How wuyanan513/pulmonary-vessel-segmentation-and-artery-vein-separation works
Main Components (Top Level)
• Input Processing Layer
• Model Architecture Layer
• Training Pipeline
• Output Processing Layer
Detailed Component Breakdown
a. Input Layer:
• Preprocessing steps (handled by utils.py)
b. Model Architecture Layer (Primary Focus):
• Encoder Path (with 3D Contextual Transformer modules)
• Decoder Path (with 3D Contextual Transformer modules)
• Skip Connections (with Double-Attention modules)
c. Training Pipeline:
• Loss Function Component (from loss.py)
• Combination of Focal Loss and Dice Loss
• Training Loop (from main.py)
• Optimization Process
d. Output Layer:
• Segmentation Output
• Vessel Segmentation Branch
• Artery-Vein Separation Branch
Implementation Details
Use these symbols and conventions:
• Rounded rectangles for data representations
Key Architectural Elements to Highlight
• 3D CoT (Contextual Transformer) Module
• Double Attention Module
• U-Net backbone structure
• Multi-task output heads
Data Flow Indicators
• Input CT scan → Preprocessing → Encoder → Decoder → Segmentation Maps
• Training flow with loss calculation and backpropagation
• Validation/Inference path
Remember to:
• Use consistent styling for similar components
• Notes for key architectural innovations (transformer and attention modules)
• Relationship between the files (main.py, model.py, loss.py, utils.py) and their corresponding components in the architecture