Volume Reconstruction from Single MRI Thick-slice Stack with Deep Learning for Fetal Brain
Poster No:
1031
Submission Type:
Abstract Submission
Authors:
Hongjia Yang1, Mingxuan Liu1, Yi Liao2, Juncheng Zhu2, Haoxiang Li1, Zihan Li1, Jize Zhang3, Jialan Zheng1, Ziyu Li4, Haibo Qu2, Qiyuan Tian1
Institutions:
1School of Biomedical Engineering, Tsinghua University, Beijing, China, 2Department of Radiology, West China Second University Hospital, Sichuan University, Chengdu, China, 3Wellcome Centre for integrative Neuroimaging, OHBA, Department of Psychiatry, University of Oxford, Oxford, United Kingdom, 4Wellcome Centre for Integrative Neuroimaging, FMRIB, Nuffield Department of Clinical Neurosciences,, Oxford, United Kingdom
First Author:
Co-Author(s):
Yi Liao
Department of Radiology, West China Second University Hospital, Sichuan University
Chengdu, China
Department of Radiology, West China Second University Hospital, Sichuan University
Chengdu, China
Juncheng Zhu
Department of Radiology, West China Second University Hospital, Sichuan University
Chengdu, China
Department of Radiology, West China Second University Hospital, Sichuan University
Chengdu, China
Jize Zhang
Wellcome Centre for integrative Neuroimaging, OHBA, Department of Psychiatry, University of Oxford
Oxford, United Kingdom
Wellcome Centre for integrative Neuroimaging, OHBA, Department of Psychiatry, University of Oxford
Oxford, United Kingdom
Ziyu Li
Wellcome Centre for Integrative Neuroimaging, FMRIB, Nuffield Department of Clinical Neurosciences,
Oxford, United Kingdom
Wellcome Centre for Integrative Neuroimaging, FMRIB, Nuffield Department of Clinical Neurosciences,
Oxford, United Kingdom
Haibo Qu
Department of Radiology, West China Second University Hospital, Sichuan University
Chengdu, China
Department of Radiology, West China Second University Hospital, Sichuan University
Chengdu, China
Introduction:
High-isotropic-resolution volume reconstruction is essential yet challenging in fetal brain MRI studies. Current methods, such as NiftyMIC (Ebner, 2020), and their deep learning-based improvements (Huang, 2023; Xu, 2023; Cordero-Grande, 2022), require thick-slice stacks acquired in at least 3 different orientations and slice-to-volume co-registration, resulting in prolonged scan and reconstruction. Additionally, if any thick-slice stack is affected by motion artifact that are common in fetal imaging, the reconstruction is prone to failure.
To tackle this challenge, we propose FetalSR, a pipeline for fast high-isotropic-resolution fetal brain volume reconstruction using only one thick-slice stack acquired in any axial/sagittal/coronal orientation. FetalSR consists of an efficient slice-to-template registration workflow, followed by a reconstruction neural network to super-resolve high-resolution images, demonstrating more stable reconstruction and providing accuracy in downstream brain segmentation tasks.
To tackle this challenge, we propose FetalSR, a pipeline for fast high-isotropic-resolution fetal brain volume reconstruction using only one thick-slice stack acquired in any axial/sagittal/coronal orientation. FetalSR consists of an efficient slice-to-template registration workflow, followed by a reconstruction neural network to super-resolve high-resolution images, demonstrating more stable reconstruction and providing accuracy in downstream brain segmentation tasks.
Methods:
Data Acquisition: With informed written consent and IRB approval, data were acquired on 2004 pregnant women (20-40 weeks gestation) with normal fetuses using a T2-weighted TSE sequence to obtain thick-slice stacks from different directions, with 4 mm slice thickness and 0.94 × 0.94 mm2 in-plane resolution.
Reconstruction Reference: High-quality thick-slice stacks were selected using ORN-IQA (Liu, 2023). NiftyMIC was then applied to reconstruct image volumes at 0.8mm isotropic resolution in the template space as reference. 1502 cases were successfully reconstructed. 456 cases (1333 stacks) were used for training and validation while the remaining 1046 cases (3103 stacks) were used for evaluation.
FetalSR Pipeline (Fig. 1A,B): The pipeline begins with slice-to-template registration, up-sampling and transforming each thick-slice stack to 0.8mm isotropic resolution in template space. This process involves brain masking using a public fetal brain mask network (Ebner, 2020), followed by determining head position (i.e., 3 axes) with proposed orientation recognition network. High-resolution volumes are then computed by the reconstruction network and then fed into a public segmentation network (Fidon, 2024) for brain segmentation into eight labels.
