. (2022 IEEE 19th International Symposium on Biomedical Imaging (ISBI)).

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution

Zhang, C, Piccini, D, Demirel, OB, Bonanno, G, Yaman, B, Stuber, M, Moeller, S & Akcakaya, M 2022, Distributed Memory-Efficient Physics-Guided Deep Learning Reconstruction for Large-Scale 3d Non-Cartesian MRI. in IEEE ISBI 2022 Proceedings - 2022 IEEE International Symposium on Biomedical Imaging. 2022 IEEE 19th International Symposium on Biomedical Imaging (ISBI), IEEE Computer Society, 19th IEEE International Symposium on Biomedical Imaging, ISBI 2022, Hybrid, Kolkata, India, 3/28/22. https://doi.org/10.1109/ISBI52829.2022.9761485

Zhang C, Piccini D, Demirel OB, Bonanno G, Yaman B, Stuber M et al. Distributed Memory-Efficient Physics-Guided Deep Learning Reconstruction for Large-Scale 3d Non-Cartesian MRI. In IEEE ISBI 2022 Proceedings - 2022 IEEE International Symposium on Biomedical Imaging. IEEE Computer Society. 2022. (2022 IEEE 19th International Symposium on Biomedical Imaging (ISBI)). doi: 10.1109/ISBI52829.2022.9761485

@inproceedings{54f51c99144c4354bc9a5aed0ccc90d7,

title = "Distributed Memory-Efficient Physics-Guided Deep Learning Reconstruction for Large-Scale 3d Non-Cartesian MRI",

abstract = "Physics-guided deep learning (PG-DL) reconstruction has emerged as a powerful strategy for accelerated MRI. However, adopting PG-DL on 3D non-Cartesian MRI remains a challenge due to GPU hardware limitations. In this paper, we utilize multiple memory-efficient techniques to accomplish PG-DL on large-scale 3D kooshball coronary MRI. We first leverage a recently proposed approach to keep only one unrolled step on GPUs. We then utilize a Toeplitz approach to represent the multi-coil encoding operator. Subsequently, we distribute the most memory-consuming data consistency operations into multiple GPUs, enabling conjugate gradient iterations without necessitating coil compression. Finally, we employ mixed-precision training to further reduce memory consumption.The combination of these methods enable training of high-quality PG-DL reconstruction for 3D kooshball trajectories, and our results show reconstruction improvement compared to existing strategies.",

keywords = "GPU, accelerated imaging, deep learning, implementation, non-Cartesian MRI",

author = "Chi Zhang and Davide Piccini and Demirel, \{Omer Burak\} and Gabriele Bonanno and Burhaneddin Yaman and Matthias Stuber and Steen Moeller and Mehmet Akcakaya",

note = "Funding Information: This work was partially supported by NIH R01HL153146, NIH P41EB027061, NIH R21EB028369, NSF CAREER CCF-1651825. Publisher Copyright: {\textcopyright} 2022 IEEE.; 19th IEEE International Symposium on Biomedical Imaging, ISBI 2022 ; Conference date: 28-03-2022 Through 31-03-2022",

year = "2022",

doi = "10.1109/ISBI52829.2022.9761485",

language = "English (US)",

series = "2022 IEEE 19th International Symposium on Biomedical Imaging (ISBI)",

publisher = "IEEE Computer Society",

booktitle = "IEEE ISBI 2022 Proceedings - 2022 IEEE International Symposium on Biomedical Imaging",

}

TY - GEN

T1 - Distributed Memory-Efficient Physics-Guided Deep Learning Reconstruction for Large-Scale 3d Non-Cartesian MRI

AU - Zhang, Chi

AU - Piccini, Davide

AU - Demirel, Omer Burak

AU - Bonanno, Gabriele

AU - Yaman, Burhaneddin

AU - Stuber, Matthias

AU - Moeller, Steen

AU - Akcakaya, Mehmet

PY - 2022

Y1 - 2022

N2 - Physics-guided deep learning (PG-DL) reconstruction has emerged as a powerful strategy for accelerated MRI. However, adopting PG-DL on 3D non-Cartesian MRI remains a challenge due to GPU hardware limitations. In this paper, we utilize multiple memory-efficient techniques to accomplish PG-DL on large-scale 3D kooshball coronary MRI. We first leverage a recently proposed approach to keep only one unrolled step on GPUs. We then utilize a Toeplitz approach to represent the multi-coil encoding operator. Subsequently, we distribute the most memory-consuming data consistency operations into multiple GPUs, enabling conjugate gradient iterations without necessitating coil compression. Finally, we employ mixed-precision training to further reduce memory consumption.The combination of these methods enable training of high-quality PG-DL reconstruction for 3D kooshball trajectories, and our results show reconstruction improvement compared to existing strategies.

AB - Physics-guided deep learning (PG-DL) reconstruction has emerged as a powerful strategy for accelerated MRI. However, adopting PG-DL on 3D non-Cartesian MRI remains a challenge due to GPU hardware limitations. In this paper, we utilize multiple memory-efficient techniques to accomplish PG-DL on large-scale 3D kooshball coronary MRI. We first leverage a recently proposed approach to keep only one unrolled step on GPUs. We then utilize a Toeplitz approach to represent the multi-coil encoding operator. Subsequently, we distribute the most memory-consuming data consistency operations into multiple GPUs, enabling conjugate gradient iterations without necessitating coil compression. Finally, we employ mixed-precision training to further reduce memory consumption.The combination of these methods enable training of high-quality PG-DL reconstruction for 3D kooshball trajectories, and our results show reconstruction improvement compared to existing strategies.

KW - GPU

KW - accelerated imaging

KW - deep learning

KW - implementation

KW - non-Cartesian MRI

UR - https://www.scopus.com/pages/publications/85129638948

UR - https://www.scopus.com/pages/publications/85129638948#tab=citedBy

U2 - 10.1109/ISBI52829.2022.9761485

DO - 10.1109/ISBI52829.2022.9761485

M3 - Conference contribution

AN - SCOPUS:85129638948

T3 - 2022 IEEE 19th International Symposium on Biomedical Imaging (ISBI)

BT - IEEE ISBI 2022 Proceedings - 2022 IEEE International Symposium on Biomedical Imaging

PB - IEEE Computer Society

T2 - 19th IEEE International Symposium on Biomedical Imaging, ISBI 2022

Y2 - 28 March 2022 through 31 March 2022

ER -

Distributed Memory-Efficient Physics-Guided Deep Learning Reconstruction for Large-Scale 3d Non-Cartesian MRI

Abstract

Publication series

Conference

Bibliographical note

Keywords

Center for Magnetic Resonance Research (CMRR) tags

Publisher link

Find It

Other files and links

Fingerprint

Cite this