Using data from the Microsoft Kinect 2 to determine postural stability in healthy subjects: A feasibility trial

Abstract

The objective of this study was to determine whether kinematic data collected by the Microsoft Kinect 2 (MK2) could be used to quantify postural stability in healthy subjects. Twelve subjects were recruited for the project, and were instructed to perform a sequence of simple postural stability tasks. The movement sequence was performed as subjects were seated on top of a force platform, and the MK2 was positioned in front of them. This sequence of tasks was performed by each subject under three different postural conditions: "both feet on the ground" (1), "One foot off the ground" (2), and "both feet off the ground" (3). We compared force platform and MK2 data to quantify the degree to which the MK2 was returning reliable data across subjects. We then applied a novel machine-learning paradigm to the MK2 data in order to determine the extent to which data from the MK2 could be used to reliably classify different postural conditions. Our initial comparison of force plate and MK2 data showed a strong agreement between the two devices, with strong Pearson correlations between the trunk centroids "Spine_Mid" (0.85 ± 0.06), "Neck" (0.86 ± 0.07) and "Head" (0.87 ± 0.07), and the center of pressure centroid inferred by the force platform. Mean accuracy for the machine learning classifier from MK2 was 97.0%, with a specific classification accuracy breakdown of 90.9%, 100%, and 100% for conditions 1 through 3, respectively. Mean accuracy for the machine learning classifier derived from the force platform data was lower at 84.4%. We conclude that data from the MK2 has sufficient information content to allow us to classify sequences of tasks being performed under different levels of postural stability. Future studies will focus on validating this protocol on large populations of individuals with actual balance impairments in order to create a toolkit that is clinically validated and available to the medical community.

Fig 1. Experimental layout for data collection: subjects sat on a rectangular box that was placed in the center of the KFP at a distance of 2.7 meters from the MK2.

The MK2 was mounted on a tripod 0.92 meters from the floor.

Fig 2. Three postural conditions.

a) Both feet on the ground. b) One foot off the ground. c) Both feet off the ground.

Fig 3

a) The 25 joint centroids that are used by the MK2 to characterize full body motion. b) The 16 joint centroids used to characterize the trunk and upper extremities. c) The four joint centroids used to characterize the spine.

Fig 4. Classification pipeline for the classification across tasks.

Fig 5. Visualization of raw data that was recorded during one subject’s performance of mBBS Task 5.

We have plotted motion data from the KFP, the “Spine Mid” MK2 centroid, the “Neck” MK2 centroid and the “Head” MK2 centroid on the three axis. Note similarities in “x” and “y” directional data from the KFP and MK2 centroids. Each time series is normalized by its standard deviation for comparison.

Fig 6. t-SNE representation of the dataset for the MK2 (left) and the KFP (right).

A clear separation between tasks can be observed for both devices, but the MK2 data are more separable. Each cluster is formed by the points of the 3 conditions performed under the corresponding task. A zoom over the cluster of task 5 highlights a lower separation of the conditions.

Fig 7

Confusion matrix showing the percent accuracy outcome of condition classification using a) MK2 centroids of the upper body, b) MK2 spinal centroids, and c) KFP center of pressure variance.

Fig 8

Confusion matrix showing the percent accuracy outcome of task classification using a) MK2 centroids of the upper body, b) MK2 spinal centroids, and c) KFP center of pressure variance.

Fig 9. Accuracy of condition classification as number of subjects used to train the model increases.