Comparing Machine-Learning Approaches for Skull Stripping in Human, Macaque, and Piglet Brains
Presented During: Poster Session 3
Friday, June 27, 2025: 01:45 PM - 03:45 PM
Friday, June 27, 2025: 01:45 PM - 03:45 PM
Presented During: Poster Session 4
Saturday, June 28, 2025: 01:45 PM - 03:45 PM
Saturday, June 28, 2025: 01:45 PM - 03:45 PM
Poster No:
1576
Submission Type:
Abstract Submission
Authors:
Heesoo Park1, Joonmin Lee1, Jinho Bae1, Seoyeong Ha1, Hangjoon Jo2
Institutions:
1Hanyang Univ., Seoul, Seoul Metropolitan City, 2College of Medicine, Hanyang Univ., Seoul, Seoul Metropolitan City
First Author:
Co-Author(s):
Introduction:
Skull stripping (SS) is a necessary preprocessing step in neuroimaging analysis. Classical SS methods generally perform well in most cases. However, when imaging data with artifacts or quality issues, these methods often produce suboptimal results. In such cases, workarounds are necessary, involving a cumbersome process of iteratively adjusting tool parameters and comparing the results through trial and error. These challenges have led to the development of machine learning (ML)-based SS methods, which have significantly improved the results. In this study, we aim to compare the performance of ML-based SS methods under variations in input brain masks. Specifically, since we expect that the accuracy of ML results will vary due to differences in the boundaries of manually drawn masks, we observed this effect by introducing changes to the brain mask boundaries from different species during training.
Methods:
We used brain T1-weighted MRI datasets from humans (N=125; 75 samples for training (T), 25 for validation (V), and 25 for estimation (E)), macaques (N=125; T=75, V=25, E=25), and piglets (N=30; T=19, V=7, E=4), using either pre-defined or manually drawn brain masks (Eskildsen et al., 2012; Neff, 2019; Stanke et al., 2023). All datasets were automatically cropped using the AFNI packages (Cox, 1996), and no additional preprocessing for image enhancement was applied to isolate the effects of mask boundary variations and differences between ML models, thereby minimizing any potential confounding influences. These datasets were employed for training, validation, and estimation in four ML models (2D U-Net, 3D U-Net, V-Net, and VoxResNet) (Azad et al., 2024; Chen et al., 2016). The gold standard masks (GSM) for each brain mask were pre-established, and the masks used for training were dilated or eroded by one voxel at a time from the GSM. This process was repeated up to five steps to generate new masks. Geometric similarity, overlap fraction, and extra fraction were calculated between the output results and the gold standard data to evaluate the robustness and accuracy of the SS results for each method. The resulting brain masks from the SS procedure were then aligned to species-specific template brains (Illinois Pig MRI Atlas version, N27, NMT version 2.0 for piglet, human, and macaque) for visualization of the results (Fil et al., 2021; Holmes et al., 1998; Jung et al., 2021).
Results:
Boundary variations caused by erosion and dilation masks yielded different results depending on the species. Human data exhibited relatively robust performance against mask boundary variations, whereas macaque and piglet data were more sensitive to these changes (Fig. 1). Analysis of evaluation metrics revealed a trend of decreasing geometric similarity and overlap fraction as mask boundaries were altered (Fig. 2). These findings indicate that 3D U-Net demonstrated the most robust performance across human, macaque, and piglet datasets.
·Figure 1
·Figure 2
Conclusions:
Our research indicates that accurately delineating brain mask boundaries contributes significantly to enhancing the robustness and accuracy of ML-based SS models' performance, particularly when training on non-human datasets. This study showed that variations in brain mask boundaries can influence model performance, with macaque and piglet data being more sensitive to subtle boundary changes compared to human data.
Modeling and Analysis Methods:
Methods Development 1
Segmentation and Parcellation 2
Keywords:
Machine Learning
MRI
Segmentation
Structures
Other - Skull Stripping
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.
Chen, H., Dou, Q., Yu, L., & Heng, P.-A. (2016). Voxresnet: Deep voxelwise residual networks for volumetric brain segmentation. arXiv preprint arXiv:1608.05895.
Cox, R. W. (1996). AFNI: software for analysis and visualization of functional magnetic resonance neuroimages. Computers and Biomedical research, 29(3), 162-173.
Eskildsen, S. F., Coupé, P., Fonov, V., Manjón, J. V., Leung, K. K., Guizard, N., Wassef, S. N., Østergaard, L. R., Collins, D. L., & Initiative, A. s. D. N. (2012). BEaST: brain extraction based on nonlocal segmentation technique. NeuroImage, 59(3), 2362-2373.
Fil, J. E., Joung, S., Zimmerman, B. J., Sutton, B. P., & Dilger, R. N. (2021). High-resolution magnetic resonance imaging-based atlases for the young and adolescent domesticated pig (Sus scrofa). Journal of Neuroscience Methods, 354, 109107.
Holmes, C. J., Hoge, R., Collins, L., Woods, R., Toga, A. W., & Evans, A. C. (1998). Enhancement of MR images using registration for signal averaging. Journal of computer assisted tomography, 22(2),
324-333.
Jung, B., Taylor, P. A., Seidlitz, J., Sponheim, C., Perkins, P., Ungerleider, L. G., Glen, D., & Messinger, A. (2021). A comprehensive macaque fMRI pipeline and hierarchical atlas. NeuroImage, 235, 117997.
Neff, E. P. (2019). PRIME-DE is primed with primate data. Lab Animal, 48(1), 26-26.
Stanke, K. L., Larsen, R. J., Rund, L., Leyshon, B. J., Louie, A. Y., & Steelman, A. J. (2023). Automated identification of piglet brain tissue from MRI images using Region-based Convolutional Neural
Networks. PloS one, 18(5), e0284951.