Config file contains an entire setting for the automated analysis via main.py file and master_script.py. It allows changing methods for blood vessels segmentation as well as changing pre-trained models to use.

config.json file

All Changeble parameters are listed bellow:

Main Key	method/value	method_parameters
segmentation_method_tumor1	thresholding2	method: [th_triangle,th_yen,th_otsu]
segmentation_method_vessel1	random_forest2	model_file: relative path
	unet2	model_file: relative path
	thresholding2	method: [th_triangle,th_yen,th_otsu]
segmentation_postprocessing_tumor	split_tumor_into_core_and_periphery3	periphery_as_ratio_of_max_distance:<0; 1>
distance_tranform		stack_size: int
pixels_to_microns	float
mlflow_logging	bool
mlflow_run_name	str

Example of possible config.json file.

{
    "segmentation_method_tumor": {  
        "method": "thresholding",          # Select segmentation method for tumor channel (here "thresholding")
        "method_parameters": {
            "method": "th_triangle"        # What thresholding method to use: yen, triangle, otsu...
        }
    },
    "segmentation_method_vessel": {  
        "method": "unet",                 # What method to use for blood vessels segmentation
        "method_parameters": {
            "model_file": "ppdm/data/unet_model.pt", # path to the pre-trained model
        }
    },
    "segmentation_postprocessing_tumor": {               # Tumor postprocesing - splitting the brains to core and periphery
        "method": "split_tumor_into_core_and_periphery", # Split tumor to the core and periphery
        "method_parameters": {}                          # No parameter necessary for this method - defaults to 0.2 (20 percent core, 80 periphery)
    },
    "distance_tranform": {                      # (Outer) Distance Transform for the blood vessels distance
        "method_parameters": {                  # How many layers stacked together inside DT aggregation
            "stack_size": 100
        }
    }
    "pixels_to_microns": 4,     # Multiplication constant for converting pixels to micrones
    "mlflow_logging": true      # Bool parameters if the results should be saved to mlflow.
    "mlflow_run_name": "DEMO"   # Name of the experiment which should be used for the logging
}

---

1. see pipeline overview ↩↩

2. see segmentation module documentation and examples ↩↩↩↩

3. Describes the fraction of tumor core and periphery (should add up to 1) - e.g. we assumed 0.2 (20 percent) periphery and 0.8 (80 percent) tumor core. ↩