Related Experiment Video
Updated: Aug 14, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
2.9K
DSCA-Net: A depthwise separable convolutional neural network with attention mechanism for medical image segmentation
Tong Shan1, Jiayong Yan2,3, Xiaoyao Cui3
1School of Health Science and Engineering, University of Shanghai for Science and Technology, Shanghai 200093, China.
Mathematical Biosciences and Engineering : MBE
|January 18, 2023
Summary
This study introduces DSCA-Net, a lightweight neural network for medical image segmentation that significantly reduces computational complexity compared to U-Net. DSCA-Net achieves improved segmentation accuracy across multiple datasets.
Area of Science:
- Medical Image Analysis
- Deep Learning
- Computer Vision
Background:
- Accurate medical image segmentation is essential for analysis.
- U-Net and its variants are popular but computationally intensive.
- High complexity limits practical applications of current models.
Purpose of the Study:
- To propose a novel lightweight neural network, DSCA-Net, for efficient medical image segmentation.
- To reduce the computational complexity associated with U-Net models.
- To enhance segmentation performance using attention mechanisms and depthwise separable convolutions.
Main Methods:
- Introduced depthwise separable convolution and attention mechanisms into a U-shaped architecture.
- Developed three attention modules: Pooling Attention (PA), Context Attention (CA), and Multiscale Edge Attention (MEA).
- Reduced network parameters by 71.6% compared to U-Net (2.2 M parameters).
Main Results:
- DSCA-Net demonstrated significant parameter reduction.
- Achieved improved Dice coefficients on four public datasets: ISIC 2018 (+5.49%), thyroid (+4.28%), lung (+1.61%), and nuclei (+9.31%) compared to U-Net.
- The proposed attention modules effectively improved segmentation performance.
Conclusions:
- DSCA-Net offers a computationally efficient alternative for medical image segmentation.
- The integration of novel attention modules enhances segmentation accuracy.
- DSCA-Net shows strong potential for practical medical image analysis applications.
