Therapy Reloaded: Temporal Neural Network-Based Platform for Processing Muscle Activity Measurement in Electromyography-Games

This paper presents a novel approach to the calibration process of an Electromyography sensor using classical and Machine Learning techniques for use within video games. This involves the use of a selection of temporal Machine Learning models such as the Recurrent Neural Network, Long Short-Term Memory and a Transformer network, in order to generalise the outputs of the Electromyography sensor, which is used to measure specific muscle activity. The platform is aimed at individuals with neurological conditions such as Cerebral Palsy, Hemiplegia, Multiple Sclerosis, Stroke and other types of rehabilitative conditions which may require this as a form of therapeutic benefit. The results of the Machine Learning architectures, show that the Recurrent Neural Network with a sequence size of 28 samples, outperformed the other models, across all evaluation metrics, for this type of temporal data.

I. Introduction

Physiotherapy and its associated techniques, are an imperative aspect of rehabilitation. Therefore, enabling faster recovery from neurological limitations such as Cerebral Palsy (CP), Hemiplegia, Multiple Sclerosis (MS), Stroke and other types of rehabilitative conditions which may require this as a form of therapeutic benefit [1]. However, it is crucial for individuals to continue using these therapeutic techniques at home, even though the process can be difficult and tedious. The effectiveness of such practices at home may vary due to different scenarios and circumstances.