2.5D Deep Learning for CT Image Reconstruction Using A Multi-GPU Implementation

Amirkoushyar Ziabari; Dong Hye Ye; Somesh Srivastava; Ken D. Sauer; Jean Baptiste Thibault; Charles A. Bouman

doi:10.1109/ACSSC.2018.8645364

2.5D Deep Learning for CT Image Reconstruction Using A Multi-GPU Implementation

Amirkoushyar Ziabari, Dong Hye Ye, Somesh Srivastava, Ken D. Sauer, Jean Baptiste Thibault, Charles A. Bouman

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

28 Scopus citations

Abstract

While Model Based Iterative Reconstruction (MBIR) of CT scans has been shown to have better image quality than Filtered Back Projection (FBP), its use has been limited by its high computational cost. More recently, deep convolutional neural networks (CNN) have shown great promise in both denoising and reconstruction applications. In this research, we propose a fast reconstruction algorithm, which we call Deep Learning MBIR (DL-MBIR), for approximating MBIR using a deep residual neural network. The DL-MBIR method is trained to produce reconstructions that approximate true MBIR images using a 16 layer residual convolutional neural network implemented on multiple GPUs using Google Tensorflow. In addition, we propose 2D, 2.5D and 3D variations on the DL-MBIR method and show that the 2.5D method achieves similar quality to the fully 3D method, but with reduced computational cost.

Original language	English
Title of host publication	Conference Record of the 52nd Asilomar Conference on Signals, Systems and Computers, ACSSC 2018
Editors	Michael B. Matthews
Publisher	IEEE Computer Society
Pages	2044-2049
Number of pages	6
ISBN (Electronic)	9781538692189
DOIs	https://doi.org/10.1109/ACSSC.2018.8645364
State	Published - Jul 2 2018
Externally published	Yes
Event	52nd Asilomar Conference on Signals, Systems and Computers, ACSSC 2018 - Pacific Grove, United States Duration: Oct 28 2018 → Oct 31 2018

Publication series

Name	Conference Record - Asilomar Conference on Signals, Systems and Computers
Volume	2018-October
ISSN (Print)	1058-6393

Conference

Conference	52nd Asilomar Conference on Signals, Systems and Computers, ACSSC 2018
Country/Territory	United States
City	Pacific Grove
Period	10/28/18 → 10/31/18

Keywords

2.5D DL-MBIR
Computed Tomography
Deep Learning (DL)
FBP
MBIR
Residual CNN

Access to Document

10.1109/ACSSC.2018.8645364

Cite this

Ziabari, A., Ye, D. H., Srivastava, S., Sauer, K. D., Thibault, J. B., & Bouman, C. A. (2018). 2.5D Deep Learning for CT Image Reconstruction Using A Multi-GPU Implementation. In M. B. Matthews (Ed.), Conference Record of the 52nd Asilomar Conference on Signals, Systems and Computers, ACSSC 2018 (pp. 2044-2049). Article 8645364 (Conference Record - Asilomar Conference on Signals, Systems and Computers; Vol. 2018-October). IEEE Computer Society. https://doi.org/10.1109/ACSSC.2018.8645364

Ziabari, Amirkoushyar ; Ye, Dong Hye ; Srivastava, Somesh et al. / 2.5D Deep Learning for CT Image Reconstruction Using A Multi-GPU Implementation. Conference Record of the 52nd Asilomar Conference on Signals, Systems and Computers, ACSSC 2018. editor / Michael B. Matthews. IEEE Computer Society, 2018. pp. 2044-2049 (Conference Record - Asilomar Conference on Signals, Systems and Computers).

@inproceedings{7c45afaa91c44bf4abee6d44305bca62,

title = "2.5D Deep Learning for CT Image Reconstruction Using A Multi-GPU Implementation",

abstract = "While Model Based Iterative Reconstruction (MBIR) of CT scans has been shown to have better image quality than Filtered Back Projection (FBP), its use has been limited by its high computational cost. More recently, deep convolutional neural networks (CNN) have shown great promise in both denoising and reconstruction applications. In this research, we propose a fast reconstruction algorithm, which we call Deep Learning MBIR (DL-MBIR), for approximating MBIR using a deep residual neural network. The DL-MBIR method is trained to produce reconstructions that approximate true MBIR images using a 16 layer residual convolutional neural network implemented on multiple GPUs using Google Tensorflow. In addition, we propose 2D, 2.5D and 3D variations on the DL-MBIR method and show that the 2.5D method achieves similar quality to the fully 3D method, but with reduced computational cost.",

keywords = "2.5D DL-MBIR, Computed Tomography, Deep Learning (DL), FBP, MBIR, Residual CNN",

author = "Amirkoushyar Ziabari and Ye, {Dong Hye} and Somesh Srivastava and Sauer, {Ken D.} and Thibault, {Jean Baptiste} and Bouman, {Charles A.}",

note = "Publisher Copyright: {\textcopyright} 2018 IEEE.; 52nd Asilomar Conference on Signals, Systems and Computers, ACSSC 2018 ; Conference date: 28-10-2018 Through 31-10-2018",

year = "2018",

month = jul,

day = "2",

doi = "10.1109/ACSSC.2018.8645364",

language = "English",

series = "Conference Record - Asilomar Conference on Signals, Systems and Computers",

publisher = "IEEE Computer Society",

pages = "2044--2049",

editor = "Matthews, {Michael B.}",

booktitle = "Conference Record of the 52nd Asilomar Conference on Signals, Systems and Computers, ACSSC 2018",

