Publication Date

2008-04-10

Availability

Open access

Degree Type

Dissertation

Degree Name

Doctor of Philosophy (PHD)

Department

Biomedical Engineering (Engineering)

Date of Defense

2008-04-04

First Committee Member

Dejan Tepavac - Committee Co-Chair

Second Committee Member

Christine Thomas - Committee Co-Chair

Third Committee Member

Ozcan Ozdamar - Committee Member

Fourth Committee Member

Jorge Bohorquez - Committee Member

Fifth Committee Member

Weizhao Zhao - Committee Member

Abstract

Involuntary electromyographic (EMG) activity has been recorded in the thenar (thumb) muscles of spinal cord injured (SCI) subjects for only short time periods (minutes), but it is unknown if this motor unit activity is ongoing. Longer duration EMG recordings can investigate the physiological significance of this neuromuscular activity. Analysis of these data is complex and time consuming. Since no software is currently capable of classifying 24 hours of data at a single motor unit level, the goal of this research was to devise an algorithm to automatically classify motor unit potentials over 24-hours. Twenty-four-hour, 2-channel thenar muscle EMG recordings were obtained from four different SCI subjects with cervical level injuries using a data logging device with custom software. The automatic motor unit classification algorithm used to classify the 24-hour recordings was a procedure consisting of four stages that included segmentation, clustering, and motor unit template uniting. All individual potentials were then classified and any superimposed potentials were resolved into their constituent classes. Finally, the algorithm found the firing patterns for each of the stable motor unit classes. The classification algorithm performance was compared to the analysis of a human operator and assessed in 2 ways: Tracking global classes over the 24 hours and correctly classifying individual motor unit potentials as to belonging to particular global classes. The algorithm was able to track an average of 13 global classes in four 24-hour recordings with a mean accuracy of 92 %. It was also able to classify individual potentials with a mean accuracy of 86% over four recordings, greater than the inter-rater reliability of two human operators (79%). The activities of the motor units tracked by the algorithm ranged from tonic firing to sporadic activity. The algorithm could analyze 24 hours of data in 2-3 weeks, while a human operator was estimated to take more than 2 years. In conclusion, the motor unit classification algorithm accomplished its goal of automatically tracking motor unit classes over a 24-hour recording with high accuracy. The 24-hour classification method developed here could be applied towards classifying long term recordings of other biological signals.

Keywords

EMG; Classification; Long Term; 24-hour; Spinal Cord Injury

Share

COinS