Machine Learning methods for EMG-based stroke patient movement analysis

Dies ist die aktuellste Version dieses Eintrags.

Kurzfassung

Stroke causes loss of brain functionality leading to restricted muscle activity on the affected body side. This directly impacts the Electromyography (EMG) signal. In order to develop a robotic rehabilitation system for stroke patients, the signal changes have to be analyzed. Hence, the goal of this master’s thesis is to identify the latent dimensionality of EMG data and to derive a criterion for the health status of stroke patients. Thus, an experiment has been conducted with seven stroke patients and two reference subjects. Different machine learning methods have been evaluated. At first, dimensionality reduction techniques c.f. PCA, NMF, and Gain Shape k-Means are utilized and evaluated using clustering scores. Those scores show, that reference subjects have higher values, whereas patients have values below a threshold. Applying PCA and NMF synergies have been identified. In order to preserve at least 95% of the information contained, data of patients need more synergies than data of reference subjects. Furthermore, patients’ patterns changes considerably on both sides compared to reference subjects’. These results show that synergies are very promising and insightful to evaluate EMG data. Further investigations are necessary to be able to generalize these findings.

Verfügbare Versionen dieses Eintrags

• Machine Learning methods for EMG-based stroke patient movement analysis. (deposited 20 Okt 2014 11:13) [Gegenwärtig angezeigt]