Characterization of sounds emanating from the human temporomandibular joints

Sounds from the temporomandibular joint were recorded on audiotape from 238 individuals by placing microphones in both ears. The recordings were later digitized at a sample rate of 1.7 kHz with 10-bit resolution and stored on computer disk. At least two open-close cycles were assessed from each individual; 2707 different individual sounds were analysed in the time and frequency domains. The sounds were classified as: (a) single, short duration (clicks), (b) multiple, short-duration (creaks) and (c) long duration (crepitus). The sounds were further subclassified into either high or low amplitude by (i) the attack, which produced hard and soft categories and (ii) comparing the amplitude between sides-bilateral sounds were those with amplitudes differing by < 40 mV; the rest were unilateral. To establish the robustness of the classification 42 acoustic events were selected to be classified visually by three observers on two separate occasions. Intraobserver agreement was 82% (kappa = 0.75) while interobserver agreement was 60% (kappa = 0.71). Statistically significant differences were noted between all classifications of sound. These were most marked in the time domain. A simple, automated classification scheme was devised that was capable of categorizing the sounds with 82% agreement (kappa = 0.71) compared to a human observer.