Diagnosis of Hearing Disorders and Screening using Artificial Neural Networks based on Distortion Product Otoacoustic Emissions

Abstract

The auditory system produces low level sound in response to stimulation. Distortion Product Otoacoustic Emissions (DPOAEs) are a series of combination tones generated in the ear when stimulated with two pure tones of frequencies f1 and f2. The most prominent DPOAE is the cubic difference tone at 2f1-f2. The objective of this work is to determine the DPOAE response characteristics from normal and hearing impaired subjects and hearing screening using artificial neural networks (ANN). The data acquisition platform controls the microphone/speaker system which generates the acoustical stimuli for presentation to the ear canal and collects the acoustic responses recorded in the ear canal. Digital signal processing technique involves estimation of weak DPOAE response at the 2f1-f2 frequency, the noise floor and the DPOAE signal to noise ratio (SNR) with f2 at frequencies 1.5, 2, 3, 4, 5 and 6 KHz. The individual pure tone audiometric data are also collected to differentiate ears with normal hearing from impaired ear. DPOAE features and pure tone audiometric data are used to train the ANN classifier. The proposed method is developed and evaluated using a dataset based on 76 hearing subjects. Supervisory learning is used for the training of the classifier. Each hearing subject is classified as normal or abnormal one. The ANN system indicates that the results are comparable to human intervention, with substantial reduction of time and resources. The DPOAE measurement provides detailed frequency specific information on cochlear function; it is a potential tool to find hearing impairment and hearing screening.