Classification of temporomandibular joint sounds using hidden markov models

The Temporomandibular Joint (TMJ) plays a critical role in essential motor functions such as chewing, speaking, and swallowing. TMJ disorders negatively affect quality of life, manifesting as pain, muscle spasms, and restricted movement. Traditional diagnostic methods largely rely on subjective assessments, leading to uncertainties in diagnosis. This study proposes a classification method based on acoustic signal analysis and supported by Hidden Markov Models (HMM). Acoustic data were collected using a custom-designed device and preprocessed to remove noise. Various band-pass digital filters were applied, and features including energy, RMS, amplitude, dominant frequency, and Mel-Frequency Cepstral Coefficients (MFCC) were extracted. Principal Component Analysis (PCA) based dimensionality reduction was employed to reduce feature space complexity. Different feature combinations and data selection methods were systematically evaluated to assess their impact on classification performance. The extracted features were used as input to the HMM, and model training and testing were performed using the Viterbi and Baum-Welch algorithms. Results demonstrated that feature extraction significantly improved classification accuracy, with MFCC and RMS features providing particularly high performance. These findings highlight the effectiveness of acoustic analysis for TMJ disorder diagnosis and suggest that the proposed method can serve as a non-invasive, rapid, and reliable supportive diagnostic tool in clinical practice.