Related Experiment Video
Updated: Nov 23, 2025
04:48
Swin-PSAxialNet: An Efficient Multi-Organ Segmentation Technique
Published on: July 5, 2024
655
An Automatic Glioma Segmentation System Using a Multilevel Attention Pyramid Scene Parsing Network
Zhenyu Zhang1, Shouwei Gao1, Zheng Huang2
1School of Mechatronic Engineering and Automation, Shanghai University, Shanghai, China.
Current Medical Imaging
|January 4, 2021
Summary
This study introduces a novel multilevel attention pyramid scene parsing network (MLAPSPNet) for improved glioma segmentation. The MLAPSPNet enhances segmentation accuracy by effectively aggregating multiscale context and multilevel features.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Computer Vision
Background:
- Glioma segmentation is challenging due to variations in tumor shape and size.
- Accurate segmentation is crucial for effective treatment planning and monitoring.
Purpose of the Study:
- To propose a novel deep learning model for automated glioma segmentation.
- To enhance segmentation performance by integrating multiscale context and multilevel features.
Main Methods:
- A multilevel attention pyramid scene parsing network (MLAPSPNet) was developed.
- Input data included combined T1 pre-contrast, FLAIR, and T1 post-contrast MRI sequences.
- The network incorporates multilevel pyramid pooling modules (PPMs) and attention gates.
Main Results:
- The MLAPSPNet achieved a Dice Similarity Coefficient (DSC) of 0.885, sensitivity of 0.933, and Jaccard score of 0.8.
- Multilevel pyramid pooling modules and attention gates improved DSC by 0.029 and 0.022, respectively.
- The model outperformed several existing networks, including Res-UNet, Dense-UNet, RCA-UNet, DeepLab V3+, and UNet++.
Conclusions:
- The proposed MLAPSPNet achieves state-of-the-art performance in glioma segmentation.
- The integration of multilevel pyramid pooling modules and attention gates significantly improves segmentation accuracy.
