Development of χ-separation pipeline for UK Biobank dataset
Poster No:
1511
Submission Type:
Abstract Submission
Authors:
Jaehyeon Koo1, Hwihun Jeong1, Jiye Kim1, Rokgi Hong1, Hyeong-Geol Shin2,3, Xu Li3,4, Yun Soo Hong5, Ye Qiao2, Jongho Lee1
Institutions:
1Department of Electrical and Computer Engineering, Seoul National University, Seoul, Korea, Republic of, 2Department of Biomedical Engineering, Johns Hopkins University, Baltimore, MD, 3F.M. Kirby Research Center for Functional Brain Imaging, Kennedy Krieger Institute, Baltimore, MD, 4Russell H. Morgan Department of Radiology and Radiological Science, Johns Hopkins University School of Medicine, Baltimore, MD, 5McKusick-Nathans Institute, Department of Genetic Medicine, Johns Hopkins University School of Medic, Baltimore, MD
First Author:
Jaehyeon Koo
Department of Electrical and Computer Engineering, Seoul National University
Seoul, Korea, Republic of
Department of Electrical and Computer Engineering, Seoul National University
Seoul, Korea, Republic of
Co-Author(s):
Hwihun Jeong
Department of Electrical and Computer Engineering, Seoul National University
Seoul, Korea, Republic of
Department of Electrical and Computer Engineering, Seoul National University
Seoul, Korea, Republic of
Jiye Kim
Department of Electrical and Computer Engineering, Seoul National University
Seoul, Korea, Republic of
Department of Electrical and Computer Engineering, Seoul National University
Seoul, Korea, Republic of
Rokgi Hong
Department of Electrical and Computer Engineering, Seoul National University
Seoul, Korea, Republic of
Department of Electrical and Computer Engineering, Seoul National University
Seoul, Korea, Republic of
Hyeong-Geol Shin
Department of Biomedical Engineering, Johns Hopkins University|F.M. Kirby Research Center for Functional Brain Imaging, Kennedy Krieger Institute
Baltimore, MD|Baltimore, MD
Department of Biomedical Engineering, Johns Hopkins University|F.M. Kirby Research Center for Functional Brain Imaging, Kennedy Krieger Institute
Baltimore, MD|Baltimore, MD
Xu Li
F.M. Kirby Research Center for Functional Brain Imaging, Kennedy Krieger Institute|Russell H. Morgan Department of Radiology and Radiological Science, Johns Hopkins University School of Medicine
Baltimore, MD|Baltimore, MD
F.M. Kirby Research Center for Functional Brain Imaging, Kennedy Krieger Institute|Russell H. Morgan Department of Radiology and Radiological Science, Johns Hopkins University School of Medicine
Baltimore, MD|Baltimore, MD
Yun Soo Hong
McKusick-Nathans Institute, Department of Genetic Medicine, Johns Hopkins University School of Medic
Baltimore, MD
McKusick-Nathans Institute, Department of Genetic Medicine, Johns Hopkins University School of Medic
Baltimore, MD
Jongho Lee
Department of Electrical and Computer Engineering, Seoul National University
Seoul, Korea, Republic of
Department of Electrical and Computer Engineering, Seoul National University
Seoul, Korea, Republic of
Introduction:
χ-separation(Shin et al., 2021) is an advanced quantitative susceptibility mapping (QSM) method that separates paramagnetic and diamagnetic susceptibility sources. UK Biobank (UKB) is a large dataset with a variety of health information(Miller et al., 2016), employing χ-separation for UKB may enable us to investigate the effects of susceptibility sources in large cohorts. However, the absence of spin-echo data hindered the application of χ-separation. We employed recently developed χ-sepnet-R2* and χ-separation toolbox(Kim et al., 2024; SNU-LIST, 2024), which produces COSMOS-quality χ-separation results using only GRE data, to apply χ-separation in UKB. For high-quality χ-separation, artifacts due to low-resolution R2* and few-echo acquisition should be corrected(Han et al., 2015), promoting us to develop an R2* improvement network. In this study, we aim to develop a pipeline to perform susceptibility source separation on the UKB dataset using χ-sepnet-R2* (Fig. 1). Additionally, we improved the quality of R2* using a deep neural network.
