Automatic Classification of Functional Gait Disorders.

Area of Science:

• Biomechanics
• Medical Engineering
• Computational Medicine

Background:

• Functional gait disorders (GDs) significantly impact mobility.
• Accurate classification of GDs is crucial for effective treatment.
• Current classification methods may lack objective, quantitative measures.

Purpose of the Study:

• To investigate the efficacy of ground reaction force (GRF) parameterizations for classifying functional GDs.
• To establish a performance baseline for automated GD classification using a large dataset.
• To analyze the impact of different GRF representations on classification accuracy.

Main Methods:

• Utilized GRF measurements from 279 GD patients and 161 healthy controls.
• Applied various GRF parameterizations: direct parameters, Principal Component Analysis (PCA)-based, and combined PCA.
• Employed linear discriminant analysis for feature discrimination and pursued binary (healthy vs. GD) and multiclass classification.

Main Results:

• Demonstrated the potential of GRF parameterizations for discriminating between healthy and impaired gaits.
• Achieved promising results in multiclass classification of GDs affecting hip, knee, ankle, and calcaneus.
• Identified imbalanced class sizes and varied measurement sessions as significant factors influencing classification accuracy.

Conclusions:

• GRF measurements offer a viable basis for automated functional gait disorder classification.
• The study provides a foundational performance benchmark for future research in this domain.
• Further development should account for data characteristics like class imbalance for improved automated GD detection.