Network and Training (Fig. 1C,D): The orientation recognition network followed the SFCN (Peng, 2021). The reconstruction network was a 3D U-Net with residual learning. Training was performed using the Adam optimizer with a combined L1 loss and perceptual loss function.
Evaluation: PSNR and SSIM were used to measure the similarity between super-resolution results versus references. Dice similarity and Pearson correlation quantified the overlap of brain segments. The relationship between brain volumes and gestational age was fitted using a generalized additive model with six knots penalized B-spline functions.
Reconstruction Reference: High-quality thick-slice stacks were selected using ORN-IQA (Liu, 2023). NiftyMIC was then applied to reconstruct image volumes at 0.8mm isotropic resolution in the template space as reference. 1502 cases were successfully reconstructed. 456 cases (1333 stacks) were used for training and validation while the remaining 1046 cases (3103 stacks) were used for evaluation.
FetalSR Pipeline (Fig. 1A,B): The pipeline begins with slice-to-template registration, up-sampling and transforming each thick-slice stack to 0.8mm isotropic resolution in template space. This process involves brain masking using a public fetal brain mask network (Ebner, 2020), followed by determining head position (i.e., 3 axes) with proposed orientation recognition network. High-resolution volumes are then computed by the reconstruction network and then fed into a public segmentation network (Fidon, 2024) for brain segmentation into eight labels.
Network and Training (Fig. 1C,D): The orientation recognition network followed the SFCN (Peng, 2021). The reconstruction network was a 3D U-Net with residual learning. Training was performed using the Adam optimizer with a combined L1 loss and perceptual loss function.
Evaluation: PSNR and SSIM were used to measure the similarity between super-resolution results versus references. Dice similarity and Pearson correlation quantified the overlap of brain segments. The relationship between brain volumes and gestational age was fitted using a generalized additive model with six knots penalized B-spline functions.
·Figure 1. FetalSR pipeline.
Results:
FetalSR successfully reconstructed high-quality, high-isotropic-resolution T2-weighted fetal brain volumes, even with missing brain regions and motion artifacts, achieving results similar to reference. The mean PSNR and SSIM between super-resolution results and references were 32.0 dB and 0.96, respectively. The reconstruction was about 1 minute.
Super-resolved volumes provided accurate brain segmentation (Fig.2B) and faithfully captured the brain growth pattern across advancing gestational age, with developmental trajectories highly similar to reference high-resolution volumes (Fig.2A).
Fig.2C shows cases where NiftyMIC failed due to motion artifacts, while FetalSR still provided high-quality results.
Super-resolved volumes provided accurate brain segmentation (Fig.2B) and faithfully captured the brain growth pattern across advancing gestational age, with developmental trajectories highly similar to reference high-resolution volumes (Fig.2A).
Fig.2C shows cases where NiftyMIC failed due to motion artifacts, while FetalSR still provided high-quality results.
·Figure 2. Reconstruction and segmentation results.
Conclusions:
FetalSR proves that high-resolution fetal brain MRI can be reconstructed from a single thick-slice stack in any orientation, providing robust images and segmentation for diagnostic and quantitative analysis.
Lifespan Development:
Normal Brain Development: Fetus to Adolescence 1
Modeling and Analysis Methods:
Image Registration and Computational Anatomy
Methods Development 2
Motion Correction and Preprocessing
Novel Imaging Acquisition Methods:
Anatomical MRI
Keywords:
Development
Machine Learning
MRI
PEDIATRIC
STRUCTURAL MRI
Workflows
Other - Fetal
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?
Provide references using APA citation style.
Huang, S. (2023). Super-resolution reconstruction of fetal brain MRI with prior anatomical knowledge. In International Conference on Information Processing in Medical Imaging (pp. 428-441). Cham: Springer Nature Switzerland.
Xu, J. (2023). NeSVoR: implicit neural representation for slice-to-volume reconstruction in MRI. IEEE transactions on medical imaging, 42(6), 1707-1719.
Cordero-Grande, L. (2022). Fetal MRI by robust deep generative prior reconstruction and diffeomorphic registration. IEEE Transactions on Medical Imaging, 42(3), 810-822.
Liu, M. (2023). Image Quality Assessment using an Orientation Recognition Network for Fetal MRI. In: 2024 ISMRM annual meeting (0726).
Fidon, L. (2024). A Dempster-Shafer approach to trustworthy AI with application to fetal brain MRI segmentation. IEEE transactions on pattern analysis and machine intelligence.
Peng, H. (2021). Accurate brain age prediction with lightweight deep neural networks. Medical image analysis, 68, 101871.