A Deep Learning Framework for Diffusion MRI-Based Cortical Surface Reconstruction
Poster No:
1284
Submission Type:
Abstract Submission
Authors:
Chengjin Li1, Yuqian Chen2, Nir Sochen3, Lauren O’Donnell2, Ofer Pasternak2, Fan Zhang1
Institutions:
1University of Electronic Science and Technology of China, Chengdu, Sichuan, 2Brigham and Women's Hospital, Harvard Medical School, Boston, MA, 3School of Mathematical Sciences, University of Tel Aviv, Tel Aviv
First Author:
Co-Author(s):
Introduction:
Diffusion MRI (dMRI) is an imaging technique to study the brain's white matter (WM) structural connectivity (Zhang et al., 2022). Reconstructing cortical surfaces is crucial for dMRI analyses such as WM tractography (Shastin et al., 2022) and multimodal MRI analysis (Yeo et al., 2011). Currently, cortical surfaces in dMRI data are obtained by reconstructing them from T1-weighted images and then registering them to the dMRI space. However, intermodality registration is challenging due to image distortions and the low dMRI resolution. Additionally, these methods are not applicable when T1-weighted data is unavailable. This study introduces DDSurfer, a novel deep-learning framework for reconstructing cortical surfaces (i.e., the WM and pial surfaces) directly from dMRI data. We show that by bypassing inter-modality MRI registration, DDSurfer largely enhances the accuracy and efficiency of cortical surface reconstruction in dMRI.
Methods:
DDSurfer includes 2 major steps (see overview in Fig 1): 1) initial surface reconstruction via tissue segmentation, and 2) surface refinement via weakly supervised learning.
Step 1: DDSurfer begins with computing dMRI-derived maps (FA, MD, 3 eigenvalues), from which probabilistic segmentation maps of WM, gray matter (GM), and cerebrospinal fluid (CSF) are computed using the recently proposed DDParcel method (Zhang et al., 2024). To improve the delineation of tissue boundaries, the probabilistic maps are further processed using distance field enhancement (DFE) (Han et al., 2004) and fuzzy C-Means with spatial constraints (FCM_S) (Wen et al., 2013). Then, from the processed probabilistic maps, cortical reconstruction using implicit surface evolution (CRUISE) (Han et al., 2004) is applied to generate signed distance fields (SDFs) for the WM and pial surfaces, followed by a fast topology correction algorithm (FastTCA) (Ma et al., 2022) to maintain genus-0 topology (Fig 1b). Finally, cortical surfaces are extracted from the SDFs using marching cubes (Lorensen & Cline, 1987)(Fig 1c).
Step 2: Although the above processing effectively captures the overall surface shape, it struggles to resolve the intricate, highly folded patterns of gyri and sulci. Therefore, we perform a surface refinement via weakly supervised learning. To do so, we extend the CoSeg method (Ma et al., 2024), which performs T1-weighted surface reconstruction by learning diffeomorphic deformations to fit an initial surface to target cortical surfaces. We modify CoSeg by integrating the dMRI-derived maps with the above-obtained cortical surfaces to train a dMRI-specific model for surface reconstruction and cortical parcellation (Fig 1d and e).
Step 1: DDSurfer begins with computing dMRI-derived maps (FA, MD, 3 eigenvalues), from which probabilistic segmentation maps of WM, gray matter (GM), and cerebrospinal fluid (CSF) are computed using the recently proposed DDParcel method (Zhang et al., 2024). To improve the delineation of tissue boundaries, the probabilistic maps are further processed using distance field enhancement (DFE) (Han et al., 2004) and fuzzy C-Means with spatial constraints (FCM_S) (Wen et al., 2013). Then, from the processed probabilistic maps, cortical reconstruction using implicit surface evolution (CRUISE) (Han et al., 2004) is applied to generate signed distance fields (SDFs) for the WM and pial surfaces, followed by a fast topology correction algorithm (FastTCA) (Ma et al., 2022) to maintain genus-0 topology (Fig 1b). Finally, cortical surfaces are extracted from the SDFs using marching cubes (Lorensen & Cline, 1987)(Fig 1c).
Step 2: Although the above processing effectively captures the overall surface shape, it struggles to resolve the intricate, highly folded patterns of gyri and sulci. Therefore, we perform a surface refinement via weakly supervised learning. To do so, we extend the CoSeg method (Ma et al., 2024), which performs T1-weighted surface reconstruction by learning diffeomorphic deformations to fit an initial surface to target cortical surfaces. We modify CoSeg by integrating the dMRI-derived maps with the above-obtained cortical surfaces to train a dMRI-specific model for surface reconstruction and cortical parcellation (Fig 1d and e).
