MS-TCNet: An effective Transformer-CNN combined network using multi-scale feature learning for 3D medical image...

Area of Science:

• Medical image processing
• Computer vision
• Artificial intelligence in healthcare

Background:

• Medical image segmentation is crucial for diagnosis and treatment planning.
• Existing Transformer-CNN methods struggle to capture essential multi-scale features.
• Multi-scale features are vital for understanding texture and context in medical images.

Purpose of the Study:

• To propose a novel Transformer-CNN combined network, MS-TCNet, for 3D medical image segmentation.
• To enhance the capture and utilization of multi-scale features for improved segmentation accuracy.
• To develop a computationally efficient model with fewer parameters.

Main Methods:

• Developed MS-TCNet, a network integrating a shunted Transformer and CNN for encoder-decoder architecture.
• Implemented a pyramid decoder enabling feature learning across six scale levels.
• Introduced a lightweight multi-scale feature fusion (MSFF) module for enhanced feature integration.

Main Results:

• MS-TCNet outperformed state-of-the-art methods on three 3D medical image segmentation datasets.
• The model demonstrated effectiveness, robustness, and superiority in segmentation tasks.
• Achieved superior performance with a reduced number of parameters and lower computational complexity.

Conclusions:

• The proposed MS-TCNet effectively learns and utilizes multi-scale features for superior 3D medical image segmentation.
• The model offers a promising solution for accurate and efficient medical image analysis.
• Open-sourced code is available for further research and development.