New Methods for the Acoustic-Signal Segmentation of the Temporomandibular Joint

Abstract

(1) Background: The stethoscope is one of the main accessory tools in the diagnosis of temporomandibular joint disorders (TMD). However, the clinical auscultation of the masticatory system still lacks computer-aided support, which would decrease the time needed for each diagnosis. This can be achieved with digital signal processing and classification algorithms. The segmentation of acoustic signals is usually the first step in many sound processing methodologies. We postulate that it is possible to implement the automatic segmentation of the acoustic signals of the temporomandibular joint (TMJ), which can contribute to the development of advanced TMD classification algorithms. (2) Methods: In this paper, we compare two different methods for the segmentation of TMJ sounds which are used in diagnosis of the masticatory system. The first method is based solely on digital signal processing (DSP) and includes filtering and envelope calculation. The second method takes advantage of a deep learning approach established on a U-Net neural network, combined with long short-term memory (LSTM) architecture. (3) Results: Both developed methods were validated against our own TMJ sound database created from the signals recorded with an electronic stethoscope during a clinical diagnostic trail of TMJ. The Dice score of the DSP method was 0.86 and the sensitivity was 0.91; for the deep learning approach, Dice score was 0.85 and there was a sensitivity of 0.98. (4) Conclusions: The presented results indicate that with the use of signal processing and deep learning, it is possible to automatically segment the TMJ sounds into sections of diagnostic value. Such methods can provide representative data for the development of TMD classification algorithms.

Keywords: auscultation; deep learning; segmentation; signal processing; temporomandibular joints.

Figure 1

Generic pipeline for heart sounds classification [7].

Figure 2

Diagram of the developed method of TMJ acoustic signal segmentation. The cut-off frequency of the low-pass filter is automatically calculated for each signal and is equal to its main frequency.

Figure 3

Segmentation result obtained for an example left temporomandibular joint (TMJ) acoustic signal recorded from a patient complaining about pain in left TMJ and diagnosed with left joint arthralgia. The red dashed line corresponds to the beginning and end of every segment.

Figure 4

Segmentation result obtained for an example right temporomandibular joint (TMJ) acoustic signal recorded from healthy patient. The red dashed line corresponds to the beginning and end of every segment.

Figure 5

A diagram of the preprocessing step of our model. One can notice that convolutions and pooling operations generate a set of signal representations because each single convolutional layer consists of multiple filters which results in multiple output. Signal representations are shorter than the signal itself because of the pooling layers which discard some processed samples.

Figure 6

The above image represents a pooling block, which consists of the max pooling layer and the designed min-abs-pooling layer with concatenation of their outputs.

Figure 7

Collection of plots representing output examples from the intermediate steps of our model. (a) the original signal (input); (b) the signal representation (an arbitrarily chosen set of features) after a set of convolutions and decimation operations; (c) output from a set of recurrent layers; (d) interpolated output from LSTMs shown in (c). The output from (d) is then classified if a particular sample belongs to a class which is presented in (e); (f) original signal together with reference mask and the model’s output mask.

Figure 8

A plot representing the training process (loss function depending on the number of epochs).