Fast GPU Implementation of a Scan-Specific Deep Learning Reconstruction for Accelerated Magnetic Resonance Imaging

Chi Zhang, Sebastian Weingartner, Steen Moeller, Kamil Ugurbil, Mehmet Akcakaya

Research output: Chapter in Book/Report/Conference proceedingConference contribution

1 Citation (Scopus)

Abstract

Parallel Imaging is a technique commonly used in most clinical Magnetic Resonance Imaging (MRI) scans to mitigate the problem of long scan-times. In parallel imaging, information from multiple receiver antennas with different spatial sensitivities is combined to allow reconstruction from undersampled image information. Robust Artificial-neural-networks for k-space Interpolation (RAKI) has been recently proposed enabling parallel imaging reconstruction in MRI using convolutional neural networks (CNN) trained solely on a calibration signal corresponding to that image. While RAKI has demonstrated improved reconstruction performance compared to established techniques, its reconstruction time is prolonged due to the repeated application of the CNN, and the necessity of a training-phase for each receiver image. In this study, we propose an optimized RAKI implementation based on GPU parallel programming. The training phase duration is substantially shortened by optimizing the number of iterations and allowing for adaptively updated learning rates without compromising visual reconstruction quality. Efficient use of GPU resources was facilitated by a parallelized implementation of the training of multiple networks using CPU multiprocessing. The proposed implementation demonstrates more than 60-fold reduction in the reconstruction speed of clinical sample data compared with conventional sequential implementation, thus, easing the integration of RAKI in clinical applications.

Original languageEnglish (US)
Title of host publication2018 IEEE International Conference on Electro/Information Technology, EIT 2018
PublisherIEEE Computer Society
Pages399-403
Number of pages5
ISBN (Electronic)9781538653982
DOIs
StatePublished - Oct 18 2018
Event2018 IEEE International Conference on Electro/Information Technology, EIT 2018 - Rochester, United States
Duration: May 3 2018May 5 2018

Publication series

NameIEEE International Conference on Electro Information Technology
Volume2018-May
ISSN (Print)2154-0357
ISSN (Electronic)2154-0373

Other

Other2018 IEEE International Conference on Electro/Information Technology, EIT 2018
CountryUnited States
CityRochester
Period5/3/185/5/18

Fingerprint

Neural networks
Interpolation
Imaging techniques
Parallel programming
Program processors
Magnetic Resonance Imaging
Deep learning
Graphics processing unit
Calibration
Antennas

Cite this

Zhang, C., Weingartner, S., Moeller, S., Ugurbil, K., & Akcakaya, M. (2018). Fast GPU Implementation of a Scan-Specific Deep Learning Reconstruction for Accelerated Magnetic Resonance Imaging. In 2018 IEEE International Conference on Electro/Information Technology, EIT 2018 (pp. 399-403). [8500090] (IEEE International Conference on Electro Information Technology; Vol. 2018-May). IEEE Computer Society. https://doi.org/10.1109/EIT.2018.8500090

Fast GPU Implementation of a Scan-Specific Deep Learning Reconstruction for Accelerated Magnetic Resonance Imaging. / Zhang, Chi; Weingartner, Sebastian; Moeller, Steen; Ugurbil, Kamil; Akcakaya, Mehmet.

2018 IEEE International Conference on Electro/Information Technology, EIT 2018. IEEE Computer Society, 2018. p. 399-403 8500090 (IEEE International Conference on Electro Information Technology; Vol. 2018-May).

