Separating background activity from noise in EMG using a neuromuscular model

Separating background activity from noise in EMG using a neuromuscular model

van der Ploeg, Bart (TU Delft Mechanical, Maritime and Materials Engineering; TU Delft Biomechanical Engineering)

Mugge, Winfred (mentor)
Rodriguez Hernandez, Karen (mentor)
De Groot, Jurriaan (mentor)
Schouten, Alfred (graduation committee)
van den Dobbelsteen, John (graduation committee)

Delft University of Technology

Biomedical Engineering | BioMechatronics

2019-10-07

To objectively diagnose the severity of spasticity, it is important to measure the muscle activity accurately. Filtered EMG has an offset above zero which is part noise and part background activity. The goal of this study is to separate these two parts, using a non-linear neuromuscular ankle model. The model uses positional and EMG data as inputs to make a prediction of the output torque around the ankle, optimizing several muscle parameters. It was assumed model error was partly caused by the lack of separation of background activity and noise in the EMG offset. The correlation between model error and the force-length relationship of the muscle was investigated, as it was hypothesized that the force-length relationship could be used to separate background activity from noise in EMG.Twelve subjects (11 male, 1 female, mean age 25.21.7 years) participated in the study, which was approved by the TU Delft Human Research Ethics Committee. The Achilles ankle perturbator and the TMSi Porti7 EMG system were used to obtain positional and EMG data around the ankle. The subjects performed several trials containing multiple ramp-and-hold phases. The trials were performed while relaxed, and while performing co-contraction at 5% and 10% of maximum voluntary co-contraction. Model validity was judged on robustness and fit. Model fit shows how close the predicted torque resembles the measured torque. High robustness is defined as a low variance of the model parameters per subject, between trials. To improve robustness, two model configurations, one using averaged data and one simplified model configuration, were compared with the original model. The most robust model configuration was used as a basis for an alternative model, which added a parameter for each EMG signal, that subtracted part of the EMG offset. This EMG offset subtraction parameter was added to remove the constant part of the noise in the EMG offset. The correlation between model error and the force-length relationship of the model was calculated for the model with the best robustness, and the alternative model.Model robustness was not improved in the averaged data model configuration or the simplified model configuration, compared to the original model. The original model was therefore used for the rest of the study. The EMG offset subtraction parameter did not yield to desired reduction of noise in the EMG offset. This was likely a result of difficulties with determining the EMG offset. Only weak correlations (R<<0.3) were found between the model error and the force-length relationship of the muscles, for both the original model and the alternative model configuration.

EMG
Neuromuscular modeling
Spasticity

http://resolver.tudelft.nl/uuid:8aeed468-b03d-459c-84fb-84a86570744f

Student theses

master thesis

© 2019 Bart van der Ploeg