Related Experiment Video
Updated: Aug 2, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
2.8K
GL-Segnet: Global-Local representation learning net for medical image segmentation.
Di Gai1,2,3, Jiqian Zhang4, Yusong Xiao4
1School of Mathematics and Computer Sciences, Nanchang University, Nanchang, China.
Frontiers in Neuroscience
|April 20, 2023
Summary
A new Global-Local representation learning net (GL-Segnet) improves medical image segmentation by capturing both global context and local details. This method effectively filters background noise and enhances crucial features for superior segmentation accuracy.
Area of Science:
- Neuroscience
- Medical Imaging
- Computer Vision
Background:
- Medical image segmentation is crucial in neuroscience but challenging due to background noise.
- Existing methods struggle with both long-range and short-range dependencies, often neglecting geometric details.
Purpose of the Study:
- To introduce a novel network, GL-Segnet, for improved medical image segmentation.
- To address limitations of current methods by integrating global and local feature learning.
Main Methods:
- Developed a Feature Encoder using Multi-Scale Convolution (MSC) and Pooling (MSP) for global semantic and local geometric information.
- Incorporated a global semantic feature extraction module to filter irrelevant background.
- Utilized an Attention-enhancing Decoder to refine features and proposed a hybrid loss function.
Main Results:
- GL-Segnet demonstrated superior performance over state-of-the-art methods on datasets including Glas, ISIC, Brain Tumors, and SIIM-ACR.
- Achieved improvements in both subjective visual quality and objective evaluation metrics.
Conclusions:
- GL-Segnet effectively segments medical images by balancing global and local feature representation.
- The proposed network offers a significant advancement in medical image segmentation accuracy and detail preservation.
