Related Experiment Video
Updated: Jul 28, 2025
08:49
Whole-Brain Single-Cell Imaging and Analysis of Intact Neonatal Mouse Brains Using MRI, Tissue Clearing, and Light-Sheet Microscopy
Published on: August 1, 2022
3.6K
Fetal brain tissue annotation and segmentation challenge results.
Kelly Payette1, Hongwei Bran Li2, Priscille de Dumast3
1Center for MR Research, University Children's Hospital Zurich, University of Zurich, Zurich, Switzerland; Neuroscience Center Zurich, University of Zurich, Zurich, Switzerland.
Medical Image Analysis
|June 2, 2023
Summary
Automatic segmentation of fetal brain MRI using deep learning is crucial for prenatal neurodevelopment analysis. The Fetal Tissue Annotation (FeTA) Challenge benchmarked 21 algorithms, with one asymmetrical U-Net showing superior performance.
Area of Science:
- Medical Imaging
- Neuroscience
- Artificial Intelligence
Background:
- In-utero fetal MRI is vital for diagnosing and analyzing the developing human brain.
- Manual segmentation of fetal brain structures is time-consuming, error-prone, and has inter-observer variability.
- Automatic segmentation is essential for quantitative analysis of prenatal neurodevelopment in research and clinical settings.
Purpose of the Study:
- To encourage the development of automatic segmentation algorithms for the fetal brain.
- To establish a benchmark for automatic multi-tissue segmentation algorithms for the developing human brain in utero.
Main Methods:
- The Fetal Tissue Annotation (FeTA) Challenge 2021 utilized the FeTA Dataset, an open dataset of fetal brain MRI reconstructions.
- Twenty international teams submitted 21 algorithms, primarily based on deep learning methods like U-Nets.
- Algorithms were evaluated based on technical and clinical perspectives for segmenting seven different fetal brain tissues.
Main Results:
- All participating deep learning algorithms, mainly U-Nets, showed similar performance across various architectures and processing steps.
- Ensemble learning methods were employed by four of the top five performing teams.
- One algorithm, utilizing an asymmetrical U-Net architecture, demonstrated significantly superior performance compared to other submissions.
Conclusions:
- Deep learning, particularly U-Net variants, shows promise for automatic fetal brain segmentation.
- The FeTA Challenge provides a valuable benchmark for future advancements in prenatal neuroimaging analysis.
- Further research into specialized architectures like asymmetrical U-Nets could enhance segmentation accuracy.
