This repository contains the code required to replicate the results from the following publication (in submission).

Video-based biomechanical analysis captures disease-specific movement signatures of different neuromuscular diseases

Parker S. Ruth*, Scott D. Uhlrich*, Constance de Monts, Antoine Falisse, Julie Muccini, Sydney Covitz, Shelby Vogt-Domke, John Day, Tina Duong†, Scott Delp†

*Contributed equally, †Contributed equally

This code has been tested on MacOS.

Code Installation

1. Install the (mini)conda or mamba Python environment manager.

2. Open a terminal and navigate to an empty working directory where you will download the code and data. Clone the GitHub repository and enter it:

3. Initialize the conda environment for the repository:

cd [path_to_empty_working_directory]
git clone https://github.com/stanfordnmbl/opencap-fshd-dm-analysis.git
cd opencap-fshd-dm-analysis
conda env create -f environment.yml -n opensim-nmd
conda activate opensim-nmd
cd ..

Data Access

The dataset is hosted on Zenodo: https://doi.org/10.5281/zenodo.13788592

Data can be downloaded through the browser, or through the command line as follows:

mkdir datadir
cd datadir
zenodo_get -d 10.5281/zenodo.13788592
cd ..

At this point, your working directory should be organized as follows:

[working_directory]
├── datadir
│   ├── README.md
│   ├── feature_key.csv
│   ├── opencap_data.zip
│   ├── opencap_markers.png
│   ├── participant_info.csv
│   └── video_features.csv
└── opencap-fshd-dm-analysis
    ├── LICENSE
    ├── README.md
    ├── environment.yml
    ├── feature_extraction
    └── figure_generation

Unzip the opencap_data .zip directory.

unzip datadir/opencap_data.zip -d datadir

Feature Extraction

The feature extraction pipeline uses the snakemake. This is a Makefile-like tool that automates feature extraction across all of the activities. This may take 10 minutes to run or more depending on your machine. Use the lines below to run the pipeline. At any time, the pipeline can be halted by pressing CTL-C. When restarted, it snakemake will resume from where it left off.

cd opencap-fshd-dm-analysis/feature_extraction
snakemake -c1

To parallelize over n cores use -c[n] (e.g. -c4 parallelizes over 4 cores).

To reduce the verbosity of the terminal output, add the --quiet flag.

Figure Generation

The code for figure generation code is in Jupyter notebooks. Use these commands to launch Jupyter Lab.

cd ../figure_generation
jupyter lab

At this point, you will be able to select individual notebooks and run all cells to generate our figures.

Intraclass correlation is calculated in fig3_icc.ipynb by calling an R script from Python via a shell. On MacOS, the Rscript command can be acquired by installing RStudio. If R is not available on the computer, an alternative is to run the fig3_icc_no_Rscript.ipynb file, which reads all of the ICC from a pre-computed CSV file.