Enhancing SNR of CEST MRI Through Noise-to-Noise Deep Learning With K-Space Data Consistency

Huabing LIU; Ziyi XIA; Lok Hin LAW; Zilin CHEN; Jianpan HUANG; Dinggang SHEN; Kannie W.Y. CHAN

doi:10.58530/2025/2296

Enhancing SNR of CEST MRI Through Noise-to-Noise Deep Learning With K-Space Data Consistency

Huabing LIU (Co-first Author)
, Ziyi XIA (Co-first Author)
, Lok Hin LAW
, Zilin CHEN
, Jianpan HUANG
, Dinggang SHEN
, Kannie W.Y. CHAN^*

^*Corresponding author for this work

Department of Biomedical Engineering

Research output: Conference Papers › RGC 32 - Refereed conference paper (without host publication) › peer-review

Abstract

Motivation: Endogenous CEST contrast is relatively small and vulnerable to imaging noise.

Goal(s): To develop a retrospective denoising method to enhance SNR of acquired CEST images.

Approach: A denoising neural network was trained using pairs of noisy CEST images through a noise-to-noise deep learning approach, distinct from conventional approaches that use clean or simulated CEST images. A data consistency layer was introduced to preserve center k-space of original CEST images to improve fidelity. A transformer module was used to exploit spatiotemporal correlations among different frequency offsets.

Results: Multipool Lorentzian fitting was performed. Compared to clean images, our method achieved mean correlation coefficient of 0.90.

Impact: Our method can considerably increase SNR of CEST images without sacrificing image fidelity. After denoising, the derived CEST maps could more reliably represent molecular changes in brain regions.

Original language	English
DOIs	https://doi.org/10.58530/2025/2296
Publication status	Published - May 2025
Event	2025 ISMRM & ISMRT Annual Meeting & Exhibition - Honolulu, United States Duration: 10 May 2025 → 15 May 2025 https://www.ismrm.org/25m/

Conference

Conference	2025 ISMRM & ISMRT Annual Meeting & Exhibition
Abbreviated title	ISMRM 2025
Place	United States
City	Honolulu
Period	10/05/25 → 15/05/25
Internet address	https://www.ismrm.org/25m/

Funding

Authors would like to acknowledge the funding supports from Research Grants Council (11102218, 11200422, RFS2223-1S02, C1134-20G), City University of Hong Kong (7005626, 7030012, 9609321 and 9610616), Tung Biomedical Sciences Centre, State Key Laboratory of Terahertz and Millimeter Waves, Innovation and Technology Commission - InnoHK - Hong Kong Centre for Cerebro-Cardiovascular Health Engineering; MHKJS (MHP/076/23).

Research Keywords

CEST / APT / NOE
Denoising

RGC Funding Information

RGC-funded

Access Output

10.58530/2025/2296

Other Access Options

Check CityUHK Library

Permanent Link

Right click to copy link

3 Finished
4 Active

ITF: Artificial Intelligence-empowered Molecular Magnetic Resonance Imaging for the Brain
CHAN, W. Y. K. (Principal Investigator / Project Coordinator), HUANG, J. (Co-Investigator) & LAU, K. K. G. (Co-Investigator)
1/03/25 → …
Project: Research
ASDC_Sub: Futian Research Project: CEST MRI for Assessing Molecular Abnormalities in Aging Brain
CHAN, W. Y. K. (Principal Investigator / Project Coordinator), LEE, Y. (Co-Investigator), YIN, H. (Co-Investigator) & ZHAO, X. (Co-Investigator)
1/04/24 → …
Project: Research
GRF: Imaging Nose-to-brain Drug Delivery to Brain Tumors Using Multiple CEST MRI Contrast
CHAN, W. Y. K. (Principal Investigator / Project Coordinator) & LEUNG, G. K. K. (Co-Investigator)
1/01/23 → …
Project: Research

2025 ISMRM & ISMRT Annual Meeting & Exhibition
LIU, H. (Advisory Chair)
10 May 2025 → 15 May 2025
Activity: Organizing or Participating in a conference / an event › Conference / Symposium

