Interpretable classification for multivariate gait analysis of cerebral palsy.

Area of Science:

• Biomedical Engineering
• Rehabilitation Science
• Data Science in Healthcare

Background:

• The Gross Motor Function Classification System (GMFCS) is standard for assessing Cerebral Palsy (CP) mobility, relying on expert visual evaluation.
• Current CP research often uses gait analysis, but the functional gait patterns' predictive potential for GMFCS remains underexplored.
• Objective gait analysis can offer deeper insights into CP's impact on mobility and inform intervention strategies.

Purpose of the Study:

• To develop and validate a multivariate functional classification method for predicting GMFCS levels from kinematic gait measures.
• To establish a link between gait patterns and GMFCS levels in individuals with and without CP.
• To enhance the objectivity and scientific basis for CP diagnosis and GMFCS level assignment.

Main Methods:

• Utilized a sparse linear functional discrimination framework for interpretable prediction models.
• Applied a multivariate functional classification approach to handle complex gait data.
• Generalized the method for multivariate functional data and multi-class GMFCS level classification.

Main Results:

• Achieved competitive or superior prediction accuracy compared to existing functional classification methods.
• Developed interpretable discriminant functions that correlate with the progression of gait in higher GMFCS levels.
• Demonstrated the effectiveness of kinematic gait measures in classifying GMFCS levels.

Conclusions:

• Successfully generalized a sparse functional linear discrimination framework for interpretable GMFCS classification using gait data.
• The proposed method aids clinicians in more consistent and systematic GMFCS level assignment for CP patients.
• Findings support a scientifically grounded approach to CP assessment and intervention planning.