Human Gait Analysis: Machine Learning-Based Classification of Gait Disorders

Abstract

Clinical gait analysis is a central approach for assessing human gait, which forms the foundation for clinicians to make accurate diagnoses and to develop effective treatment plans. The complexity of gait analysis data and corresponding clinical tasks has motivated researchers to investigate the suitability of machine learning (ML) methods to solve gait analysis tasks. The use of ML methods aims to improve the efficiency of clinical gait analysis and to contribute to better informed decision-making. The present thesis addresses existing gaps and limitations and makes a significant methodological contribution to explainable ML approaches for complex multi-class gait classification tasks. For this purpose, traditional ML and deep learning approaches are developed and investigated in two real-world use cases utilizing clinical datasets containing data from patients with functional gait disorders and cerebral palsy. This thesis proposes for the first time explainability approaches for ML methods for clinical gait data that enable clinicians to trace decisions. Additionally, it demonstrates the usefulness of explainability methods in identifying biases. In addition to a systematic evaluation of data handling strategies concerning feature scaling, feature extraction, and data imbalances, this thesis investigates the discriminative power of ground reaction force and joint angle data. A significant contribution of the current thesis lies also in the publication of a large-scale real- world dataset named GaitRec. This dataset serves as a benchmark, providing a foundation for assessing the performance of ML approaches in a standardized way. Finally, the present thesis identifies future research directions that have the potential to advance the field of automated classification of clinical gait data.