Abstract:
Electromyography (EMG) signal analysis plays a major role in the diagnosis of
neuromuscular diseases, such as amyotrophic lateral sclerosis (ALS) and myopathy.
With the recorded EMG data generally, either direct analysis is performed or rst
the motor unit action potentials (MUAPs) are extracted and then MUAP based
analysis is carried out. In this thesis, two schemes based on these two approaches
are proposed for neuromuscular disease classi cation. In the direct frame by frame
analysis of EMG data, unlike the conventional methods, instead of considering only
the local information obtained from a single frame of EMG recording, we propose to
utilize global statistics which is obtained based on information collected from some
consecutive frames. Di erent time and frequency domain features are investigated.
A discrete wavelet transform (DWT) based feature extraction scheme is developed,
where a few high energy DWT coe cients alongwith the maximum value are used,
which drastically reduces the feature dimension. With the objective of reducing
computational complexity, a MUAP based classi cation scheme is then presented.
In the proposed MUAP based classi cation scheme, rst MUAPs are extracted from
EMG data via EMG decomposition using a template matching scheme. It is well
known that not all MUAPs obtained via decomposition are capable of uniquely
representing a class. Thus, an energy content based dominant MUAP selection
criterion is proposed and only the dominant MUAP is used for the classi cation.
Conventional morphological features of dominant MUAPs are investigated. A fea-
ture extraction scheme based on some statistical properties of the DWT coe cients
of dominant MUAPs is proposed. Moreover some spectral domain features based on
discrete cosine transform are also introduced. For the purpose of classi cation, the
K-nearest neighborhood (KNN) classi er is employed in supervised classi cation. In
order to investigate the performance of the proposed methods, a publicly available
clinical EMG database is used. The leave-one-out cross validation technique is used
in order to verify the performance in classifying a test data among three classes, nor-
mal, ALS and myopathy. It is found that the proposed schemes provide extremely
satisfactory results in comparison to that obtained by some of the existing methods
in terms of speci city, sensitivity, and overall classi cation accuracy.
