Skip to content

Code for the paper "Vision-based Estimation of MDS-UPDRS Gait Scores for Assessing Parkinson’s Disease Motor Severity"

License

Notifications You must be signed in to change notification settings

mlu355/PD-Motor-Severity-Estimation

Repository files navigation

Vision-based Estimation of MDS-UPDRS Gait Scores for Assessing Parkinson's Disease Motor Severity

This repository is the official implementation of:

Vision-based Estimation of MDS-UPDRS Gait Scores for Assessing Parkinson's Disease Motor Severity
Mandy Lu, Kathleen Poston, Adolf Pfefferbaum, Edith V. Sullivan, Li Fei-Fei, Kilian M. Pohl, Juan Carlos Niebles, Ehsan Adeli
MICCAI 2020

3D input poses of four subjects with increasing PD motor severity (top) and saliency of classifier (bottom).

The oral presentation of this work at MICCAI can be viewed here.

Requirements

To install requirements, run:

pip install -r requirements.txt

with Python 3 (3.7 used).

Pose Extraction

The Stanford Medicine Gait Dataset is not publically available, but the CASIA Gait Database is available upon request. Any similar video dataset with human movement can be used as input.

To obtain the input poses for the classifier, clone and follow the instructions in the VIBE repo. The script extract_joints <path_to_data> <output_path> takes a folder of videos, runs VIBE on them, and produces outputs joints in the correct format for the classifier. The output files we use are the vibe_output.pkl files generated by VIBE. An example joint file generated by VIBE from the CASIA dataset is in data/sample with an image from the source video. The initial paper detailed a longer data preprocessing sequence which has been condensed in VIBE, so the simpler version is provided.

Data Preprocessing

data_preprocessing.ipynb provides step-by-step instructions for preprocessing joint data.

Training

To train the model(s) in the paper, run this command:

python train.py --model_dir <params_and_output_path> --seed <random_seed>

or use

python train.py

for the default mode. The model_dir should contain a params.json file formatted as in jobs/default/params.json.

References

If you use this code in your research, please cite our paper.

@inproceedings{lu2020vision,
  title={Vision-based Estimation of MDS-UPDRS Gait Scores for Assessing Parkinson’s Disease Motor Severity},
  author={Lu, Mandy and Poston, Kathleen and Pfefferbaum, Adolf and Sullivan, Edith V and Fei-Fei, Li and Pohl, Kilian M and Niebles, Juan Carlos and Adeli, Ehsan},
  booktitle={International Conference on Medical Image Computing and Computer-Assisted Intervention},
  pages={637--647},
  year={2020},
  organization={Springer}
}

These resources were used or cited within the code:

Contact for Questions

Mandy Lu, mlu@cs.stanford.edu

About

Code for the paper "Vision-based Estimation of MDS-UPDRS Gait Scores for Assessing Parkinson’s Disease Motor Severity"

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages