An advanced machine learning approach for high accuracy automated diagnosis of otitis media with effusion in different age groups using 3D wideband acoustic immittance

•A two-stage ML approach is used to classify normal and OME ears in different age groups from the WAI data.•The first stage is a CNN model, and the second stage is a self-attention model focussing on key regions in the WAI.•The key regions are the areas with the most statistically significant differences between the normal and OME training data.•Data augmentation (e.g., interpolation and perturbation) is employed to generate synthetic (more training) data.•A high accuracy is achieved for classifying ears with otitis media with effusion. Wideband Acoustic Immittance (WAI) is a diagnostic tool for identifying middle ear dysfunction. The challenge to its widespread use is difficulty in interpreting the complex data. This study aimed to develop advanced Machine Learning (ML) tools to automatically diagnose ears with otitis media with effusion (OME) in different age groups from the WAI data. A total of 1177 sets of WAI data were collected from 551 normal middle ears and 626 ears with OME, divided into three age groups. A Titan IMP440 was used to measure wideband absorbance at frequencies from 226 to 8000 Hz, and pressure between +200 daPa and −300 daPa. A two-stage ML approach was used to achieve a highly accurate diagnosis of OME in each age group. In the first stage, a convolutional neural network (CNN) was developed to classify the WAI data set. In the second stage, another neural network with a self-attention mechanism was used to classify the most discriminative regions of the data. These regions were extracted areas that had the top 2.5 % most statistically significant difference between normal and OME ears in the training WAI data. Final classification considered outputs from the two stages. The two-stage ML approach achieved classification accuracy of 96.6 %, 94.1 %, and 90.7 % for the three age groups, respectively. The importance of this research is its contribution to the development of an automated diagnostic tool for OME. This tool will be easy to use, highly accurate, works across age groups and which will support clinicians in their diagnostic decisions.