An Automated Classification of Pathological Gait Using Unobtrusive Sensing Technology
Summary
This study uses affordable sensors and machine learning to detect pathological gait patterns after stroke or brain injury. Upper limb, lower limb, and trunk movements are key indicators for identifying abnormal walking.
Area of Science:
- Biomedical Engineering
- Machine Learning
- Gait Analysis
Background:
- Gait analysis is crucial for diagnosing neurological conditions like stroke and acquired brain injury.
- Traditional gait analysis methods can be expensive and intrusive.
- Developing accessible and unobtrusive technologies for gait assessment is a significant need.
Purpose of the Study:
- To integrate affordable sensing technology with machine learning for discriminating healthy and pathological gait patterns.
- To identify the role of specific body parts in gait pattern differentiation.
- To evaluate the performance of instance-based and generative machine learning classifiers for gait analysis.
Main Methods:
- Utilized Kinect skeletal tracking to capture gait sequences under self-paced, distracted, and fast-paced walking conditions.
- Calculated gait features including orientations of the trunk, upper limb, and lower limb.
- Employed two machine learning classifiers: an instance-based k-nearest neighbor and a Gaussian Process Latent Variable Model (generative classifier), evaluated using nested cross-validation.
Main Results:
- The instance-based discriminative model significantly outperformed the generative model in F1-score and macro-averaged error across all walking conditions.
- Feature analysis using ReliefF identified upper limb, lower limb, and trunk movements as the most informative features for detecting pathological gait.
- While the discriminative model excelled in overall accuracy, the generative model showed better performance in micro-averaged error.
Conclusions:
- Affordable sensing technology combined with machine learning effectively discriminates between healthy and pathological gait.
- Gait features related to the upper limb, lower limb, and trunk are critical for identifying gait abnormalities post-stroke or brain injury.
- The findings support the development of accessible tools for neurological gait disorder assessment.