Results:
We perform experiments using the Human Connectome Project Young Adult (HCP-YA) data, including 23 subjects for training and 10 subjects for testing. Fig 2a demonstrates the overall performance of DDSurfer, showing a high degree of alignment with the underlying anatomical structures. Figs 2b and c give comparison results with the T1w-derived surfaces registered to dMRI and the DDParcel-derived surfaces, where we show that DDSurfer can better capture the tissue boundaries. Fig 2d gives the quantitative result of our surface quality against T1w-based surfaces derived by FreeSurfer using metrics 90th-percentile Hausdorff distance (HD90), average symmetric surface distance (ASSD), cortical thickness, and volume. DDSurfer achieves low HD90 (1.3564 mm WM, 2.1710 mm Pial) and ASSD (0.8776 mm WM, 1.1493 mm Pial), minimal volume differences, and cortical thickness (2.7630 mm) close to the reference (2.5881 mm), confirming its reliability and accuracy. Regarding computational time, DDSurfer performs inference of a testing subject within 5 minutes, providing a highly efficient tool for large-scale data processing.
Conclusions:
This study presents DDSurfer, a deep-learning-based method that can accurately and efficiently achieve cortical surface reconstruction directly from dMRI data.
Modeling and Analysis Methods:
Diffusion MRI Modeling and Analysis 1
Methods Development 2
Segmentation and Parcellation
Novel Imaging Acquisition Methods:
Diffusion MRI
Multi-Modal Imaging
Keywords:
Cortex
Machine Learning
Modeling
Segmentation
STRUCTURAL MRI
White Matter
WHITE MATTER IMAGING - DTI, HARDI, DSI, ETC
Other - Cortical Surface Reconstruction
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.
Han, X., Pham, D. L., Tosun, D., Rettmann, M. E., Xu, C., & Prince, J. L. (2004). CRUISE: cortical reconstruction using implicit surface evolution. NeuroImage, 23(3), 997–1012.
Lorensen, W. E., & Cline, H. E. (1998). Marching cubes: A high resolution 3D surface construction algorithm. In Seminal graphics: pioneering efforts that shaped the field (pp. 347-353).
Ma, Q., Li, L., Robinson, E. C., Kainz, B., & Rueckert, D. (2024). Weakly supervised learning of cortical surface reconstruction from segmentations. In Lecture Notes in Computer Science (pp. 766–777). Springer Nature Switzerland.
Ma, Q., Li, L., Robinson, E. C., Kainz, B., Rueckert, D., & Alansary, A. (2022). CortexODE: Learning Cortical Surface Reconstruction by Neural ODEs. In IEEE Transactions on Medical Imaging. https://ieeexplore.ieee.org/document/9888110
Shastin, D., Genc, S., Parker, G. D., Koller, K., Tax, C. M. W., Evans, J., Hamandi, K., Gray, W. P., Jones, D. K., & Chamberland, M. (2022). Surface-based tracking for short association fibre tractography. NeuroImage, 260(119423), 119423.
Wen, Y., He, L., von Deneen, K. M., & Lu, Y. (2013). Brain tissue classification based on DTI using an improved fuzzy C-means algorithm with spatial constraints. Magnetic Resonance Imaging, 31(9), 1623–1630.
Yeo, B. T. T., Krienen, F. M., Sepulcre, J., Sabuncu, M. R., Lashkari, D., Hollinshead, M., Roffman, J. L., Smoller, J. W., Zöllei, L., Polimeni, J. R., Fischl, B., Liu, H., & Buckner, R. L. (2011). The organization of the human cerebral cortex estimated by intrinsic functional connectivity. Journal of Neurophysiology, 106(3), 1125–1165.
Zhang, F., Cho, K. I. K., Seitz-Holland, J., Ning, L., Legarreta, J. H., Rathi, Y., Westin, C.-F., O’Donnell, L. J., & Pasternak, O. (2024). DDParcel: Deep learning anatomical brain parcellation from diffusion MRI. IEEE Transactions on Medical Imaging, 43(3), 1191–1202.
Zhang, F., Daducci, A., He, Y., Schiavi, S., Seguin, C., Smith, R. E., Yeh, C.-H., Zhao, T., & O’Donnell, L. J. (2022). Quantitative mapping of the brain’s structural connectivity using diffusion MRI tractography: A review. NeuroImage, 249, 118870.
Acknowledgment:
National Key R&D Program of China (No. 2023YFE0118600), National Natural Science Foundation of China (No. 62371107), National Institutes of Health (R01MH108574, P41EB015902, R01MH074794, R01MH125860, R01MH119222, R01MH132610, R01NS125781).