Related Experiment Video
Updated: Aug 26, 2025
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
A novel MCF-Net: Multi-level context fusion network for 2D medical image segmentation
Lizhu Liu1, Yexin Liu2, Jian Zhou2
1Engineering Research Center of Automotive Electrics and Control Technology, College of Mechanical and Vehicle Engineering, Hunan University, Changsha 410082, China; National Engineering Laboratory of Robot Visual Perception and Control Technology, School of Robotics, Hunan University, Changsha 410082, China.
This study introduces a novel Multi-Level Context Fusion Network (MCF-Net) to enhance medical image segmentation. MCF-Net improves upon U-Net by better integrating multi-level contextual information for more accurate disease diagnosis and analysis.
Area of Science:
- Medical imaging
- Deep learning
- Computer vision
Background:
- Medical image segmentation is vital for disease diagnosis and analysis.
- U-Net models show promise but struggle with multi-level context and feature extraction.
Purpose of the Study:
- To introduce the novel Multi-Level Context Fusion Network (MCF-Net).
- To enhance U-Net's performance in medical image segmentation tasks.
- To improve the integration of multi-level contextual information and feature extraction capabilities.
Main Methods:
- Designed three modules: Hybrid Attention-based Residual Atrous Convolution (HARA), Multi-Scale Feature Memory (MSFM), and Multi-Receptive Field Fusion (MRFF).
- HARA module combines atrous spatial pyramid pooling and attention for multi-receptive field features.
- MSFM and MRFF modules fuse features across different levels to extract contextual information.
Main Results:
- MCF-Net was evaluated on ISIC 2018, DRIVE, BUSI, and Kvasir-SEG datasets.
- Demonstrated competitive performance against other U-Net models.
- Showcased effectiveness on images with varying sizes and anatomy.
Conclusions:
- MCF-Net offers significant improvements for 2D medical image segmentation.
- The network has potential as a general-purpose deep learning model for medical image analysis.
- Enhanced contextual information fusion leads to superior segmentation performance.
