Clinical gait analysis: comparing explicit state duration HMMs using a reference-based index.

Area of Science:

• Biomechanics
• Medical Informatics
• Machine Learning

Background:

• Clinical gait analysis relies on optical motion capture to compare patient gait to healthy references.
• Interpreting high-dimensional gait data is challenging, necessitating advanced analytical tools.
• Machine learning offers potential for identifying key gait phases and joint angles, and quantifying gait deviations.

Purpose of the Study:

• To develop and evaluate a novel machine learning framework for analyzing clinical gait data.
• To quantify the differences between healthy and pathological gait using a stochastic approach.
• To introduce a reference-based measure for assessing gait similarity and generating visualizations.

Main Methods:

• An explicit state duration hidden Markov model (HMM) was employed to model gait timeseries data.
• A reference-based measure was developed to compare observations within defined states.
• The model's accuracy and measure's performance were assessed using data from healthy subjects and arthritis patients.

Main Results:

• The explicit state duration HMM effectively models gait timeseries data for individuals and groups.
• The reference-based measure allows for quantitative comparison of healthy and pathological gait across states, joints, and subjects.
• An overall gait index was derived, proving useful for group comparisons and individual gait assessment.

Conclusions:

• The proposed HMM framework provides a robust method for interpreting complex gait datasets.
• The developed measures enable precise quantification of gait pathology and similarity.
• This approach enhances clinical gait analysis by offering objective insights into patient mobility and disease impact.