Deep Learning-Based Motion-Compensated Reconstruction for Accelerating 4-Dimensional Magnetic Resonance Fingerprinting

Lu WANG; Chenyang LIU; Yinghui WANG; Xiang WANG; Peilin WANG; Weihang LIAO; Xinzhi TENG; Andy Lai Yin CHEUNG; Victor Ho Fun LEE; Shaohua ZHI; Ge REN; Jing QIN; Peng CAO; Tian LI; Jing CAI

doi:10.1016/j.ijrobp.2025.10.001

Deep Learning-Based Motion-Compensated Reconstruction for Accelerating 4-Dimensional Magnetic Resonance Fingerprinting

Lu WANG
, Chenyang LIU
, Yinghui WANG
, Xiang WANG
, Peilin WANG
, Weihang LIAO
, Xinzhi TENG
, Andy Lai Yin CHEUNG
, Victor Ho Fun LEE
, Shaohua ZHI
, Ge REN
, Jing QIN
, Peng CAO
, Tian LI^*
, Jing CAI

^*Corresponding author for this work

School of Interdisciplinary Studies

Research output: Journal Publications › Journal Article (refereed) › peer-review

Abstract

Purpose: To develop and validate DeepMocor, a deep learning–based method for motion-compensated 4-dimensional magnetic resonance fingerprinting (4D-MRF) reconstruction to accelerate conventional 4D-MRF reconstruction, enabling more efficient clinical treatment planning.

Methods and Materials: This prospective study enrolled 19 hepatocellular carcinoma patients (mean age, 62 years; 14 males) between June 2021 and October 2024. Abdominal free-breathing raw k-space data were acquired using a 3T magnetic resonance imaging scanner. DeepMocor involves motion field initialization, motion field refinement, and final 4D-MRF reconstruction. A 3-fold cross-validation strategy was employed for training and testing. Performance was evaluated against 2 alternatives (stage-I&III-only; stage-III-only) in terms of image quality, tissue property accuracy, tumor-to-tissue contrast, and tumor motion measurement. Image quality was assessed by peak signal-to-noise ratio (PSNR) and structural similarity index (SSIM). Tissue property accuracy was evaluated by mean absolute percentage error (MAPE). Tumor-to-tissue contrast was quantified by contrast-to-noise ratio (CNR) of the tumor region and the surrounding area. Tumor motion tracking was assessed by average motion difference (AMD) and Pearson correlation coefficients (PCC) in the superior-inferior and anterior-posterior directions. The Wilcoxon signed rank test was used for comparison with P < .05.

Results: For T1 maps, DeepMocor demonstrates PSNR of 25.49 ± 1.30, SSIM of 0.84 ± 0.03, MAPE of 3.5% to 5.9%, and CNR of 6.14 ± 3.54. For T2 maps, DeepMocor achieves PSNR of 25.57 ± 1.24, SSIM of 0.88 ± 0.02, MAPE of 3.1% to 15.8%, and CNR of 8.42 ± 13.72. DeepMocor achieves AMD of 0.62 ± 0.86 mm with PCC of 0.96 ± 0.07 in the superior-inferior direction and AMD of 0.32 ± 0.37 mm with PCC of 0.94 ± 0.06 in the anterior-posterior direction. DeepMocor shows superior performance across most metrics compared to stage-III-only and a subset of metrics compared to stage-I&III-only significantly.

Conclusions: The proposed DeepMocor method enables a 24-fold acceleration compared to the conventional reference method, highlighting its potential for liver radiation therapy planning.

Original language	English
Pages (from-to)	1103-1114
Number of pages	12
Journal	International Journal of Radiation Oncology Biology Physics
Volume	124
Issue number	4
Early online date	30 Oct 2025
DOIs	https://doi.org/10.1016/j.ijrobp.2025.10.001
Publication status	Published - 15 Mar 2026

Bibliographical note

Publisher Copyright:
Crown Copyright © 2025 Published by Elsevier Inc. All rights are reserved, including those for text and data mining, AI training, and similar technologies.

Funding

Disclosures: none. This work was supported in part by the National Natural Science Foundation of China Young Scientist Fund (grant number 82202941); the General Research Fund (grant numbers 15102219, 15104822, 15104323); the Health and Medical Research Fund (HMRF 10211606); and the Health Bureau, Innovation and Technology Support Program (ITS/049/22FP).

