KIKI-net

cross-domain convolutional neural networks for reconstructing undersampled magnetic resonance images

Taejoon Eo, Yohan Jun, Taeseong Kim, Jinseong Jang, Ho Joon Lee, Do Sik Hwang

Research output: Contribution to journalArticle

10 Citations (Scopus)

Abstract

Purpose: To demonstrate accurate MR image reconstruction from undersampled k-space data using cross-domain convolutional neural networks (CNNs). Methods: Cross-domain CNNs consist of 3 components: (1) a deep CNN operating on the k-space (KCNN), (2) a deep CNN operating on an image domain (ICNN), and (3) an interleaved data consistency operations. These components are alternately applied, and each CNN is trained to minimize the loss between the reconstructed and corresponding fully sampled k-spaces. The final reconstructed image is obtained by forward-propagating the undersampled k-space data through the entire network. Results: Performances of K-net (KCNN with inverse Fourier transform), I-net (ICNN with interleaved data consistency), and various combinations of the 2 different networks were tested. The test results indicated that K-net and I-net have different advantages/disadvantages in terms of tissue-structure restoration. Consequently, the combination of K-net and I-net is superior to single-domain CNNs. Three MR data sets, the T2 fluid-attenuated inversion recovery (T2 FLAIR) set from the Alzheimer's Disease Neuroimaging Initiative and 2 data sets acquired at our local institute (T2 FLAIR and T1 weighted), were used to evaluate the performance of 7 conventional reconstruction algorithms and the proposed cross-domain CNNs, which hereafter is referred to as KIKI-net. KIKI-net outperforms conventional algorithms with mean improvements of 2.29 dB in peak SNR and 0.031 in structure similarity. Conclusion: KIKI-net exhibits superior performance over state-of-the-art conventional algorithms in terms of restoring tissue structures and removing aliasing artifacts. The results demonstrate that KIKI-net is applicable up to a reduction factor of 3 to 4 based on variable-density Cartesian undersampling.

Original languageEnglish
Pages (from-to)2188-2201
Number of pages14
JournalMagnetic Resonance in Medicine
Volume80
Issue number5
DOIs
Publication statusPublished - 2018 Nov 1

Fingerprint

Magnetic Resonance Spectroscopy
Computer-Assisted Image Processing
Fourier Analysis
Neuroimaging
Artifacts
Alzheimer Disease
Datasets

All Science Journal Classification (ASJC) codes

  • Radiology Nuclear Medicine and imaging

Cite this

Eo, Taejoon ; Jun, Yohan ; Kim, Taeseong ; Jang, Jinseong ; Lee, Ho Joon ; Hwang, Do Sik. / KIKI-net : cross-domain convolutional neural networks for reconstructing undersampled magnetic resonance images. In: Magnetic Resonance in Medicine. 2018 ; Vol. 80, No. 5. pp. 2188-2201.
@article{8d12e7109a6b432592054b10032181cd,
title = "KIKI-net: cross-domain convolutional neural networks for reconstructing undersampled magnetic resonance images",
abstract = "Purpose: To demonstrate accurate MR image reconstruction from undersampled k-space data using cross-domain convolutional neural networks (CNNs). Methods: Cross-domain CNNs consist of 3 components: (1) a deep CNN operating on the k-space (KCNN), (2) a deep CNN operating on an image domain (ICNN), and (3) an interleaved data consistency operations. These components are alternately applied, and each CNN is trained to minimize the loss between the reconstructed and corresponding fully sampled k-spaces. The final reconstructed image is obtained by forward-propagating the undersampled k-space data through the entire network. Results: Performances of K-net (KCNN with inverse Fourier transform), I-net (ICNN with interleaved data consistency), and various combinations of the 2 different networks were tested. The test results indicated that K-net and I-net have different advantages/disadvantages in terms of tissue-structure restoration. Consequently, the combination of K-net and I-net is superior to single-domain CNNs. Three MR data sets, the T2 fluid-attenuated inversion recovery (T2 FLAIR) set from the Alzheimer's Disease Neuroimaging Initiative and 2 data sets acquired at our local institute (T2 FLAIR and T1 weighted), were used to evaluate the performance of 7 conventional reconstruction algorithms and the proposed cross-domain CNNs, which hereafter is referred to as KIKI-net. KIKI-net outperforms conventional algorithms with mean improvements of 2.29 dB in peak SNR and 0.031 in structure similarity. Conclusion: KIKI-net exhibits superior performance over state-of-the-art conventional algorithms in terms of restoring tissue structures and removing aliasing artifacts. The results demonstrate that KIKI-net is applicable up to a reduction factor of 3 to 4 based on variable-density Cartesian undersampling.",
author = "Taejoon Eo and Yohan Jun and Taeseong Kim and Jinseong Jang and Lee, {Ho Joon} and Hwang, {Do Sik}",
year = "2018",
month = "11",
day = "1",
doi = "10.1002/mrm.27201",
language = "English",
volume = "80",
pages = "2188--2201",
journal = "Magnetic Resonance in Medicine",
issn = "0740-3194",
publisher = "John Wiley and Sons Inc.",
number = "5",

}

KIKI-net : cross-domain convolutional neural networks for reconstructing undersampled magnetic resonance images. / Eo, Taejoon; Jun, Yohan; Kim, Taeseong; Jang, Jinseong; Lee, Ho Joon; Hwang, Do Sik.

