Dual encoder network with transformer-CNN for multi-organ segmentation.

Area of Science:

• Medical Imaging
• Computer Vision
• Artificial Intelligence

Background:

• Medical image segmentation is crucial for diagnostics and treatment planning.
• Traditional Convolutional Neural Networks (CNNs) struggle with global context, while Transformers excel but miss fine details.
• Existing U-shaped transformer models in medical imaging face limitations in capturing intricate details.

Purpose of the Study:

• To propose a novel dual-encoder network for accurate multi-organ medical image segmentation.
• To leverage the strengths of both CNNs and Transformers for improved segmentation performance.
• To enhance the fusion of global and local features for precise boundary delineation.

Main Methods:

• A dual-encoder framework integrating a Swin-Transformer for global context and a CNN for local features.
• Introduction of fusion modules to combine convolutional and transformer-derived features.
• Feature fusion is implemented via skip connections to smooth segmentation boundaries.

Main Results:

• Achieved Dice Similarity Coefficient (DSC) of 80.68% on the synapse multi-organ CT dataset.
• Achieved DSC of 91.12% on the Automated Cardiac Diagnosis Challenge (ACDC) dataset.
• Demonstrated superior performance and more consistent boundary matching compared to state-of-the-art methods.

Conclusions:

• The proposed dual-encoder network effectively combines CNN and Transformer capabilities for superior medical image segmentation.
• The method shows significant improvements in multi-organ segmentation accuracy, particularly on challenging 2D images.
• Ablation studies confirm the effectiveness of the key components in the proposed segmentation framework.