UN SDGs

This output contributes to the following UN Sustainable Development Goals (SDGs)

SDG 3 Good Health and Well-being

Access to Document

10.1016/j.ijrobp.2025.10.001

Cite this

WANG, L., LIU, C., WANG, Y., WANG, X., WANG, P., LIAO, W., TENG, X., CHEUNG, A. L. Y., LEE, V. H. F., ZHI, S., REN, G., QIN, J., CAO, P., LI, T., & CAI, J. (2026). Deep Learning-Based Motion-Compensated Reconstruction for Accelerating 4-Dimensional Magnetic Resonance Fingerprinting. International Journal of Radiation Oncology Biology Physics, 124(4), 1103-1114. https://doi.org/10.1016/j.ijrobp.2025.10.001

@article{2945ea7ae9bc4550aa762cdfb7edbb04,

title = "Deep Learning-Based Motion-Compensated Reconstruction for Accelerating 4-Dimensional Magnetic Resonance Fingerprinting",

abstract = "Purpose: To develop and validate DeepMocor, a deep learning–based method for motion-compensated 4-dimensional magnetic resonance fingerprinting (4D-MRF) reconstruction to accelerate conventional 4D-MRF reconstruction, enabling more efficient clinical treatment planning.Methods and Materials: This prospective study enrolled 19 hepatocellular carcinoma patients (mean age, 62 years; 14 males) between June 2021 and October 2024. Abdominal free-breathing raw k-space data were acquired using a 3T magnetic resonance imaging scanner. DeepMocor involves motion field initialization, motion field refinement, and final 4D-MRF reconstruction. A 3-fold cross-validation strategy was employed for training and testing. Performance was evaluated against 2 alternatives (stage-I\&III-only; stage-III-only) in terms of image quality, tissue property accuracy, tumor-to-tissue contrast, and tumor motion measurement. Image quality was assessed by peak signal-to-noise ratio (PSNR) and structural similarity index (SSIM). Tissue property accuracy was evaluated by mean absolute percentage error (MAPE). Tumor-to-tissue contrast was quantified by contrast-to-noise ratio (CNR) of the tumor region and the surrounding area. Tumor motion tracking was assessed by average motion difference (AMD) and Pearson correlation coefficients (PCC) in the superior-inferior and anterior-posterior directions. The Wilcoxon signed rank test was used for comparison with P < .05.Results: For T1 maps, DeepMocor demonstrates PSNR of 25.49 ± 1.30, SSIM of 0.84 ± 0.03, MAPE of 3.5\% to 5.9\%, and CNR of 6.14 ± 3.54. For T2 maps, DeepMocor achieves PSNR of 25.57 ± 1.24, SSIM of 0.88 ± 0.02, MAPE of 3.1\% to 15.8\%, and CNR of 8.42 ± 13.72. DeepMocor achieves AMD of 0.62 ± 0.86 mm with PCC of 0.96 ± 0.07 in the superior-inferior direction and AMD of 0.32 ± 0.37 mm with PCC of 0.94 ± 0.06 in the anterior-posterior direction. DeepMocor shows superior performance across most metrics compared to stage-III-only and a subset of metrics compared to stage-I\&III-only significantly. Conclusions: The proposed DeepMocor method enables a 24-fold acceleration compared to the conventional reference method, highlighting its potential for liver radiation therapy planning.",

author = "Lu WANG and Chenyang LIU and Yinghui WANG and Xiang WANG and Peilin WANG and Weihang LIAO and Xinzhi TENG and CHEUNG, \{Andy Lai Yin\} and LEE, \{Victor Ho Fun\} and Shaohua ZHI and Ge REN and Jing QIN and Peng CAO and Tian LI and Jing CAI",

note = "Publisher Copyright: Crown Copyright {\textcopyright} 2025 Published by Elsevier Inc. All rights are reserved, including those for text and data mining, AI training, and similar technologies.",

year = "2026",

month = mar,

day = "15",

doi = "10.1016/j.ijrobp.2025.10.001",

language = "English",

volume = "124",

pages = "1103--1114",

journal = "International Journal of Radiation Oncology Biology Physics",

issn = "0360-3016",

publisher = "Elsevier Inc.",

number = "4",

}

WANG, L, LIU, C, WANG, Y, WANG, X, WANG, P, LIAO, W, TENG, X, CHEUNG, ALY, LEE, VHF, ZHI, S, REN, G, QIN, J, CAO, P, LI, T & CAI, J 2026, 'Deep Learning-Based Motion-Compensated Reconstruction for Accelerating 4-Dimensional Magnetic Resonance Fingerprinting', International Journal of Radiation Oncology Biology Physics, vol. 124, no. 4, pp. 1103-1114. https://doi.org/10.1016/j.ijrobp.2025.10.001

TY - JOUR

T1 - Deep Learning-Based Motion-Compensated Reconstruction for Accelerating 4-Dimensional Magnetic Resonance Fingerprinting

