GU-Net: Causal relationship-based generative medical image segmentation model.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Medical image segmentation is crucial but challenging due to anatomical variations.
• Current convolutional neural networks often lack interactive training and feedback, limiting their adaptability.
• Existing models may only perform segmentation for specific diseases, lacking generalizability.

Purpose of the Study:

• To propose GU-Net, a causal relationship-based generative model for improved medical image segmentation.
• To address limitations of existing methods by integrating interactive training and feedback mechanisms.
• To enhance the robustness and accuracy of medical image segmentation, particularly in complex scenarios.

Main Methods:

• Developed GU-Net, integrating a U-Net decoder with a counterfactual attention mechanism and CBAM.
• Employed a Generative Adversarial Network (GAN) framework with discriminator-based backpropagation for alternate training.
• Implemented feedback mechanisms to improve feature representation and model stability.

Main Results:

• GU-Net demonstrated superior segmentation performance across various datasets, including those with limited data.
• Achieved consistent improvements in Dice scores compared to existing attention-based U-Net models.
• Showcased enhanced capabilities in handling challenging segmentation tasks and reducing overfitting.

Conclusions:

• GU-Net offers a more robust and accurate solution for medical image segmentation.
• The proposed generative approach with interactive training enhances model expressiveness and learning.
• GU-Net shows significant potential for improving diagnostic accuracy in diverse clinical applications.