GC-Net: Global context network for medical image segmentation.

Area of Science:

• Medical image analysis
• Deep learning for computer vision
• Biomedical imaging

Background:

• Medical image segmentation is crucial for clinical applications but challenging due to image variability and complex object structures.
• Current deep learning methods often overlook global context features, limiting their segmentation performance.
• There is a need for advanced neural networks that leverage global context for improved medical image segmentation.

Purpose of the Study:

• To develop a novel neural network architecture that incorporates global context feature information for diverse medical image segmentation tasks.
• To address the limitations of existing deep learning models in capturing comprehensive contextual information for segmentation.

Main Methods:

• A global context network (GC-Net) featuring distinct feature encoding and decoding modules was proposed.
• The decoding module utilizes a global context attention (GCA) block and a squeeze and excitation pyramid pooling (SEPP) block for enhanced feature representation and multi-scale fusion.
• A weighted cross-entropy loss function was implemented to improve the balance between segmented and non-segmented regions.

Main Results:

• The GC-Net demonstrated superior performance on multiple medical segmentation tasks, including intracranial and retinal vessel segmentation, cell contour segmentation, and lung segmentation.
• Experiments were conducted on three public and one local dataset, validating the network's effectiveness.
• The proposed network outperformed state-of-the-art methods based on standard evaluation metrics.

Conclusions:

• The developed deep convolutional neural network, incorporating a global context attention mechanism, effectively achieves accurate medical image segmentation across various tasks.
• The GC-Net offers a promising solution for enhancing the accuracy and reliability of medical image segmentation in clinical practice.