Auditory Brainstem Response Classification Using Wavelet Transform and Multilayer Feed-forward Networks

The auditory brainstem response (ABR) has become a routine clinical tool for hearing and neurological assessments. In order to diagnose disorders and injuries affecting hearing threshold and functionality, the ABR signal peaks should be labeled by experts and signals should be classified in normal or abnormal categories. This work can be very complex and time consuming, especially in hearing screening of a population. In this study a method combining the wavelet analysis (dual-tree complex wavelet transform) and the multilayer perceptron network is introduced to classify a set of ABR signals into three different classes. The important features of the ABR are extracted by implementing dual-tree complex wavelet transform (DTCWT) to the signals. These extracted features are then used as input variables to a multilayer feedforward network with a hidden layer for the classification. The signals are classified in three major classes: normal Response (R), No Response (NR) and signals without a normal V peak (-V). 320 ABRs from 81 individual are applied in this study, about 70% of them are used for training stage and the remaining 30% used for testing and validating the algorithm.