Related Experiment Video
Updated: Nov 3, 2025
06:48
Automated Segmentation of Cortical Grey Matter from T1-Weighted MRI Images
Published on: January 7, 2019
9.1K
Multi-Scale Squeeze U-SegNet with Multi Global Attention for Brain MRI Segmentation
Chaitra Dayananda1, Jae-Young Choi2, Bumshik Lee1
1Department of Information and Communications Engineering, Chosun University, Gwangju 61452, Korea.
Sensors (Basel, Switzerland)
|June 2, 2021
Summary
This study introduces an attention-based U-SegNet for brain MRI segmentation, improving accuracy and reducing parameters. The novel approach enhances feature extraction for precise segmentation of cerebrospinal fluid, gray matter, and white matter.
Area of Science:
- Medical image analysis
- Artificial intelligence in neuroimaging
- Deep learning for segmentation
Background:
- Convolutional Neural Networks (CNNs) are widely used for medical image segmentation but face challenges with redundant feature extraction and inaccurate long-range dependency modeling.
- Conventional encoder-decoder architectures in CNNs can lead to inefficiencies and suboptimal segmentation accuracy for brain tissues.
- Accurate segmentation of brain tissues in Magnetic Resonance Images (MRIs) is crucial for diagnosis and treatment planning.
Purpose of the Study:
- To propose a novel attention-based U-SegNet architecture for enhanced multi-scale feature extraction in brain MRI segmentation.
- To address the limitations of conventional CNNs, including redundant feature learning and poor long-range dependency modeling.
- To improve the accuracy and computational efficiency of automatic brain tissue segmentation.
Main Methods:
- Developed a multi-scale feature extraction method using attention-based convolutional learning within the U-SegNet architecture.
- Integrated a global attention module to refine feature extraction and enhance the representational power of the CNN.
- Incorporated an attention-based multi-scale fusion strategy to combine local features with global dependencies and utilized fire modules to reduce computational complexity.
Main Results:
- Achieved high segmentation accuracies: 94.81% for cerebrospinal fluid (CSF), 95.54% for gray matter (GM), and 96.33% for white matter (WM).
- Demonstrated a 2.5% improvement in Dice Similarity Index (DSI) compared to previous U-SegNet approaches.
- Reduced the number of learnable parameters by 4.5 times, indicating significant computational efficiency.
Conclusions:
- The proposed attention-based U-SegNet effectively segments brain tissues in MRIs with improved accuracy and reduced computational cost.
- The novel attention mechanisms enhance feature representation and integration of multi-scale information for precise segmentation.
- This approach offers a reliable and efficient solution for automatic brain MRI segmentation, outperforming existing methods.
