config.md

Config Setting

1. subfolder_name: the name of subfolder in the input/output folder containing organs .nii.gz files. For example:

   INPUT / OUTPUT (--input_folder / --output_folder)
   └── case_001
       └── segmentations <- (subfolder_name)
               ├── liver.nii.gz
               ...
               └── veins.nii.gz
   

2. class_map: the label mapping dict of organ and their labels.

   [!WARNING] This parameter will be deprecated soon.

   All organs on this list will be read and loaded, but only the ones listed in target_organs will be processed by ShapeKit.

3. target_organs: the organs selected for postprocessing.

   By adding or deleting the organs listed, you can choose which organs you want to process. For example:

       target_organs:
           - bladder
           - colon
           - duodenum
           - femur
           - intestine
           - kidney
           - liver
           - lung
           - pancreas
   

4. organ_adjacency_map: a dictionary used in the reassign_false_positives function.

   This section identifies organs that sit close together where the AI might mislabel a border. By listing these anatomical neighbors, you help the software distinguish between touching structures—like the liver and pancreas—to ensure your results are accurate.

   Exmaple:

   organ_adjacency_map:
       lung_left: [postcava]
       lung_right: [postcava]
       liver: [kidney_right, pancreas]
   

   This means that during segmentation: (1) Parts of the predicted lung_left may be false positives that actually belong to postcava. (2) Similarly, liver may mistakenly include areas from kidney_right or pancreas.

   Note: This map is one-directional, i.e., if lung_left → postcava is defined, it does not imply the reverse (postcava → lung_left). This directionality reflects common misclassification patterns, not anatomical symmetry.

5. affine_reference_file_name: file to load affine reference info.

6. if_save_combined_label: boolean parameter that controls whether to save the combined labels as a .nii.gz file after processing. For example:

   OUTPUT
   └── case_001
       ├── combined_labels.nii.gz <- (if_save_combined_label)
       └── segmentations
               ├── liver.nii.gz
               ...
               └── veins.nii.gz
   

7. vertebrae_engine: which vertebrae module to run. shapekit (default) is the existing mask-based module. shapekit_pro is the evidence-gated engine that repairs vertebra labels against the case CT by recoloring inside the prediction envelope (no deletion of predicted bone); it requires the case CT and falls back to shapekit when the CT is absent.

8. ct_file_name / ct_root: how shapekit_pro finds the CT. The engine first looks for <input_case>/<ct_file_name>; when ct_root is set it also tries <ct_root>/<case_id>/<ct_file_name>.