Explaining the differences of gait patterns between high and low-mileage runners with machine learning

Area of Science:

• Biomechanics
• Sports Science
• Computational Neuroscience

Background:

• Running gait analysis is crucial for understanding injury mechanisms in runners with varying weekly mileages.
• Limited research exists on the direct relationship between specific running gait patterns and weekly mileage.
• Machine learning models offer gait pattern recognition but often lack interpretability (black box problem).

Purpose of the Study:

• To investigate differences in running gait patterns between higher-mileage and low-mileage runners.
• To utilize a Deep Neural Network (DNN) with Layer-wise Relevance Propagation (LRP) for interpretable gait analysis.
• To identify key gait features and body segments contributing to mileage-based gait distinctions.

Main Methods:

• Employed a Deep Neural Network (DNN) model for gait pattern classification.
• Integrated Layer-wise Relevance Propagation (LRP) to interpret the DNN's predictions.
• Analyzed running gait data focusing on differences between high-mileage and low-mileage runners.

Main Results:

• Ankle and knee movements, especially in the sagittal and transverse planes, are significant for distinguishing gait patterns.
• The early stance phase of running contains crucial information for gait pattern recognition.
• LRP provided feasible interpretations of the DNN model's results, enhancing understanding of gait features.

Conclusions:

• Distinct running gait patterns exist between higher-mileage and low-mileage runners, potentially explaining differential injury patterns.
• The ankle and knee joints are key contributors to these mileage-related gait differences.
• LRP technology offers valuable insights for interpreting complex gait analysis models, aiding in injury prevention strategies.