Automatic classification of audiological expert knowledge summarized by common audiological functional parameters (CAFPAs)
* Presenting author
In clinics, the reliability of audiological diagnostics depends on the expertise of the examining ENT specialists. Particularly less experienced physicians would benefit from a data-driven assistance system. Thus, we developed an automatic classification system that uses audiological data for supporting indication and rehabilitation finding. The data in this study were simulated from experts’ knowledge, collected by a survey. Experts designated the range of patients’ responses for predefined audiological measurements for 14 diagnostic cases and treatments. Furthermore, the experts were asked to define values for the common audiological functional parameters (CAFPAs) for these cases. CAFPAs serve as generalization, i.e., data reduction, of audiological measurements characterizing the hearing capabilities of patients, i.e., thresholds in quiet, supra-thresholds, binaural hearing, neural and cognitive components and the socio-economic status.To generate individual patient data from these expert data, the indicated ranges were treated as independent probability distributions of the measurements. Hence it was possible to simulate patient data by randomly drawing data points from these distributions. The data from either pure measurements or the CAFPAs were fed to machine learning algorithms for classification of audiological indications and rehabilitations. The results are compared and show that CAFPAs comprise a suitable reduction of data without information loss.