Cite this

@conference{77b1e750507747ed8ad025f6d1137bc2,

title = "Enhancing SNR of CEST MRI Through Noise-to-Noise Deep Learning With K-Space Data Consistency",

abstract = "Motivation: Endogenous CEST contrast is relatively small and vulnerable to imaging noise. Goal(s): To develop a retrospective denoising method to enhance SNR of acquired CEST images. Approach: A denoising neural network was trained using pairs of noisy CEST images through a noise-to-noise deep learning approach, distinct from conventional approaches that use clean or simulated CEST images. A data consistency layer was introduced to preserve center k-space of original CEST images to improve fidelity. A transformer module was used to exploit spatiotemporal correlations among different frequency offsets. Results: Multipool Lorentzian fitting was performed. Compared to clean images, our method achieved mean correlation coefficient of 0.90. Impact: Our method can considerably increase SNR of CEST images without sacrificing image fidelity. After denoising, the derived CEST maps could more reliably represent molecular changes in brain regions.",

keywords = "CEST / APT / NOE, Denoising",

author = "Huabing LIU and Ziyi XIA and LAW, \{Lok Hin\} and Zilin CHEN and Jianpan HUANG and Dinggang SHEN and CHAN, \{Kannie W.Y.\}",

year = "2025",

month = may,

doi = "10.58530/2025/2296",

language = "English",

note = "2025 ISMRM \& ISMRT Annual Meeting \& Exhibition, ISMRM 2025 ; Conference date: 10-05-2025 Through 15-05-2025",

url = "https://www.ismrm.org/25m/",

}

Enhancing SNR of CEST MRI Through Noise-to-Noise Deep Learning With K-Space Data Consistency. / LIU, Huabing (Co-first Author); XIA, Ziyi (Co-first Author); LAW, Lok Hin et al.
2025. Paper presented at 2025 ISMRM & ISMRT Annual Meeting & Exhibition, Honolulu, Hawaii, United States.

Research output: Conference Papers › RGC 32 - Refereed conference paper (without host publication) › peer-review

TY - CONF

T1 - Enhancing SNR of CEST MRI Through Noise-to-Noise Deep Learning With K-Space Data Consistency

AU - LIU, Huabing

AU - XIA, Ziyi

AU - LAW, Lok Hin

AU - CHEN, Zilin

AU - HUANG, Jianpan

AU - SHEN, Dinggang

AU - CHAN, Kannie W.Y.

PY - 2025/5

Y1 - 2025/5

N2 - Motivation: Endogenous CEST contrast is relatively small and vulnerable to imaging noise. Goal(s): To develop a retrospective denoising method to enhance SNR of acquired CEST images. Approach: A denoising neural network was trained using pairs of noisy CEST images through a noise-to-noise deep learning approach, distinct from conventional approaches that use clean or simulated CEST images. A data consistency layer was introduced to preserve center k-space of original CEST images to improve fidelity. A transformer module was used to exploit spatiotemporal correlations among different frequency offsets. Results: Multipool Lorentzian fitting was performed. Compared to clean images, our method achieved mean correlation coefficient of 0.90. Impact: Our method can considerably increase SNR of CEST images without sacrificing image fidelity. After denoising, the derived CEST maps could more reliably represent molecular changes in brain regions.

AB - Motivation: Endogenous CEST contrast is relatively small and vulnerable to imaging noise. Goal(s): To develop a retrospective denoising method to enhance SNR of acquired CEST images. Approach: A denoising neural network was trained using pairs of noisy CEST images through a noise-to-noise deep learning approach, distinct from conventional approaches that use clean or simulated CEST images. A data consistency layer was introduced to preserve center k-space of original CEST images to improve fidelity. A transformer module was used to exploit spatiotemporal correlations among different frequency offsets. Results: Multipool Lorentzian fitting was performed. Compared to clean images, our method achieved mean correlation coefficient of 0.90. Impact: Our method can considerably increase SNR of CEST images without sacrificing image fidelity. After denoising, the derived CEST maps could more reliably represent molecular changes in brain regions.

KW - CEST / APT / NOE

KW - Denoising

U2 - 10.58530/2025/2296

DO - 10.58530/2025/2296

M3 - RGC 32 - Refereed conference paper (without host publication)

T2 - 2025 ISMRM & ISMRT Annual Meeting & Exhibition

Y2 - 10 May 2025 through 15 May 2025

ER -

Enhancing SNR of CEST MRI Through Noise-to-Noise Deep Learning With K-Space Data Consistency

Abstract

Conference

Funding

Research Keywords

RGC Funding Information

Access Output

Other Access Options

Permanent Link

Fingerprint

Projects

ITF: Artificial Intelligence-empowered Molecular Magnetic Resonance Imaging for the Brain

ASDC_Sub: Futian Research Project: CEST MRI for Assessing Molecular Abnormalities in Aging Brain

GRF: Imaging Nose-to-brain Drug Delivery to Brain Tumors Using Multiple CEST MRI Contrast

Activities

2025 ISMRM & ISMRT Annual Meeting & Exhibition

Cite this