Related Experiment Video
Updated: Jul 9, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
2.8K
TCI-UNet: transformer-CNN interactive module for medical image segmentation
Xuan Bian1, Guanglei Wang1, Yanlin Wu1
1College of Electronic and Information Engineering, Hebei University, Hebei 071002, China.
Biomedical Optics Express
|November 29, 2023
Summary
This study introduces TCI-UNet, a novel medical image segmentation model that enhances feature extraction by integrating transformer and convolutional neural network (CNN) components. TCI-UNet improves global context capture and local detail processing for superior diagnostic accuracy.
Area of Science:
- Medical image analysis
- Computer vision
- Deep learning for healthcare
Background:
- Medical image segmentation is vital for disease diagnosis and treatment.
- UNet is a popular network but struggles with long-range dependencies.
- Existing transformer-UNet models have limitations in feature extraction flexibility.
Purpose of the Study:
- To develop an improved UNet architecture for medical image segmentation.
- To enhance the modeling of global and local features in medical images.
- To overcome the limitations of fixed receptive fields and single feature extraction methods in current models.
Main Methods:
- Proposed a transformer-CNN interactive (TCI) feature extraction module.
- Integrated TCI into the UNet architecture to create TCI-UNet.
- Enhanced transformer self-attention for better attention map guidance and resource allocation.
- Incorporated local multi-scale information to complement global context.
Main Results:
- Achieved high Dice Coefficient Image Enhancement (DCIE) values: 93.81% on LiTS-2017 and 88.22% on ISIC-2018.
- Ablation studies confirmed the effectiveness of the TCI module.
- TCI-UNet demonstrated superior accuracy and generalization compared to state-of-the-art networks.
Conclusions:
- TCI-UNet effectively captures global context and local details in medical images.
- The proposed TCI module enhances feature extraction capabilities.
- TCI-UNet offers improved performance for medical image segmentation tasks.
