*** All users be aware: UWSpace has been experiencing unusually long wait times during the depositing process. If you are a graduate student depositing a thesis, it is recommended that while the browser is loading that you do not try to close the connection. If you receive an error or a timeout message, please logout and then log back in. Please do not recreate and resend a new thesis deposit. In most cases, despite the error message, your deposit has successfully been sent to be reviewed. You can verify this by checking under the ‘deposits being reviewed' section of the deposits page. We apologize for the inconvenience. We are working hard to resolve this issue quickly. ***
Characterization of Neuromuscular Disorders Using Quantitative Electromyographic Techniques
MetadataShow full item record
This thesis presents a multifaceted effort to develop a system that allows electrodiagnostic clinicians to perform a quantitative analysis of needle detected electromyographic (EMG) signals for characterization of neuromuscular disorders. Currently, the most widely adopted practise for evaluation of patients with suspected neuromuscular disorders is based on qualitative visual and auditory assessment of EMG signals. The resulting characterizations from this qualitative assessment are criticized for being subjective and highly dependent on the skill and experience of the examiner. The proposed system can be decomposed functionally into three stages: (1) extracting relevant information from the EMG signals, (2) representing the extracted information in formats suitable for qualitative, semi-quantitative and quantitative assessment, and (3) supporting the clinical decision, i.e., characterizing the examined muscle by estimating the likelihood of it being affected by a specific category of neuromuscular disorders. The main contribution of the thesis to the extraction stage is the development of an automated decomposition algorithm specifically tailored for characterization of neuromuscular disorders. The algorithm focuses on identifying as many representative motor unit potential trains as possible in times comparable to the times needed to complete a qualitative assessment. The identified trains are shown to reliably capture important aspects of the motor unit potential morphology and morphological stability. With regards to the representation stage, the thesis proposes ten new quantitative EMG features that are shown to be discriminative among the different disease categories. Along with eight traditional features, the features can be grouped into subsets, where each subset reflects a different aspect of the underlying motor unit structure and/or function. A muscle characterization obtained using a feature set in which every relevant aspect is included using the most representative feature is more structured, simple, and generalizable. All the investigated features are clinically relevant. An examiner can easily validate their values by visual inspection; interpret them from an anatomical, physiological, and pathological basis; and is aware of their limitations and dependence on the acquisition setup. The second main contribution to the representation stage is the evaluation of the possibility of detecting neurogenic disorders using a newly proposed set of quantitative features describing the firing patterns of the identified motor units. The last contribution to the representation stage is the development of novel methods that allow an examiner to detect contributions from fibres close to the detection surface of a needle electrode and to track them across a motor unit potential train. The work in this thesis related to the decision support stage aims at improving existing methods for obtaining transparent muscle characterization. Transparent methods do not only estimate the likelihood of the muscle being affected by a specific disorder, but also induce a set of rules explaining the likelihood estimates. The results presented in this thesis show that remodelling the characterization problem using an appropriate binarization mapping can overcome the decrease in accuracy associated with quantizing features, which is used to induce transparency rules. To attain the above mentioned objectives, different signal processing and machine learning methods are utilized and extended. This includes spectral clustering, Savitzky-Golay filtering, dynamic time warping, support vector machines, classification based on event association rules and Gaussian mixture models. The performance of the proposed methods has been evaluated with four different sets of examined limb muscles (342 muscles in total). Also, it has been evaluated using simulated EMG signals calculated using physiologically and anatomically sound models. A system capable of achieving the aforementioned objectives is expected to promote further clinical adoption of quantitative electromyographic techniques. These techniques have potential advantages over existing qualitative assessments including resolving equivocal cases, formalizing communication and evaluating prognosis.
Cite this work
Meena AbdelMaseeh Adly Fouad (2016). Characterization of Neuromuscular Disorders Using Quantitative Electromyographic Techniques. UWSpace. http://hdl.handle.net/10012/10247