A hybrid scheme for AEP based hearing deficiency diagnosis: CWT and convoluted k-nearest neighbour (CKNN) pipeline

Abstract

The auditory evoked potential (AEP) has been considered a standard clinical instrument for hearing and neurological evaluation. Although several approaches for learning EEG signal characteristics have been established earlier, the hybridization concept has rarely been explored to produce novel representations of AEP features and achieve further performance enhancement for AEP signals. Moreover, the classification of auditory attention within a concise time interval is still facing some challenges. To address this concern, this study has proposed a hybridization scheme, represented as a hybrid convoluted k-nearest neighbour (CKNN) algorithm, consisting of concatenating the convolutional layer of CNN with k-nearest neighbour (k-NN) classifier. The proposed architecture helps in improving the accuracy of KNN from 83.23% to 92.26% with a 3-second decision window. The effect of several concise decision windows is also investigated in this analysis. The proposed architecture is validated by a publicly benchmark AEP dataset, and the outcomes indicate that the CKNN significantly outperforms other state-of-the-art techniques with a concise decision window. The proposed framework shows superior performance in a concise decision window that can be effectively used for early hearing deficiency diagnosis. This paper also presents several discoveries that could be helpful to the neurological community.

Keywords

• Brain-computer interface (BCI)
• Auditory evoked potentials (AEP)
• Machine learning (ML)
• Deep learning (DL)
• Hybridization