}

Ziabari, A, Ye, DH, Srivastava, S, Sauer, KD, Thibault, JB & Bouman, CA 2018, 2.5D Deep Learning for CT Image Reconstruction Using A Multi-GPU Implementation. in MB Matthews (ed.), Conference Record of the 52nd Asilomar Conference on Signals, Systems and Computers, ACSSC 2018., 8645364, Conference Record - Asilomar Conference on Signals, Systems and Computers, vol. 2018-October, IEEE Computer Society, pp. 2044-2049, 52nd Asilomar Conference on Signals, Systems and Computers, ACSSC 2018, Pacific Grove, United States, 10/28/18. https://doi.org/10.1109/ACSSC.2018.8645364

2.5D Deep Learning for CT Image Reconstruction Using A Multi-GPU Implementation. / Ziabari, Amirkoushyar; Ye, Dong Hye; Srivastava, Somesh et al.
Conference Record of the 52nd Asilomar Conference on Signals, Systems and Computers, ACSSC 2018. ed. / Michael B. Matthews. IEEE Computer Society, 2018. p. 2044-2049 8645364 (Conference Record - Asilomar Conference on Signals, Systems and Computers; Vol. 2018-October).

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

TY - GEN

T1 - 2.5D Deep Learning for CT Image Reconstruction Using A Multi-GPU Implementation

AU - Ziabari, Amirkoushyar

AU - Ye, Dong Hye

AU - Srivastava, Somesh

AU - Sauer, Ken D.

AU - Thibault, Jean Baptiste

AU - Bouman, Charles A.

PY - 2018/7/2

Y1 - 2018/7/2

N2 - While Model Based Iterative Reconstruction (MBIR) of CT scans has been shown to have better image quality than Filtered Back Projection (FBP), its use has been limited by its high computational cost. More recently, deep convolutional neural networks (CNN) have shown great promise in both denoising and reconstruction applications. In this research, we propose a fast reconstruction algorithm, which we call Deep Learning MBIR (DL-MBIR), for approximating MBIR using a deep residual neural network. The DL-MBIR method is trained to produce reconstructions that approximate true MBIR images using a 16 layer residual convolutional neural network implemented on multiple GPUs using Google Tensorflow. In addition, we propose 2D, 2.5D and 3D variations on the DL-MBIR method and show that the 2.5D method achieves similar quality to the fully 3D method, but with reduced computational cost.

AB - While Model Based Iterative Reconstruction (MBIR) of CT scans has been shown to have better image quality than Filtered Back Projection (FBP), its use has been limited by its high computational cost. More recently, deep convolutional neural networks (CNN) have shown great promise in both denoising and reconstruction applications. In this research, we propose a fast reconstruction algorithm, which we call Deep Learning MBIR (DL-MBIR), for approximating MBIR using a deep residual neural network. The DL-MBIR method is trained to produce reconstructions that approximate true MBIR images using a 16 layer residual convolutional neural network implemented on multiple GPUs using Google Tensorflow. In addition, we propose 2D, 2.5D and 3D variations on the DL-MBIR method and show that the 2.5D method achieves similar quality to the fully 3D method, but with reduced computational cost.

KW - 2.5D DL-MBIR

KW - Computed Tomography

KW - Deep Learning (DL)

KW - FBP

KW - MBIR

KW - Residual CNN

UR - http://www.scopus.com/inward/record.url?scp=85062938754&partnerID=8YFLogxK

U2 - 10.1109/ACSSC.2018.8645364

DO - 10.1109/ACSSC.2018.8645364

M3 - Conference contribution

AN - SCOPUS:85062938754

T3 - Conference Record - Asilomar Conference on Signals, Systems and Computers

SP - 2044

EP - 2049

BT - Conference Record of the 52nd Asilomar Conference on Signals, Systems and Computers, ACSSC 2018

A2 - Matthews, Michael B.

PB - IEEE Computer Society

T2 - 52nd Asilomar Conference on Signals, Systems and Computers, ACSSC 2018

Y2 - 28 October 2018 through 31 October 2018

ER -

Ziabari A, Ye DH, Srivastava S, Sauer KD, Thibault JB, Bouman CA. 2.5D Deep Learning for CT Image Reconstruction Using A Multi-GPU Implementation. In Matthews MB, editor, Conference Record of the 52nd Asilomar Conference on Signals, Systems and Computers, ACSSC 2018. IEEE Computer Society. 2018. p. 2044-2049. 8645364. (Conference Record - Asilomar Conference on Signals, Systems and Computers). doi: 10.1109/ACSSC.2018.8645364

2.5D Deep Learning for CT Image Reconstruction Using A Multi-GPU Implementation

Abstract

Publication series

Conference

Keywords

Access to Document

Other files and links

Fingerprint

Cite this