Determination of temporomandibular joint disorder by using signal processing and artificial intelligence techniques

Tempromandibular Joint (TMJ) is the joint between lower jaw bone (mandible) condyle and mandibular fossa of temporal bone. TMJ has ability to perform complex movements. Any health problem arising from TMJ is called Temporomandibular Joint Disorder (TMD). TMD is a frequent health problem and %75 of the population may suffer from some degree of TMD. One of the classical diagnose method of TMD is listening TMJ sound during the clinical examination of the patient by the dentist using a stethescope. TMD sound are grouped into three main categories known as crepitation, clicking and popping. In this study, a method is developed to classify the TMJ sounds as healthy and not-healthy by using artificial intelligience techniques. First a non-invasive device is designed and TMJ sounds of healthy and non-healthy people are recorded. In first phase, to remove noise and insignificant parts digital signal processing is applied to sound data and then 100 frequency based features are extracted from each data set. Data is classified by an Artificial Neural Network (ANN) and a success rate of around 78% is obtained. In second phase, statistical features are extracted from sound data set. Extracted features are used to classify the sound data set by means of ANN. A success rate of around %87 to 89 is obtained. In the third phase deep learning methods are used for classification. A band pass filter is applied to chosen data and insignificant parts of the higher frequency parts are removed by decreasing sampling rate. Known success of deep learning methods classifying picture data direct the transforming of sound data to spectrogram picture data by using Short Time Fourier Transform. Deep learning algorithms are applied to picture data. Parameter and network structure adjustments are made to increase the network consistency. The success rate of deep learning algorithm is regularly increased over 90%. If the results are compared, it is observed that classification method based on deep learning is more successful then previous two methods.