Automatic classification of the temporomandibular joint sounds using scale and time-shift invariant representation of their time-frequency distributions

Abstract

Stethoscope auscultation can ascertain the presence of abnormal temporomandibular joint (TMJ) sounds, but auscultatory findings are inherently subjective and difficult to document for accurate comparison of diagnoses among different clinicians, or within the same patient over time. Time-frequency analysis of the TMJ sounds using reduced interference distributions (RIDs) indicate that several types of sounds can be correlated to TMJ pathology. The nearest neighbor (NN) and a previously introduced nearest linear combination (NLC) and nearest constrained linear combination (NCLC) pattern recognition methods are tested and mutually compared. Scale and time-shift invariant representations (STIRs) of the TMJ sound RIDs were used in order to perform their automatic classification. Results imply that linear combinations improved classification of the RIDs of TMJ sound.