SUnet: A multi-organ segmentation network based on multiple attention.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Accurate organ segmentation in computed tomography (CT) is vital for medical diagnosis, surgical planning, and treatment decisions.
• Existing methods face challenges in efficiency and accuracy for multi-organ segmentation in abdominal and thoracic regions.

Purpose of the Study:

• To propose SUnet, a novel and efficient fully attention-based neural network for multi-organ segmentation in abdominal and thoracic CT images.
• To enhance feature extraction, reduce model parameters, and improve cross-scale feature integration for better segmentation performance.

Main Methods:

• Development of SUnet, a fully attention-based neural network incorporating an efficient spatial reduction attention (ESRA) module.
• Integration of a multiple attention-based feature fusion module for effective cross-scale feature integration.
• Inclusion of an enhanced attention gate (EAG) module with grouped convolution and residual connections for richer semantic features.

Main Results:

• SUnet achieved an average Dice score of 84.29% on the synapse multiple organ segmentation dataset.
• SUnet achieved an average Dice score of 92.25% on the automated cardiac diagnostic challenge dataset.
• The proposed model outperformed existing methods of similar complexity and size, demonstrating state-of-the-art results.

Conclusions:

• SUnet offers an efficient and effective solution for multi-organ segmentation in abdominal and thoracic CT images.
• The attention-based architecture, including ESRA and EAG modules, significantly enhances segmentation accuracy and feature representation.
• SUnet represents a significant advancement in medical image analysis, with potential to improve clinical decision-making.