ReGA

Overall framework

Component

Dataset

The datasets can be downloaded from the following links:

• CLINIC: Zenodo
• KITS19: GitHub
• MSD_T10: Google Drive

To reproduce the visual examples illustrated in the manuscript, please perform inference on the following cases using their associated pre-trained weights:

• Case A: Case 203 from the KITS19 dataset
• Case B: Case 5 from the MSD_T10 dataset
• Case C: Case 246 from the KITS19 dataset

How to use

Source training

For training details, refer to the instructions on the CTPelvic1K GitHub repository.
The source model weights pre-trained on the CLINIC dataset are provided at ReGA_main/weights/cascade_fullres_CTpelvic_fold0.path.

ReGA

Use

cd ReGA_min/code
python TA_eval.py --mode train_only
python TA_eval.py --mode test_only

to get the test-time adaptation results.
The model weights adapted with ReGA on the KITS19 and MSD_T10 datasets are available at the following locations:

• ReGA_main/weights/final_adapted_model_Kits.path
• ReGA_main/weights/final_adapted_model_MSD.path