CPFTransformer: transformer fusion context pyramid medical image segmentation network

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Convolutional Neural Networks (CNNs) show promise in medical image segmentation but struggle with boundary blurring and capturing global context.
• The inherent locality of CNNs limits their ability to process long-range semantic information.
• Transformer models excel at capturing global information, offering a potential solution to CNN limitations.

Purpose of the Study:

• To develop an improved medical image segmentation network that overcomes the limitations of traditional CNNs.
• To effectively fuse global and multi-scale contextual information for enhanced segmentation accuracy.
• To introduce a novel network architecture that integrates transformer capabilities with context pyramid mechanisms.

Main Methods:

• A Context Pyramid Fusion transformer (CPFTransformer) network is proposed, utilizing the Swin Transformer for feature extraction.
• An Edge-Aware module based on a context pyramid is introduced to emphasize local features like edges and corners.
• A layered Swin Transformer serves as the encoder, and a symmetric Swin Transformer acts as the decoder for resolution restoration.

Main Results:

• The CPFTransformer achieved a segmentation accuracy of 79.87% DSC and 20.83% HD on the Synapse multi-organ segmentation task.
• Experimental evaluations on the Synapse and ACDC datasets demonstrated the method's effectiveness.
• The proposed method shows significant improvements in medical image segmentation tasks.

Conclusions:

• The CPFTransformer effectively combines context pyramid mechanisms and transformers for fast and accurate medical image segmentation.
• This approach significantly enhances the precision and reliability of medical diagnosis.
• The method has potential for extension to the segmentation of other organs.