Related Experiment Video
Updated: Sep 28, 2025
05:33
Three-Dimensional Shape Modeling and Analysis of Brain Structures
Published on: November 14, 2019
7.2K
Automated Skull Stripping in Mouse Functional Magnetic Resonance Imaging Analysis Using 3D U-Net
Guohui Ruan1,2, Jiaming Liu1,2, Ziqi An1,2
1School of Biomedical Engineering, Southern Medical University, Guangzhou, China.
Frontiers in Neuroscience
|April 4, 2022
Summary
A new 3D U-Net method accurately automates mouse brain extraction for functional MRI (fMRI) analysis. This approach significantly improves efficiency and consistency over manual methods, crucial for mouse fMRI studies.
Area of Science:
- Neuroimaging
- Computational Neuroscience
- Medical Image Analysis
Background:
- Skull stripping is a vital preprocessing step in mouse functional magnetic resonance imaging (fMRI) analysis.
- Manual brain extraction is labor-intensive and prone to variability, hindering large-scale mouse fMRI studies.
- Developing automated, efficient, and accurate skull-stripping methods is essential for advancing mouse fMRI research.
Purpose of the Study:
- To investigate the efficacy of a 3D U-Net based deep learning model for automated brain extraction in mouse fMRI.
- To compare the performance of the 3D U-Net method against established skull-stripping techniques (RATS, SHERM) and manual segmentation.
- To evaluate the impact of automated segmentation on subsequent resting-state fMRI analyses.
Main Methods:
- Two 3D U-Net models were independently trained on T2-weighted anatomical and T2*-weighted functional mouse brain images.
- The trained models were rigorously tested on both internal and external datasets to assess generalization.
- Performance was quantitatively evaluated using metrics such as Dice, Jaccard index, and Hausdorff distance.
Main Results:
- The 3D U-Net models achieved high accuracy for brain extraction from T2-weighted images (Dice > 0.984) and T2*-weighted images (Dice > 0.964).
- The proposed method demonstrated superior performance compared to RATS and SHERM across all evaluated metrics.
- Resting-state fMRI analyses (seed-based and group ICA) using 3D U-Net segmentation showed high consistency with manual segmentation results.
Conclusions:
- The 3D U-Net based method provides an accurate and reliable automated solution for mouse brain extraction in fMRI studies.
- This deep learning approach can effectively replace time-consuming manual segmentation, improving efficiency and reproducibility.
- The validated 3D U-Net method facilitates robust downstream analysis of mouse fMRI data.
