Related Experiment Video
Updated: Jun 10, 2025
Author Spotlight: Bridging Gaps in Anatomy and Establishing a Foundation for Algorithmic Studies
Published on: December 15, 2023
MG-Net: A fetal brain tissue segmentation method based on multiscale feature fusion and graph convolution attention
Keying Qi1, Chenchen Yan2, Donghao Niu1
1Research Center for Medical AI, Shenzhen Institute of Advanced Technology, Chinese Academy of Sciences, Shenzhen 518055, China; Department of Artificial Intelligence, University of Chinese Academy of Sciences, Beijing 100049, China.
This study introduces an automated deep learning model for fetal brain tissue segmentation, significantly improving accuracy and efficiency for medical professionals. The novel approach enhances feature extraction for better diagnostic support.
Area of Science:
- Medical Imaging
- Neuroscience
- Artificial Intelligence
Background:
- Fetal brain tissue segmentation is crucial for understanding neurodevelopment and congenital diseases.
- Manual segmentation is time-consuming and challenging due to changing fetal anatomy and tissue contrast.
- Automated methods are needed to improve efficiency and accuracy in clinical workflows.
Purpose of the Study:
- To develop an automated deep learning model for efficient and accurate fetal brain tissue segmentation.
- To enhance the diagnostic capabilities of medical professionals through improved segmentation workflows.
- To address the challenges posed by dynamic changes in fetal brain tissue during development.
Main Methods:
- A novel deep learning model incorporating Dual Dilated Attention Block (DDAB) in the encoder for local feature extraction.
- Integration of a Multi-scale Deformable Transformer (MSDT) in the bottleneck for global information processing.
- Utilizing a Graph Convolution Attention Block (GCAB) in the decoder to enhance feature representation.
Main Results:
- The proposed model was trained and tested on the FeTA 2021 and FeTA 2022 datasets.
- Outperformed state-of-the-art models (nnFormer, SwinUNETR, 3DUX-net) in Dice, IoU, and MAE metrics.
- Achieved high performance, e.g., Dice of 0.8666 and IoU of 0.7646 on FeTA 2021, and Dice of 0.8552 and IoU of 0.7470 on FeTA 2022.
Conclusions:
- A novel 3D fetal brain tissue segmentation model based on multi-scale feature fusion and graph convolution attention was developed.
- The model demonstrates significant potential to improve the speed and accuracy of fetal brain tissue segmentation.
- Accurate segmentation is vital for precise diagnosis and understanding of fetal neurodevelopment.
