/CTSSeg

Primary LanguagePython

CTSSeg: Consistent Teacher-Student model for magnetic resonance image Segmentation

C. Zhang, Q. He, K. Yan, M. Ma, D. Liu and P. Wang, "CTSSeg: Consistent Teacher-Student model for magnetic resonance image Segmentation," 2023 IEEE International Conference on Multimedia and Expo (ICME), Brisbane, Australia, 2023, pp. 2351-2356, doi: 10.1109/ICME55011.2023.00401.

How to use

  1. Prepare text file by:
{image_path_0},{label_path_0},{mask_path_0}\n
{image_path_1},{label_path_1},{mask_path_1}\n
...
  1. Specify the directory and filename of the text file in the config file by:
dataset:
  txt_dir: {the directory}
  train_txts: [
    [{resample_times_0}, {train_text_filename_0}],
    [{resample_times_1}, {train_text_filename_1}],
    ...
  ]
  eval_txts: [
    [1, {test_text_filename_0}],
    [1, {test_text_filename_1}],
    ...
  ]

You can refer to configs/base.yaml and configs/thigh_full_label.yaml when writing your own config.

  1. python main.py --config_file {config_file_path}

For reproduction on the Atrial Segmentation Challenge dataset

The train/test split is origin from SASSNet

  1. Pretrain model with labeled data only for 25,000 iterations.

  2. Load the pretrained model from step 1, finetune the model with labeled data and unlabeled data simultaneously for 25,000 iterations.