الفهرس | Only 14 pages are availabe for public view |
Abstract The present work is an attempt to apply modern signal processing and pattern recognition techniques to the electromyographic signal recorded from the masseter muscle to determine if there are significant differences in the masseter muscle activity before and after trial denture base insertion. Five techniques were utilized for feature extraction. These include: features extracted from isolation of MUAP?s patterns, power spectral analysis, autoregressive (AR) modeling, discrete wavelet transform (DWT), and wavelet packet transform (WP). An attempt also was made to reduce the feature dimensionality using the principal component analysis (PCA). Two methods are utilized to detect if there were significant differences in the masseter muscle activity before and after wearing trial denture base. These are statistical analysis and pattern recognition techniques. Statistical analysis involved parametric and nonparametric tests were performed on sets of features extracted for each group. Parametric tests including the student?s ttest, Ftest, and Least Significant Difference (LSD) test have been applied to the five feature vectors. Results of the student?s ttest at 5% level of significance and Least Significant Difference (LSD) test at 5% and 1% levels of significance did not prove any significant difference between each feature element before and after trial denture base insertion. Using the Ftest at 5% level of significance, some feature elements showed significant differences before and after wearing trial denture bases. The nonparametric KolmogorovSmirnov (KS) two sample test has been applied for two cases; small sample size and large sample size. The test for small sample size did not prove any significant difference between the feature distributions before and after patients wearing trail denture bases. However, in the case of large sample size, the results have shown significant differences in the distributions of EMG signals before and after inserting trial denture bases. Neural network classifiers have been utilized using a feed forward multilayer neural network. The designed network has been trained and tested using the ?leave oneout? method. The results of classification have shown significant differences between feature vectors before and after inserting trial denture base. The highest classification has been obtained when using WP with PCA. |