DECTNet: Dual Encoder Network combined convolution and Transformer architecture for medical image segmentation.

Area of Science:

• Medical Image Analysis
• Computer Vision
• Artificial Intelligence

Background:

• Accurate medical image segmentation is crucial for diagnosis and treatment planning.
• Convolutional Neural Networks (CNNs) and Transformer-based models have shown promise in medical image segmentation.
• Challenges remain due to ambiguous boundaries and complex anatomical structures.

Purpose of the Study:

• To propose a novel Dual Encoder Network (DECTNet) for improved medical image segmentation.
• To address limitations in extracting fine spatial details and global contextual information.
• To enhance the accuracy of structure extraction and segmentation in medical images.

Main Methods:

• DECTNet utilizes a dual-encoder architecture with a CNN-based encoder for spatial details and a Swin Transformer-based encoder for global context.
• A feature fusion decoder integrates multi-scale representations using a channel attention mechanism.
• A deep supervision module is incorporated to accelerate model convergence.

Main Results:

• DECTNet achieved state-of-the-art performance on four diverse medical image segmentation tasks.
• The method outperformed seven other existing models in experimental evaluations.
• Demonstrated superior accuracy in segmenting skin lesions, polyps, COVID-19 lesions, and cardiac MRI.

Conclusions:

• The proposed DECTNet effectively integrates convolutional and Transformer features for robust medical image segmentation.
• DECTNet offers a significant advancement for applications in disease diagnosis and treatment planning.
• The model's ability to handle complex structures and ambiguous boundaries paves the way for more precise clinical applications.