Methods:
Seventeen subjects (train:valid:test = 10:3:4) were scanned using the same GRE3mm protocol with the UKB protocol (TR/TE1/TE2 = 27/9.42/19.7ms, resolution = 0.8×0.8×3mm3, FOV = 256×288×48mm3). Since the real UKB data lacks a reference for validation, we acquired additional GRE1mm (TR/TE1/TE2/TE3/TE4 = 27/4.98/9.42/14.56/19.7ms, resolution = 0.8×0.8×1mm3, FOV = 256×288×48mm3) to create a reference by following the UKB protocol.
The R2* network takes the input of low-quality R2* and GRE magnitude images to produce high-quality R2*. Input magnitude images were normalized by the mean of zero-echo magnitude map.
The label was created by resampling the R2* obtained from ARLO(Pei et al., 2015) of GRE1mm data to a resolution of 0.8×0.8×3mm3. 3D U-Net was utilized with loss function includes L1 loss, gradient loss, and reconstruction loss to enable the model to learn the relationship between the R2* and the magnitude image.
We tested for two R2* options: one for R2* fitted from GRE3mm (R2*orig) and the other for R2* enhanced by R2* network (R2*net). A local field is derived from the GRE phase image(Eckstein et al., 2018; Abdul-Rahamn et al., 2007; Dymerska et al., 2021; Özbay et al., 2017), and both R2* and the local field are resampled into 1×1×1mm3 to prevent the loss of high-frequency information when performing B0 alignment under conditions of anisotropy. Then, the χ-separation maps are reconstructed using χ-sepnet-R2* from R2* and the local field (Fig. 1).
To validate the quality of R2* and χ-separation maps, we used the χ-separation maps obtained from χ-sepnet-R2* using GRE1mm as a reference. NRMSE, PSNR, and SSIM were calculated with respect to the reference data. To evaluate the applicability to the UKB dataset, the proposed method was also applied to real UKB data, including multiple sclerosis (MS) patients.
The R2* network takes the input of low-quality R2* and GRE magnitude images to produce high-quality R2*. Input magnitude images were normalized by the mean of zero-echo magnitude map.
The label was created by resampling the R2* obtained from ARLO(Pei et al., 2015) of GRE1mm data to a resolution of 0.8×0.8×3mm3. 3D U-Net was utilized with loss function includes L1 loss, gradient loss, and reconstruction loss to enable the model to learn the relationship between the R2* and the magnitude image.
We tested for two R2* options: one for R2* fitted from GRE3mm (R2*orig) and the other for R2* enhanced by R2* network (R2*net). A local field is derived from the GRE phase image(Eckstein et al., 2018; Abdul-Rahamn et al., 2007; Dymerska et al., 2021; Özbay et al., 2017), and both R2* and the local field are resampled into 1×1×1mm3 to prevent the loss of high-frequency information when performing B0 alignment under conditions of anisotropy. Then, the χ-separation maps are reconstructed using χ-sepnet-R2* from R2* and the local field (Fig. 1).
To validate the quality of R2* and χ-separation maps, we used the χ-separation maps obtained from χ-sepnet-R2* using GRE1mm as a reference. NRMSE, PSNR, and SSIM were calculated with respect to the reference data. To evaluate the applicability to the UKB dataset, the proposed method was also applied to real UKB data, including multiple sclerosis (MS) patients.
Results:
Fig. 2a shows that R2* reconstructed with the R2* network demonstrated a less noisy image. Quantitative analysis also confirmed that R2*net outperformed R2*orig when it comes to correspondence with reference R2* (NRMSE: 22.3 % for R2*net vs. 55.5 % for R2*orig). χ-separation maps also showed the best performance with R2*net (NRMSE: 39.3, 38.9 % for R2*net vs. 48.5, 47.7 % for R2*orig). In the application of real UKB data, (Fig. 2c) χ separation maps reconstructed with R2*net exhibited better contrast in the deep gray matter regions compared to those reconstructed with R2*orig. For MS patient data, results are shown in Fig. 2d.
Conclusions:
In this study, we developed a pipeline to apply χ-sepnet to the UKB dataset and successfully achieved susceptibility source separation. Despite the anisotropic resolution, thick thickness, and 2-echo acquisition, we obtained high-quality χ-separation maps by applying χ-sepnet-R2*. This demonstrates the feasibility of routinely applying the pipeline to a large dataset.
