A facial pattern recognition approach for detection of temporomandibular disorder

The aim of this study is to automatically classify individuals with temporomandibular disorder and healthy subjects. The process of automated classification requires measurement of features that can be used to distinguish between different classes. We used maximum Lyapunov exponents to measure the changes in the dynamics of the chewing pattern, the number of peaks in the normalized highpass filtered data to find the abnormalities in both opening and closing of mouth, normalized skewness and kurtosis to measure the distribution profile of the data samples, likelihood information to quantify the probability of the click events in either opening or closing process, and peak amplitude to measure how severe the abnormality is. Finally, using the above features together with Support vector machine to classify all subjects as belonging to individuals with TMD or not. The early experiments show encouraging results.