/Oneshot_landmark_detection

Official code for "Oneshot Medical Landmark Detection' (MICCAI 2021 early accepted)

Primary LanguagePythonApache License 2.0Apache-2.0

oneshot-medical-landmark

Implementation of "One-shot Medical Landmark Detection" -- MICCAI 2021

Paper link: https://doi.org/10.1007/978-3-030-87196-3_17

Usage

Environment

python >= 3.9

pip install -r req.txt

Data Preparation

We up load the four X-ray datasets in google drive, including our proposed BMPLE dataset (Leg X-ray). Please download the files and unzip into the diretory ``dataset''

head dataset (Cephalometric)

hand dataset

chest dataset

leg dataset (BMPLE)

Training

Stage 1: self-supervised training

python -m scripts_SSL.train --tag xxtag --dataset head

Automatically choose the template

python -m scripts_SSL.SCP --tag xxtag --dataset head

You can set the selected id by SCP and set to config/SSL_{dataset}.yaml and config/TPL_{dataset}.yaml, otherwise you can choose to use the best template.

python -m scripts_SSL.train --tag xxtag --dataset head --finetune 1

Stage 2: self-training

python -m scripts_TPL.train --tag xxtag --dataset head

Citation

Please ite our paper if it helps.

@article{yao2021one,
  title={One-Shot Medical Landmark Detection},
  author={Yao, Qingsong and Quan, Quan and Xiao, Li and Zhou, S Kevin},
  journal={Medical Image Computing and Computer Assisted Intervention},
  year={2021}
}

License

This code is released under the Apache 2.0 license. Please see the LICENSE file for more information

We actively welcome your issues!