A Machine Learning Pipeline for Gait Analysis in a Semi Free-Living Environment

Abstract

This paper presents a novel approach to creating a graphical summary of a subject's activity during a protocol in a Semi Free-Living Environment. Thanks to this new visualization, human behavior, in particular locomotion, can now be condensed into an easy-to-read and user-friendly output. As time series collected while monitoring patients in Semi Free-Living Environments are often long and complex, our contribution relies on an innovative pipeline of signal processing methods and machine learning algorithms. Once learned, the graphical representation is able to sum up all activities present in the data and can quickly be applied to newly acquired time series. In a nutshell, raw data from inertial measurement units are first segmented into homogeneous regimes with an adaptive change-point detection procedure, then each segment is automatically labeled. Then, features are extracted from each regime, and lastly, a score is computed using these features. The final visual summary is constructed from the scores of the activities and their comparisons to healthy models. This graphical output is a detailed, adaptive, and structured visualization that helps better understand the salient events in a complex gait protocol.

Keywords: Human Activity Recognition; IMU; change point detection; free-living; graphical feedback; wearable sensor.

Figure 1

Successive steps of our processing pipeline to render a graphical feedback from a semi-FLE acquisition.

Figure 2

Description of the semi-controlled protocol. Numbers displayed indicate the position of the subject during its path.

Figure 3

One aCC signal and its associated unbiased autocorrelation. Definition of $P{1}_{CC}$ (blue dot) and $P{2}_{CC}$ features (red dot).

Figure 4

Confusion matrix for the segmentation and classification steps of our processing pipeline.

Figure 5

Graphical feedbacks from (a) HSU, (b) PSU1, (c) PSU2.

Figure 6

Evaluation of pre-hospitalization acquisition and post-hospitalization acquisition for a subject who has undergone knee ligamentoplasty. Post-operation evaluation displays a worse state. (a) Pathological Subject 3 graphical feedback Pre-surgery PSU3A; (b) Pathological Subject 3 graphical feedback: Post-surgery PSU3B.

Figure 7

Evaluation of the robustness of selected features. Features with low dispersion and high discrimination between classes. The blue horizontal line shows the average value of the feature for all healthy subjects, the red horizontal line shows the median value of the feature for all healthy subjects, and the dotted lines correspond to the 75th/25th percentiles. Each boxplot corresponds to 10 computations of the feature on a walking regime on 10 degraded ranges. Boxplots are displayed with specific colors depending on their associated subject: blue for a healthy subject, red for PSU1, and green for PSU2.