Related Experiment Video
Updated: Jul 28, 2026
06:48
Automated Segmentation of Cortical Grey Matter from T1-Weighted MRI Images
Published on: January 7, 2019
8.9K
Efficient segmentation of fetal brain MRI based on the physical resolution
Yunzhi Xu1, Jiaxin Li1, Xue Feng2
1Key Laboratory for Biomedical Engineering of Ministry of Education, College of Biomedical Engineering & Instrument Science, Zhejiang University, Hangzhou, China.
Medical Physics
|July 15, 2024
Summary
Accurate fetal brain segmentation is achievable using magnetic resonance imaging (MRI) physical resolution. A deep learning module enables high apparent resolution segmentation, resulting in smaller and faster models with negligible accuracy loss.
Area of Science:
- Medical Imaging
- Neuroscience
- Artificial Intelligence
Background:
- Fetal brain magnetic resonance imaging (MRI) resolution impacts brain development measures.
- Reconstructed 3D fetal brain MRI often show higher apparent resolution than the original physical resolution.
Purpose of the Study:
- Demonstrate accurate fetal brain segmentation based on MRI physical resolution.
- Achieve high apparent resolution segmentation using a simple deep learning module.
Main Methods:
- Retrospective study with 150 adult and 80 fetal brain MRIs.
- Downsampled high-resolution adult MRIs to assess segmentation accuracy impacts.
- Estimated fetal MRI physical resolution, downsampled images, and used upsampling strategies for restoration.
- Evaluated segmentation accuracy of ConvNet models on original and downsampled images using Dice coefficients.
Main Results:
- Negligible degradation in fetal brain segmentation accuracy when apparent resolution exceeded physical resolution (0.26%–1.8% reduction with downsampling factors of 4/3, 2, and 4).
- Proposed method yielded 7x smaller and 10x faster models with a downsampling factor of 4.
- No significant differences in accuracy compared to original data.
Conclusions:
- Efficient and accurate fetal brain segmentation models can be developed.
- Models should be based on the physical resolution of MRI acquisitions.
