Analysing wideband absorbance immittance in normal and ears with otitis media with effusion using machine learning

Abstract

Wideband Absorbance Immittance (WAI) has been available for more than a decade, however its clinical use still faces the challenges of limited understanding and poor interpretation of WAI results. This study aimed to develop Machine Learning (ML) tools to identify the WAI absorbance characteristics across different frequency-pressure regions in the normal middle ear and ears with otitis media with effusion (OME) to enable diagnosis of middle ear conditions automatically. Data analysis included pre-processing of the WAI data, statistical analysis and classification model development, and key regions extraction from the 2D frequency-pressure WAI images. The experimental results show that ML tools appear to hold great potential for the automated diagnosis of middle ear diseases from WAI data. The identified key regions in the WAI provide guidance to practitioners to better understand and interpret WAI data and offer the prospect of quick and accurate diagnostic decisions.

A Deep Learning Approach for Detecting Otitis Media From Wideband Tympanometry Measurements

“…The method proposed by Merchant et al [15] has the lowest performance, and is also the simplest method, as it is based on principal component analysis and logistic regression. The performance of the 2D CNN for WBT classification proposed by Grais et al [20] is comparable to our performance, but still lower. The proposed CNN architecture is simpler than ours, as they employ fewer layers (both convolutional and fully connected layers) and larger convolution kernels in each layer.…”

“…Their analysis used a random forest classifier on selected features (peak admittance, peak pressure, width of the tympanogram, and ear canal volume) from a standard 226 Hz tympanogram, which was combined using majority voting with the output of a convolutional neural network predicting diagnosis based on the otoscopy image of the patient. Grais et al [20] employed several machine learning methods to analyze the WBT measurements, and found the convolutional neural network to be the best performing approach. They also used a random forest model to produce class activation maps that were used to interpret the diagnostic decision.…”

“…Besides our proposed methods, we have also run experiments with the methods proposed by Merchant et al [15] and Grais et al [20] for comparison. These methods were trained and tested using our dataset to ensure a proper comparison.…”

“…We trained the binary classification model to predict OM or NOE, and tested it on our test dataset. Grais et al [20] compared several machine learning methods for the classification of OM and NOE based on WBT measurements. They show that the CNN is superior to a fully connected neural network, random forest model, support vector machine, and a k-NN.…”

“…It has not previously been shown or demonstrated that it is possible use WBT to distinguish the two types of otitis media. Other studies such as [8], [10], [20] only include OME cases, and not AOM. Helenius et al [32] investigated discrimination of diagnosis based on standard 226 Hz tympanometry, and found that this measurement can be used to distinguish between NOE and OM cases, but not to diagnose specific types of OM.…”