GFNet

GFNet performs automatic segmentation of COVID-19 lung infections from chest CT images to delineate infected regions and enhance boundary detection.

Key Features:

• VGG16 Backbone: Uses VGG16 for hierarchical feature extraction from CT slices.
• Edge-guidance Module (Eg): Fuses features from each layer to emphasize boundary details and improve segmentation of fuzzy boundaries.
• Reverse Attention Mechanism: Applies reverse attention to refine features and improve identification of infection boundaries before integration with the Edge-guidance outputs.
• Multi-layer Feature Fusion: Integrates multi-layer outputs into a final segmentation map for comprehensive delineation of infected areas.
• Deep learning-based segmentation: Employs a deep neural network architecture to automatically segment infected regions in CT images.
• Robustness to variability and low contrast: Targets high variability in infection characteristics and low contrast between infected and normal lung tissues.

Scientific Applications:

• COVID-19 lung infection segmentation: Automatic and precise delineation of infected regions in chest CT images to support radiological assessment.
• Benchmarking and evaluation: Performance evaluated against UNet, UNet++, Inf-Net and few-shot learning methods using Dice coefficient, Sensitivity, and Specificity.
• Generalization testing: Assessed on a "never seen" dataset to evaluate generalization across datasets.

Methodology:

Feature extraction with a VGG16 backbone; fusion across layers via an Edge-guidance module emphasizing boundaries; refinement using a reverse attention mechanism and multi-layer feature fusion to produce the final segmentation.