AU - WANG, Lu

AU - LIU, Chenyang

AU - WANG, Yinghui

AU - WANG, Xiang

AU - WANG, Peilin

AU - LIAO, Weihang

AU - TENG, Xinzhi

AU - CHEUNG, Andy Lai Yin

AU - LEE, Victor Ho Fun

AU - ZHI, Shaohua

AU - REN, Ge

AU - QIN, Jing

AU - CAO, Peng

AU - LI, Tian

AU - CAI, Jing

PY - 2026/3/15

Y1 - 2026/3/15

N2 - Purpose: To develop and validate DeepMocor, a deep learning–based method for motion-compensated 4-dimensional magnetic resonance fingerprinting (4D-MRF) reconstruction to accelerate conventional 4D-MRF reconstruction, enabling more efficient clinical treatment planning.Methods and Materials: This prospective study enrolled 19 hepatocellular carcinoma patients (mean age, 62 years; 14 males) between June 2021 and October 2024. Abdominal free-breathing raw k-space data were acquired using a 3T magnetic resonance imaging scanner. DeepMocor involves motion field initialization, motion field refinement, and final 4D-MRF reconstruction. A 3-fold cross-validation strategy was employed for training and testing. Performance was evaluated against 2 alternatives (stage-I&III-only; stage-III-only) in terms of image quality, tissue property accuracy, tumor-to-tissue contrast, and tumor motion measurement. Image quality was assessed by peak signal-to-noise ratio (PSNR) and structural similarity index (SSIM). Tissue property accuracy was evaluated by mean absolute percentage error (MAPE). Tumor-to-tissue contrast was quantified by contrast-to-noise ratio (CNR) of the tumor region and the surrounding area. Tumor motion tracking was assessed by average motion difference (AMD) and Pearson correlation coefficients (PCC) in the superior-inferior and anterior-posterior directions. The Wilcoxon signed rank test was used for comparison with P < .05.Results: For T1 maps, DeepMocor demonstrates PSNR of 25.49 ± 1.30, SSIM of 0.84 ± 0.03, MAPE of 3.5% to 5.9%, and CNR of 6.14 ± 3.54. For T2 maps, DeepMocor achieves PSNR of 25.57 ± 1.24, SSIM of 0.88 ± 0.02, MAPE of 3.1% to 15.8%, and CNR of 8.42 ± 13.72. DeepMocor achieves AMD of 0.62 ± 0.86 mm with PCC of 0.96 ± 0.07 in the superior-inferior direction and AMD of 0.32 ± 0.37 mm with PCC of 0.94 ± 0.06 in the anterior-posterior direction. DeepMocor shows superior performance across most metrics compared to stage-III-only and a subset of metrics compared to stage-I&III-only significantly. Conclusions: The proposed DeepMocor method enables a 24-fold acceleration compared to the conventional reference method, highlighting its potential for liver radiation therapy planning.

AB - Purpose: To develop and validate DeepMocor, a deep learning–based method for motion-compensated 4-dimensional magnetic resonance fingerprinting (4D-MRF) reconstruction to accelerate conventional 4D-MRF reconstruction, enabling more efficient clinical treatment planning.Methods and Materials: This prospective study enrolled 19 hepatocellular carcinoma patients (mean age, 62 years; 14 males) between June 2021 and October 2024. Abdominal free-breathing raw k-space data were acquired using a 3T magnetic resonance imaging scanner. DeepMocor involves motion field initialization, motion field refinement, and final 4D-MRF reconstruction. A 3-fold cross-validation strategy was employed for training and testing. Performance was evaluated against 2 alternatives (stage-I&III-only; stage-III-only) in terms of image quality, tissue property accuracy, tumor-to-tissue contrast, and tumor motion measurement. Image quality was assessed by peak signal-to-noise ratio (PSNR) and structural similarity index (SSIM). Tissue property accuracy was evaluated by mean absolute percentage error (MAPE). Tumor-to-tissue contrast was quantified by contrast-to-noise ratio (CNR) of the tumor region and the surrounding area. Tumor motion tracking was assessed by average motion difference (AMD) and Pearson correlation coefficients (PCC) in the superior-inferior and anterior-posterior directions. The Wilcoxon signed rank test was used for comparison with P < .05.Results: For T1 maps, DeepMocor demonstrates PSNR of 25.49 ± 1.30, SSIM of 0.84 ± 0.03, MAPE of 3.5% to 5.9%, and CNR of 6.14 ± 3.54. For T2 maps, DeepMocor achieves PSNR of 25.57 ± 1.24, SSIM of 0.88 ± 0.02, MAPE of 3.1% to 15.8%, and CNR of 8.42 ± 13.72. DeepMocor achieves AMD of 0.62 ± 0.86 mm with PCC of 0.96 ± 0.07 in the superior-inferior direction and AMD of 0.32 ± 0.37 mm with PCC of 0.94 ± 0.06 in the anterior-posterior direction. DeepMocor shows superior performance across most metrics compared to stage-III-only and a subset of metrics compared to stage-I&III-only significantly. Conclusions: The proposed DeepMocor method enables a 24-fold acceleration compared to the conventional reference method, highlighting its potential for liver radiation therapy planning.

UR - https://www.scopus.com/pages/publications/105023386656

U2 - 10.1016/j.ijrobp.2025.10.001

DO - 10.1016/j.ijrobp.2025.10.001

M3 - Journal Article (refereed)

C2 - 41138788

AN - SCOPUS:105023386656

SN - 0360-3016

VL - 124

SP - 1103

EP - 1114

JO - International Journal of Radiation Oncology Biology Physics

JF - International Journal of Radiation Oncology Biology Physics

IS - 4

ER -

Deep Learning-Based Motion-Compensated Reconstruction for Accelerating 4-Dimensional Magnetic Resonance Fingerprinting

Abstract

Bibliographical note

Funding

UN SDGs

Access to Document

Other files and links

Fingerprint

Cite this