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
Multi-scale attention and deep supervision-based 3D UNet for automatic liver segmentation from CT.
Jinke Wang1,2, Xiangyang Zhang2, Liang Guo2
1Department of Software Engineering, Harbin University of Science and Technology, Rongcheng 264300, China.
Mathematical Biosciences and Engineering : MBE
|January 18, 2023
Summary
This study introduces MAD-UNet, a novel deep learning network for accurate automatic liver segmentation in CT scans. The method enhances feature extraction and spatial context learning, significantly improving clinical applicability for hepatoma treatment.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Computer Vision
Background:
- Accurate automatic liver segmentation is crucial for hepatoma treatment.
- Current methods face limitations in accuracy and stability, hindering clinical use.
Purpose of the Study:
- To develop a novel network, MAD-UNet, for improved automatic liver segmentation from CT images.
- To enhance the accuracy and stability of liver segmentation for clinical applications.
Main Methods:
- Proposed MAD-UNet, based on 3D UNet, incorporating multi-scale attention and deep supervision.
- Modified encoder with convolution instead of pooling, added residual modules, and used long-short skip connections (LSSC).
- Aggregated multi-scale features and employed attention mechanisms in the decoder for spatial context capture.
Main Results:
- Achieved high Dice scores: 0.9727 (LiTS17), 0.9752 (SLiver07), and 0.9691 (3DIRCADb).
- Outperformed existing state-of-the-art (SOTA) methods in liver segmentation accuracy.
- Demonstrated effective utilization of multi-stage features and enhanced spatial learning.
Conclusions:
- MAD-UNet effectively leverages multi-stage feature information and enhances spatial learning.
- The method achieves high accuracy, proving to be a promising tool for automatic liver segmentation in clinical settings.
- The proposed approach addresses limitations in current liver segmentation techniques.
