Date on Master's Thesis/Doctoral Dissertation

6-2012

Document Type

Master's Thesis

Degree Name

M. Eng.

Department

Bioengineering

Committee Chair

El-Baz, Ayman Sabry

Author's Keywords

EMG; Local wavelet filtering; De-noising; ECG

Subject

Electromyography; Spinal cord--Wounds and injuries

Abstract

Introduction: Approximately 12,000 new cases of spinal cord injury (SCI) are reported each year in the US. Currently, the most widely used method of assessing the recovery of voluntary capability after spinal cord injury is the American Spinal Injury Association Impairment Scale (AIS). However, this test is not objective and is not sensitive enough to detect small activities. Recent studies have been using surface electromyography (EMG) to develop objective and sensitive spinal cord injury (SCI) characterization protocols. EMG recordings from the trunk muscles are contaminated with the electrical activity of the heart (ECG, electrocardiography). Depending on the level, spinal cord injury may disrupt the control of the trunk muscles, and EMG recordings from these muscles will be very weak compared to those in non-injured individuals. The elimination of ECG artifacts play critical role in precise evaluation of the trunk muscles in these individuals. While common global digital filters may generally remove some of the ECG corruption from the signal, these filters also remove or alter valuable EMG signal, which makes the physiological importance of these signals irrelevant. Methods: Local filtering approach was developed to remove this ECG noise, without significantly altering the EMG signal. The local filtering approach uses externally recorded ECG signals, in a separate lead configuration, as a mask to locate the area of ECG spikes within the noisy EMG signal. The areas of the signal containing the ECG noise are decomposed into 128 sub-wavelets using custom-scaled Morlet Wavelet Transform. Sub-wavelets pertaining to ECG within the signal at the ECG spike location are then removed, and the signal is reconstructed to create a clean EMG signal. This process is analytically tested for robustness and accuracy, using customized validation metrics, on simulated phantom signals. It is compared with a global Morlet Wavelet filter that does not localize its filtering process on the ECG spikes. Results: It was found that the localized filtering significantly reduced the Root-Mean-Squared (RMS) of the area of the signal containing ECG spikes. The Localized Filter also significantly reduced the error produced from removal of EMG signal in the areas outside of ECG spikes compared to global filter. The proposed local filter doesn’t degrade the signal, even at low ECG amplitudes (~60% improvement), compared to the global filter, which destroys the signal at this low amplitude ECG (-100% improvement). The proposed local filter is also far more efficient at removing larger amplitude ECG (more critical) than the global filter, which has a narrow range of signals that it can efficiently remove ECG. Hence, the proposed local filter is more robust and clinical-ready than the global filter. Conclusion: Proposed approach is far superior in terms of ECG removal accuracy, and introduction of artifact error from processing, compared to comparable global filter. It provides a mean to improve analysis of EMG signals as a tool to assess recovery from SCI.

Download

Share

COinS