Development of segmentation/classification model using kidney cancer pathology images

Segmentation Model Description

• Purpose: Display normal, benign, and malignant tumor segments (Segmentaiton)
• Model architecture: U-Net (efficientb0 backbone)
• Input: Image with completed data preprocessing
• Output: Mask image with area marked
• training_dataset: 10,000 WSI (80% of total), image size adjusted 1,024
• training element: a. Loss function: bce_jaccard_loss b. Optimizer: Adam c. Epoch: 120 (default) d. Learning rate: 1e-4 e. Batch size: 1 f. Evaluation metric: iou_score

1. Host environments

• OS==Windows 11 Home
• RAM==32 GB
• GPU== NVIDIA Geforce RTX 4070
• Storage== 3TB
• Docker==20.10.21

2. Required Libraries

• segmentation-models==1.0.1
• numpy==1.26.1
• openslide-python==1.3.1
• opencv-python==4.8.1.78
• jsons==1.6.3
• pandas==2.1.4
• scikit-learn==1.3.2