Dataset for "Development and validation of FootNet; a new kinematic algorithm to improve foot-strike and toe-off detection in treadmill running"

This dataset includes the input features and target labels needed to train and test FootNet. The input features include the distal tibia anteroposterior velocity, ankle plantar/dorsi flexion angle and foot centre of mass anteroposterior and vertical velocities. Additionally, ground reaction force data and trial names are also included.

Keywords:
treadmill running, biomechanics, running gait, locomotion, gait, running, kinematics, ground reaction forces
Subjects:
Medical and health interface

Cite this dataset as:
Rodriguez Rivadulla, A., Chen, X., Weir, G., Cazzola, D., Trewartha, G., Hamill, J., Preatoni, E., 2021. Dataset for "Development and validation of FootNet; a new kinematic algorithm to improve foot-strike and toe-off detection in treadmill running". Bath: University of Bath Research Data Archive. Available from: https://doi.org/10.15125/BATH-00965.

Export

[QR code for this page]

Data

StepDetectionStudy … 001.zip
application/zip (716MB)
Software: GNU LGPL 3.0

Data and Google Colab (Jupyter) notebooks to replicate algorithm development and testing

GitHub repository including further details on the algorithm, its development and use

Creators

Xi Chen
University of Bath

Gillian Weir
University of Massachusetts

Dario Cazzola
University of Bath

Joseph Hamill
University of Massachusetts

Ezio Preatoni
University of Bath

Contributors

University of Bath
Rights Holder

Documentation

Data collection method:

This dataset includes data coming from five different datasets collected in three independent laboratories (see associated publication for more details). It includes treadmill running kinematics and kinetics processed to obtain the previously mentioned variables and chopped in running gait cycles.

Data processing and preparation activities:

The original datasets were fully reprocessed as described in the Methods section of the associated publication.

Additional information:

The project directory StepDetectionStudy is organised as follows: - Data > OriginalDatasets: Folder containing the entire datasets (*_dataset.npy files). - Data > DataFolds.npy: File containing the training data grouped in 5 folds. - Data > TestingSet.npy: File containing the testing set. Data are organised as Python dictionaries containing the kinematic input features ['X'], label vectors ['Y'], metadata about the trials ['meta'] and vertical GRF ['GRFv']. Each of those dictionary keys contains a list with nested lists with the structure participant > trial > stride. For instance, `dataset['X'][0][0][0]` accesses the kinematic input features characterising the first stride recorded in the first trial of the first participant in dataset. - CrossValidation > Models: Folder containing the five models developed during cross validation. - CrossValidation > Results: Folder containing the summary performance metrics for each model on its corresponding validation set and Bland-Altman plots comparing foot strike, toe off and contact times as predicted by FootNet vs gold standard method. - FinalTest > FootNet_best_candidate: Folder containing the best set of parameters resulting from cross validation. Summary performance metrics on testing set and Bland-Altman plots comparing foot strike, toe off and contact times as predicted by FootNet vs gold standard method. - FinalTest > y_and_yhat.mat: File containing testing predictions, target labels and metadata from testing stride cycles for posterior analyses in Matlab presented in the paper. - FinalModel: Folder containing the final updated model resulting from FinalTest as a SavedModel directory (Tensorflow model format) and as .h5. - Notebooks > TrainTest_Split.ipynb: Google Colab (Jupyter) notebook demonstrating how the dataset splitting was performed, including training and testing (70/30) and further folding of training dataset in 5 folds. - Notebooks > CrossValidation.ipnyb. Google Colab (Jupyter) notebook that performs 5-fold cross-validation and selects the best set of weights as best candidate for the final test. - Notebooks > FinalTest.ipnyb: Google Colab (Jupyter) notebook that updates the best candidate model resulting from cross-validation with the 5 folds as training set and performs the final test on the testing set.

Documentation Files

README.md
text/plain (4kB)
Software: GNU LGPL 3.0

README file with instruction on how to use FootNet and how to navigate the associated Github repository (https://github.com/adrianrivadulla/FootNet)

Technical-Details.md
text/plain (1kB)
Software: GNU LGPL 3.0

Instructions on how to run the notebooks

Funders

PhD studentship

NURVV

PhD studentship

Publication details

Publication date: 26 July 2021
by: University of Bath

Version: 1

DOI: https://doi.org/10.15125/BATH-00965

URL for this record: https://researchdata.bath.ac.uk/id/eprint/965

Related papers and books

Rivadulla, A., Chen, X., Weir, G., Cazzola, D., Trewartha, G., Hamill, J., and Preatoni, E., 2021. Development and validation of FootNet; a new kinematic algorithm to improve foot-strike and toe-off detection in treadmill running. PLOS ONE, 16(8), e0248608. Available from: https://doi.org/10.1371/journal.pone.0248608.

Related datasets and code

Rivadulla, A. R., and Needham, L., 2021. adrianrivadulla/FootNet: First version of FootNet. Version v1.0.0. Zenodo. Available from: https://doi.org/10.5281/ZENODO.5532715.

Contact information

Please contact the Research Data Service in the first instance for all matters concerning this item.

Contact person: Adrian Rodriguez Rivadulla

Departments:

Faculty of Humanities & Social Sciences
Health