This work was supported by IITP-2023-RS-2023-00256081, grant funded by the Korea government(MSIT)
This work was supported by IITP-2023-RS-2023-00256081, grant funded by the Korea government(MSIT)
Modeling and Analysis Methods:
Methods Development 1
Other Methods 2
Keywords:
Machine Learning
MRI
MRI PHYSICS
Myelin
Open Data
White Matter
Other - Chi-separation, Quantitative Susceptibility Mapping (QSM)
1|2Indicates the priority used for review
Abstract Information
By submitting your proposal, you grant permission for the Organization for Human Brain Mapping (OHBM) to distribute your work in any format, including video, audio print and electronic text through OHBM OnDemand, social media channels, the OHBM website, or other electronic publications and media.
The Open Science Special Interest Group (OSSIG) is introducing a reproducibility challenge for OHBM 2025. This new initiative aims to enhance the reproducibility of scientific results and foster collaborations between labs. Teams will consist of a “source” party and a “reproducing” party, and will be evaluated on the success of their replication, the openness of the source work, and additional deliverables. Click here for more information. Propose your OHBM abstract(s) as source work for future OHBM meetings by selecting one of the following options:
Please indicate below if your study was a "resting state" or "task-activation” study.
Healthy subjects only or patients (note that patient studies may also involve healthy subjects):
Was this research conducted in the United States?
Were any human subjects research approved by the relevant Institutional Review Board or ethics panel? NOTE: Any human subjects studies without IRB approval will be automatically rejected.
Were any animal research approved by the relevant IACUC or other animal research panel? NOTE: Any animal studies without IACUC approval will be automatically rejected.
Please indicate which methods were used in your research:
For human MRI, what field strength scanner do you use?
Which processing packages did you use for your study?
Provide references using APA citation style.
Dymerska, B., Eckstein, K., Bachrata, B., Siow, B., Trattnig, S., Shmueli, K., & Robinson, S. D. (2021). Phase unwrapping with a rapid opensource minimum spanning tree algorithm (ROMEO). Magnetic resonance in medicine, 85(4), 2294-2308.
Eckstein, K., Dymerska, B., Bachrata, B., Bogner, W., Poljanc, K., Trattnig, S., & Robinson, S. D. (2018). Computationally efficient combination of multi‐channel phase data from multi‐echo acquisitions (ASPIRE). Magnetic resonance in medicine, 79(6), 2996-3006.
Han, D., Nam, Y., Gho, S. M., & Kim, D. H. (2015). Volumetric R2* mapping using z‐shim multi‐echo gradient echo imaging. Magnetic Resonance in Medicine, 73(3), 1164-1170.
Kim, M., Ji, S., Kim, J., Min, K., Jeong, H., Youn, J., ... & Lee, J. (2024). \chi-sepnet: Deep neural network for magnetic susceptibility source separation. arXiv preprint arXiv:2409.14030.
Miller, K. L., Alfaro-Almagro, F., Bangerter, N. K., Thomas, D. L., Yacoub, E., Xu, J., ... & Smith, S. M. (2016). Multimodal population brain imaging in the UK Biobank prospective epidemiological study. Nature neuroscience, 19(11), 1523-1536.
Özbay, P. S., Deistung, A., Feng, X., Nanz, D., Reichenbach, J. R., & Schweser, F. (2017). A comprehensive numerical analysis of background phase correction with V‐SHARP. NMR in Biomedicine, 30(4), e3550.
Pei, M., Nguyen, T. D., Thimmappa, N. D., Salustri, C., Dong, F., Cooper, M. A., ... & Wang, Y. (2015). Algorithm for fast monoexponential fitting based on auto‐regression on linear operations (ARLO) of data. Magnetic resonance in medicine, 73(2), 843-850.
Shin, H. G., Lee, J., Yun, Y. H., Yoo, S. H., Jang, J., Oh, S. H., ... & Lee, J. (2021). χ-separation: Magnetic susceptibility source separation toward iron and myelin mapping in the brain. Neuroimage, 240, 118371.
SNU-LIST. (2024). chi-separation toolbox (χ-separation, x-separation) Version 1.1.3. GitHub. Retrieved October 14, 2024, from https://github.com/SNU-LIST/chi-separation