Search Algorithms using Multi-Electrode and Machine Learning Features to improve Multiple Channel NMES systems

Sammanfattning

Neuromuscular Electrical Stimulation (NMES) is used for rehabilitation and Venous Thromboembolism (VTE) prevention by inducing muscle contractions through electrical pulses. The methods used today are bulky, require clinical expertise, and take a long time to prepare, which reduces user compliance. The device developed at Matrix Muscle Support AB would be easier to set up, and the user should be able to wear it throughout the day. The thesis explores two strategies aimed at shortening the search time for finding a pair of Motor Points (MPs) by combining electrodes and finding a search pattern with Machine Learning (ML) methods. This should minimise discomfort, energy consumption and battery device size, and maximize user friendliness and treatment effect. This should also make it possible to increase the number of electrodes without significantly increasing the search time. The multi-electrode algorithm is theoretically faster than the binary algorithm on which I wanted to improve on. The energy for iterating over the multi-electrodes can save between 28 and 78% compared to the binary search. For the ML algorithm, the resulting predictions of MP positions were too low to make a meaningful comparison with the other search algorithms. For future training and testing for the ML more data that can impact the MP positions. In future multi-electrode testing in practice, we need to measure that the impedance is similar over all the electrodes and the electrode connections on the skin. The current needed to trigger the MP needs to be compared for one electrode pair versus a multi-electrode pair.