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Zhang, C, Weingartner, S, Moeller, S, Ugurbil, K & Akcakaya, M 2018, Fast GPU Implementation of a Scan-Specific Deep Learning Reconstruction for Accelerated Magnetic Resonance Imaging. in 2018 IEEE International Conference on Electro/Information Technology, EIT 2018., 8500090, IEEE International Conference on Electro Information Technology, vol. 2018-May, IEEE Computer Society, pp. 399-403, 2018 IEEE International Conference on Electro/Information Technology, EIT 2018, Rochester, United States, 5/3/18. https://doi.org/10.1109/EIT.2018.8500090
Zhang C, Weingartner S, Moeller S, Ugurbil K, Akcakaya M. Fast GPU Implementation of a Scan-Specific Deep Learning Reconstruction for Accelerated Magnetic Resonance Imaging. In 2018 IEEE International Conference on Electro/Information Technology, EIT 2018. IEEE Computer Society. 2018. p. 399-403. 8500090. (IEEE International Conference on Electro Information Technology). https://doi.org/10.1109/EIT.2018.8500090
Zhang, Chi ; Weingartner, Sebastian ; Moeller, Steen ; Ugurbil, Kamil ; Akcakaya, Mehmet. / Fast GPU Implementation of a Scan-Specific Deep Learning Reconstruction for Accelerated Magnetic Resonance Imaging. 2018 IEEE International Conference on Electro/Information Technology, EIT 2018. IEEE Computer Society, 2018. pp. 399-403 (IEEE International Conference on Electro Information Technology).
@inproceedings{4473d62e88bf4ec689e64ced40969850,
title = "Fast GPU Implementation of a Scan-Specific Deep Learning Reconstruction for Accelerated Magnetic Resonance Imaging",
abstract = "Parallel Imaging is a technique commonly used in most clinical Magnetic Resonance Imaging (MRI) scans to mitigate the problem of long scan-times. In parallel imaging, information from multiple receiver antennas with different spatial sensitivities is combined to allow reconstruction from undersampled image information. Robust Artificial-neural-networks for k-space Interpolation (RAKI) has been recently proposed enabling parallel imaging reconstruction in MRI using convolutional neural networks (CNN) trained solely on a calibration signal corresponding to that image. While RAKI has demonstrated improved reconstruction performance compared to established techniques, its reconstruction time is prolonged due to the repeated application of the CNN, and the necessity of a training-phase for each receiver image. In this study, we propose an optimized RAKI implementation based on GPU parallel programming. The training phase duration is substantially shortened by optimizing the number of iterations and allowing for adaptively updated learning rates without compromising visual reconstruction quality. Efficient use of GPU resources was facilitated by a parallelized implementation of the training of multiple networks using CPU multiprocessing. The proposed implementation demonstrates more than 60-fold reduction in the reconstruction speed of clinical sample data compared with conventional sequential implementation, thus, easing the integration of RAKI in clinical applications.",
author = "Chi Zhang and Sebastian Weingartner and Steen Moeller and Kamil Ugurbil and Mehmet Akcakaya",
year = "2018",
month = "10",
day = "18",
doi = "10.1109/EIT.2018.8500090",
language = "English (US)",
series = "IEEE International Conference on Electro Information Technology",
publisher = "IEEE Computer Society",
pages = "399--403",
booktitle = "2018 IEEE International Conference on Electro/Information Technology, EIT 2018",

}

TY - GEN

T1 - Fast GPU Implementation of a Scan-Specific Deep Learning Reconstruction for Accelerated Magnetic Resonance Imaging

AU - Zhang, Chi

AU - Weingartner, Sebastian

AU - Moeller, Steen

AU - Ugurbil, Kamil

AU - Akcakaya, Mehmet

PY - 2018/10/18

Y1 - 2018/10/18

N2 - Parallel Imaging is a technique commonly used in most clinical Magnetic Resonance Imaging (MRI) scans to mitigate the problem of long scan-times. In parallel imaging, information from multiple receiver antennas with different spatial sensitivities is combined to allow reconstruction from undersampled image information. Robust Artificial-neural-networks for k-space Interpolation (RAKI) has been recently proposed enabling parallel imaging reconstruction in MRI using convolutional neural networks (CNN) trained solely on a calibration signal corresponding to that image. While RAKI has demonstrated improved reconstruction performance compared to established techniques, its reconstruction time is prolonged due to the repeated application of the CNN, and the necessity of a training-phase for each receiver image. In this study, we propose an optimized RAKI implementation based on GPU parallel programming. The training phase duration is substantially shortened by optimizing the number of iterations and allowing for adaptively updated learning rates without compromising visual reconstruction quality. Efficient use of GPU resources was facilitated by a parallelized implementation of the training of multiple networks using CPU multiprocessing. The proposed implementation demonstrates more than 60-fold reduction in the reconstruction speed of clinical sample data compared with conventional sequential implementation, thus, easing the integration of RAKI in clinical applications.

AB - Parallel Imaging is a technique commonly used in most clinical Magnetic Resonance Imaging (MRI) scans to mitigate the problem of long scan-times. In parallel imaging, information from multiple receiver antennas with different spatial sensitivities is combined to allow reconstruction from undersampled image information. Robust Artificial-neural-networks for k-space Interpolation (RAKI) has been recently proposed enabling parallel imaging reconstruction in MRI using convolutional neural networks (CNN) trained solely on a calibration signal corresponding to that image. While RAKI has demonstrated improved reconstruction performance compared to established techniques, its reconstruction time is prolonged due to the repeated application of the CNN, and the necessity of a training-phase for each receiver image. In this study, we propose an optimized RAKI implementation based on GPU parallel programming. The training phase duration is substantially shortened by optimizing the number of iterations and allowing for adaptively updated learning rates without compromising visual reconstruction quality. Efficient use of GPU resources was facilitated by a parallelized implementation of the training of multiple networks using CPU multiprocessing. The proposed implementation demonstrates more than 60-fold reduction in the reconstruction speed of clinical sample data compared with conventional sequential implementation, thus, easing the integration of RAKI in clinical applications.

UR - http://www.scopus.com/inward/record.url?scp=85057089772&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85057089772&partnerID=8YFLogxK

U2 - 10.1109/EIT.2018.8500090

DO - 10.1109/EIT.2018.8500090

M3 - Conference contribution

AN - SCOPUS:85057089772

T3 - IEEE International Conference on Electro Information Technology

SP - 399

EP - 403

BT - 2018 IEEE International Conference on Electro/Information Technology, EIT 2018

PB - IEEE Computer Society

ER -