ÃÛ¶¹ÊÓƵ

The ability to generate and control movement is essential for human interactions and overall well-being. Neuromuscular impairment caused by neurological injuries and diseases, such as stroke or limb amputation, severely limits the ability of affected individuals to interact with the surrounding environment, necessitating advanced strategies for rehabilitation and restoration of motor function. Voluntary muscle contractions are initiated by motor neurons that transmit electrical impulses, or action potentials, which propagate along nerves and throughout muscle fibres. A single motor neuron and its associated muscle fibres form a motor unit, which creates a direct link between the nervous system output and electrical signals generated by muscles. High-density surface electromyography (HDsEMG) is a powerful non-invasive technique for analysing neuromuscular function by capturing the electrical activity of motor unit action potentials. However, extracting meaningful information from HDsEMG recordings remains a significant challenge due to the complex spatial and temporal overlap of motor unit activity, particularly in regions with high muscle density, such as the forearm.
This dissertation consists of an introduction, background, and four papers presenting novel methods and algorithms for analysing HDsEMG data. Paper â…  introduces a new algorithm for motor unit decomposition, combining Fast Independent Component Analysis with an iterative peel-off scheme based on spike-triggered averaging and Principal Component Compression. Paper â…¡ presents a method for localizing motor unit action potentials, identified with the decomposition method in Paper â… , estimating the centre of electrical activity using a surface fitting approach and an analytical volume conductor model. Paper â…¢ shifts the focus from motor units, and the reliance on decomposition, to localization of individual time-domain peaks directly in the HDsEMG data, and introduces a new technique for constructing a three-dimensional model of muscle activity. In this model, individual muscles are represented as distinct volumes, which are subsequently used to classify additional HDsEMG signals into specific muscle activations. Paper â…£ extends this approach, applying the volume-based muscle representations to gesture classification and evaluates their generalizability across participants and for novel gestures beyond the original model.