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

Area of Science:

• Biomedical Engineering
• Human-Computer Interaction
• Signal Processing

Background:

• Monitoring patients in semi free-living environments generates long, complex time-series data.
• Analyzing human behavior, especially locomotion, from this data is challenging.
• Existing methods lack user-friendly and condensed visualization for complex activity data.

Purpose of the Study:

• To develop a novel graphical approach for summarizing subject activity during protocols in semi free-living environments.
• To create an easy-to-read and user-friendly visualization of human behavior, particularly locomotion.
• To enable rapid analysis of newly acquired time-series data using a learned graphical representation.

Main Methods:

• Utilized a pipeline of signal processing and machine learning algorithms.
• Applied adaptive change-point detection to segment raw data from inertial measurement units.
• Automated segment labeling, feature extraction, and activity scoring, comparing against healthy models.

Main Results:

• Developed an innovative pipeline to process complex time-series data from inertial measurement units.
• Generated a graphical representation that condenses human behavior and locomotion into an understandable format.
• The visualization effectively summarizes activities and aids in understanding salient events in gait protocols.

Conclusions:

• The novel graphical summary provides a detailed, adaptive, and structured visualization of subject activity.
• This approach simplifies the analysis of complex gait protocols in semi free-living environments.
• The method enhances the understanding of human behavior and locomotion patterns through accessible data representation.