Evaluation of muscle tone in upper limbs using electromyography and machine learning

Abstract

Muscle tone is commonly defined as the resistance to passive stretch while a patient attempts to stay relaxed, primarily regulated by the Central Nervous System (CNS). Assessing muscle tone is crucial for clinical diagnosis and treatment monitoring in individuals with neurological disorders. Traditionally, muscle tone is evaluated using ordinal scales such as the Modified Ashworth Scale (MAS) and the Unified Parkinson’s Disease Rating Scale (UPDRS). However, these scales are often criticized for their subjectivity. Several studies have explored objective assessments using kinematic components, torque, and electromyography (EMG), primarily focusing on spasticity and rigidity (hypertonia), with hypotonia receiving less attention. Moreover, there is a lack of objective methods capable of evaluating multiple types of muscle tone abnormalities simultaneously. This thesis aims to identify and determine parameters and characteristics for evaluating the full spectrum of muscle tone, ranging from hypotonia to hypertonia, and establish the best protocol for this assessment. This is achieved using electromyography (EMG) data from the biceps and triceps brachii muscles combined with machine learning (ML) techniques to classify the signals. Initially, a literature review was conducted to identify the most appropriate and consistent tools for this purpose. Based on the findings, a comprehensive protocol was designed, incorporating a variety of stretches, including active, slow passive, and fast passive movements, to capture different aspects of muscle tone. The study included 39 participants: 10 with spasticity, 10 with rigidity, 9 with hypotonia, and 10 healthy individuals. Following data collection, datasets were created based on the stretches and combinations of stretches performed. These datasets were then subjected to machine learning classification algorithms (KNN, RF, GBM and SVM), to cluster individuals into groups. All datasets demonstrated accuracies above 90%. Notably, the dataset combining active stretches and slow passive movements achieved 99% accuracy and was identified as the most suitable option for clinical application due to its practicality. This approach requires the therapist to perform only one type of passive movement, offering a shorter and more efficient protocol compared to the inclusion of all three types of movements. In conclusion, the study successfully identified key parameters and characteristics of the EMG signal that can be used for muscle tone evaluation. A comprehensive method was developed, highlighting the most effective protocol along with the necessary processing and classification steps. This method provides a valuable tool for the objective assessment of muscle tone abnormalities, aiding in the evaluation of interventions and treatments to enhance patient recovery and quality of life.