Related Experiment Video
Updated: Sep 24, 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
3.8K
Attention-modulated multi-branch convolutional neural networks for neonatal brain tissue segmentation
Xunli Fan1, Shixi Shan1, Xianjun Li2
1School of Information Science and Technology, Northwest University, Xi'an, 710127, China.
Computers in Biology and Medicine
|May 7, 2022
Summary
This study introduces a novel deep learning network for precise neonatal brain tissue segmentation, improving upon traditional methods. The attention-modulated multi-branch convolutional neural network achieves state-of-the-art results in segmenting brain structures in infants.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Neuroscience
Background:
- Accurate measurement of neonatal brain structures is crucial for assessing growth and development.
- Conventional manual segmentation methods are time-consuming and inefficient.
- Existing deep learning models struggle with immature neonatal brain MR images due to unique attributes.
Purpose of the Study:
- To develop a novel attention-modulated multi-branch convolutional neural network (AMCNN) for accurate neonatal brain tissue segmentation.
- To overcome the limitations of current methods in segmenting complex and immature neonatal brain structures.
- To improve the efficiency and accuracy of neonatal brain magnetic resonance image (MRI) analysis.
Main Methods:
- Proposed an encoder-decoder framework incorporating multi-scale convolutions for feature extraction and multi-branch attention modules for contextual information capture.
- Introduced spatial attention connections to enhance feature propagation and accelerate model convergence.
- Implemented and compared the proposed network against baseline methods on three neonatal brain datasets (dCBP2021, NBAtlas, dHCP2017).
Main Results:
- The AMCNN achieved high performance across datasets, with average Dice similarity coefficients (DSC) and Hausdorff distances (HD) demonstrating superior segmentation accuracy.
- Specific results included average DSC/HD of 0.9116/8.1289 on dCBP2021, 0.8786/11.7863 on NBAtlas, and 0.9253/7.7968 on dHCP2017 for gray matter, white matter, and cerebrospinal fluid segmentation.
- The proposed method significantly outperformed baseline approaches in segmenting neonatal brain tissues.
Conclusions:
- The novel attention-modulated multi-branch convolutional neural network presents a competitive state-of-the-art solution for neonatal brain tissue segmentation.
- The method effectively addresses the challenges posed by immature neonatal brain characteristics in MRI.
- The developed network offers a more efficient and accurate alternative to conventional manual segmentation techniques.
