Related Experiment Video
Updated: Dec 18, 2025
09:06
Whole-brain Segmentation and Change-point Analysis of Anatomical Brain MRI—Application in Premanifest Huntington's Disease
Published on: June 9, 2018
12.5K
AssemblyNet: A large ensemble of CNNs for 3D whole brain MRI segmentation
Pierrick Coupé1, Boris Mansencal1, Michaël Clément1
1CNRS, Univ. Bordeaux, Bordeaux INP, LABRI, UMR5800, F-33400, Talence, France.
Neuroimage
|June 12, 2020
Summary
AssemblyNet, an ensemble deep learning method, improves whole brain segmentation by using multiple U-Nets. This novel approach achieves reliable and robust fine-grained brain structure segmentation, outperforming existing methods.
Area of Science:
- Neuroimaging
- Artificial Intelligence
- Computational Neuroscience
Background:
- Accurate whole brain segmentation is crucial for neuroscience research.
- Deep learning (DL) methods face challenges with high anatomical label counts and limited training data.
- Existing DL approaches often use single or few Convolutional Neural Networks (CNNs).
Purpose of the Study:
- To develop a novel ensemble deep learning framework for whole brain segmentation.
- To address the challenge of segmenting fine-grained brain structures with numerous labels.
- To improve the accuracy, robustness, and reliability of brain segmentation.
Main Methods:
- Introduced AssemblyNet, an ensemble framework comprising two assemblies of U-Nets.
- Implemented knowledge sharing among neighboring U-Nets.
- Utilized a two-stage decision process: initial segmentation and higher-resolution refinement ('amendment').
- Employed majority voting for final segmentation decisions.
- Investigated semi-supervised learning to enhance performance.
Main Results:
- AssemblyNet demonstrated competitive performance against state-of-the-art methods (U-Net, Joint Label Fusion, SLANT).
- The method showed high scan-rescan consistency.
- AssemblyNet proved robust to disease effects.
- Semi-supervised learning further improved segmentation performance.
Conclusions:
- AssemblyNet offers a reliable and effective solution for challenging whole brain segmentation tasks.
- The ensemble approach with knowledge sharing and refinement enhances segmentation accuracy.
- The framework's robustness and consistency highlight its clinical and research potential.
