Automated classification of pathological gait after stroke using ubiquitous sensing technology

Area of Science:

• Biomechanics
• Medical Technology
• Machine Learning

Background:

• Gait analysis is crucial for diagnosing neurological disorders.
• Distinguishing pathological gait from normal gait can be challenging in clinical settings.
• Objective, data-driven methods are needed to improve gait assessment accuracy.

Purpose of the Study:

• To develop and evaluate a machine learning model for classifying pathological and healthy gait.
• To assess the effectiveness of combining k nearest neighbor (kNN) and dynamic time warping (DTW) algorithms.
• To determine the performance of the classification model in both original and reduced feature spaces.

Main Methods:

• Collected multi-dimensional gait sequences from post-stroke and healthy individuals using Kinect sensors.
• Utilized trajectories of rotational angle and global velocity of key body joints over time.
• Employed k nearest neighbor (kNN) and dynamic time warping (DTW) for classification, with leave-one-subject-out cross-validation.
• Applied Principal Component Analysis (PCA) for feature space dimensionality reduction.

Main Results:

• The combination of kNN (k=1) and DTW (25% warping window) yielded optimal classification performance.
• The binary classifier achieved a high F1-score, indicating robust discrimination between gait types.
• Using PCA for dimensionality reduction, the model achieved an F1-score of 96% in discriminating pathological gait.

Conclusions:

• Machine learning, specifically kNN and DTW, effectively distinguishes pathological gait from healthy gait.
• The developed method demonstrates high accuracy and potential for clinical application in gait analysis.
• Dimensionality reduction using PCA maintains high classification performance, suggesting an efficient model.