In: Magnetic Resonance in Medicine, Vol. 80, No. 5, 01.11.2018, p. 2188-2201.

Research output: Contribution to journalArticle

TY - JOUR

T1 - KIKI-net

T2 - cross-domain convolutional neural networks for reconstructing undersampled magnetic resonance images

AU - Eo, Taejoon

AU - Jun, Yohan

AU - Kim, Taeseong

AU - Jang, Jinseong

AU - Lee, Ho Joon

AU - Hwang, Do Sik

PY - 2018/11/1

Y1 - 2018/11/1

N2 - Purpose: To demonstrate accurate MR image reconstruction from undersampled k-space data using cross-domain convolutional neural networks (CNNs). Methods: Cross-domain CNNs consist of 3 components: (1) a deep CNN operating on the k-space (KCNN), (2) a deep CNN operating on an image domain (ICNN), and (3) an interleaved data consistency operations. These components are alternately applied, and each CNN is trained to minimize the loss between the reconstructed and corresponding fully sampled k-spaces. The final reconstructed image is obtained by forward-propagating the undersampled k-space data through the entire network. Results: Performances of K-net (KCNN with inverse Fourier transform), I-net (ICNN with interleaved data consistency), and various combinations of the 2 different networks were tested. The test results indicated that K-net and I-net have different advantages/disadvantages in terms of tissue-structure restoration. Consequently, the combination of K-net and I-net is superior to single-domain CNNs. Three MR data sets, the T2 fluid-attenuated inversion recovery (T2 FLAIR) set from the Alzheimer's Disease Neuroimaging Initiative and 2 data sets acquired at our local institute (T2 FLAIR and T1 weighted), were used to evaluate the performance of 7 conventional reconstruction algorithms and the proposed cross-domain CNNs, which hereafter is referred to as KIKI-net. KIKI-net outperforms conventional algorithms with mean improvements of 2.29 dB in peak SNR and 0.031 in structure similarity. Conclusion: KIKI-net exhibits superior performance over state-of-the-art conventional algorithms in terms of restoring tissue structures and removing aliasing artifacts. The results demonstrate that KIKI-net is applicable up to a reduction factor of 3 to 4 based on variable-density Cartesian undersampling.

AB - Purpose: To demonstrate accurate MR image reconstruction from undersampled k-space data using cross-domain convolutional neural networks (CNNs). Methods: Cross-domain CNNs consist of 3 components: (1) a deep CNN operating on the k-space (KCNN), (2) a deep CNN operating on an image domain (ICNN), and (3) an interleaved data consistency operations. These components are alternately applied, and each CNN is trained to minimize the loss between the reconstructed and corresponding fully sampled k-spaces. The final reconstructed image is obtained by forward-propagating the undersampled k-space data through the entire network. Results: Performances of K-net (KCNN with inverse Fourier transform), I-net (ICNN with interleaved data consistency), and various combinations of the 2 different networks were tested. The test results indicated that K-net and I-net have different advantages/disadvantages in terms of tissue-structure restoration. Consequently, the combination of K-net and I-net is superior to single-domain CNNs. Three MR data sets, the T2 fluid-attenuated inversion recovery (T2 FLAIR) set from the Alzheimer's Disease Neuroimaging Initiative and 2 data sets acquired at our local institute (T2 FLAIR and T1 weighted), were used to evaluate the performance of 7 conventional reconstruction algorithms and the proposed cross-domain CNNs, which hereafter is referred to as KIKI-net. KIKI-net outperforms conventional algorithms with mean improvements of 2.29 dB in peak SNR and 0.031 in structure similarity. Conclusion: KIKI-net exhibits superior performance over state-of-the-art conventional algorithms in terms of restoring tissue structures and removing aliasing artifacts. The results demonstrate that KIKI-net is applicable up to a reduction factor of 3 to 4 based on variable-density Cartesian undersampling.

UR - http://www.scopus.com/inward/record.url?scp=85044921256&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85044921256&partnerID=8YFLogxK

U2 - 10.1002/mrm.27201

DO - 10.1002/mrm.27201

M3 - Article

VL - 80

SP - 2188

EP - 2201

JO - Magnetic Resonance in Medicine

JF - Magnetic Resonance in Medicine

SN - 0740-3194

IS - 5

ER -