Laplacian-guided hierarchical transformer: A network for medical image segmentation

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Accurate medical image segmentation is vital for clinical diagnosis and treatment planning.
• Transformer models face challenges in capturing high-frequency features crucial for detailed segmentation, like edge texture extraction.

Purpose of the Study:

• To develop a novel model architecture that enhances Transformer capabilities for medical image segmentation.
• To improve the integration of high-frequency and low-frequency features for superior segmentation accuracy.

Main Methods:

• A Laplacian pyramid extracts high-frequency features, while a Local-Global Feature Aggregation Module captures low-frequency features.
• A Feature Interaction Fusion module integrates these features, and a bridging module transfers spatial information using layer-wise attention.
• The model was evaluated on the Synapse dataset using Dice Similarity Coefficient and Hausdorff Distance.

Main Results:

• The proposed model achieved state-of-the-art performance in 2D medical image segmentation.
• Achieved a Dice Similarity Coefficient of 84.10% and a Hausdorff Distance of 12.78.
• Demonstrated significant improvements over existing medical image segmentation methods.

Conclusions:

• The novel architecture effectively captures and integrates both high- and low-frequency features for advanced medical image segmentation.
• The model's performance on the Synapse dataset indicates its potential to enhance clinical diagnosis and treatment planning.