Valid and invalid foot contacts on force platforms during gait analysis - A dataset for automated classification

Background
The incorporation of force platform data, i.e., ground reaction force (GRF) and center of pressure (COP), in biomechanical gait analysis requires valid foot contacts on the force platforms. Foot contacts are considered valid if the foot has complete and exclusive contact with a force platform while the other foot does not touch this force platform. Compliance with these criteria is usually assessed subjectively by visual inspection by the person conducting the gait analysis.

Research question
Can the assessment to distinguish between invalid and valid foot contacts on a force platform during gait analysis be automated using a machine learning model?

Methods
Twenty healthy participants (10 female and 10 male) underwent gait analysis using GRF and COP measurements during the stance phases on one force platform (Kistler, Switzerland). Six typical cases of invalid foot
contacts in force platform measurements were simulated, with simple and diffcult valid and invalid foot contacts recorded in each case. Each measurement was classified by two examiners through visual inspection and two video recordings (Qualisys, Sweden) of the lower body. A Support Vector Machine (SVM) was trained to distinguish valid and invalid foot contacts on the force platform based on preprocessed GRF and COP time-series signals. different combinations of GRF and COP data as input to the SVM were evaluated.

Results
Using a combination of anterior-posterio and medio-lateral COP as input to the SVM achieved the highest accuracy of 96.6% (100% of simple cases and 93.2% of diffcult cases).

Significance
The development of an automated classification model based on machine learning has the potential to enhance the precision of foot contact assessments on force platforms during gait analysis. This can benefit experimental procedures by improving the quality of data and increasing the usability of (publicly) available datasets through simplified data cleaning.

Keywords:
ground reaction force, GRF, center of pressure, COP, support vector machines, SVM, pattern recognition

Published Dataset
The data collected for this study is shared publicly for reproducibility and as a benchmark for similar approaches. The .c3d files contain the raw, unprocessed analog force plate data (Kistler 9287CA, 60x90cm, 400Hz). The pre-processed data is shared as .csv file(s) - either separated by channel or all in one file. The recorded videos cannot be published for data privacy protection reasons. The metadata for each subject and each study is also available in .csv format.