Performance analysis of a framework for auditory steady-state response detection

Abstract

The Auditory Steady-State Response (ASSR) is a periodic electrical response from the brain which is generated by a healthy or impaired ear without conduction hearing loss subject. This response is evoked by presenting a subject to a periodically varying continuous acoustic signal, typically a sinusoidally modulated tone. The response consists of a waveform whose constituent discrete frequency components have the same periodicity as that of the acoustic signal and remain constant in amplitude and phase over an infinitely long time period. ASSR has been used to objectively assess hearing thresholds for hearing impairment diagnosis, in contrast to traditional subjective methods such as auditory brain-stem responses and audiometry. The objective of this study is to implement experimental setups that detect simulated ASSRs in single and multiple response recordings in order to asses detector performance in different approaches of signal recording parameters and signal processing techniques. The experimental setups were implemented using the MATLAB environment, in which three test scenarios were also developed: (i) a single channel ASSR detector in which the statistical detection tests F test, phase coherence (PC), magnitude-squared coherence (MSC) and Hotteling’s T 2 circular (T2) were used for comparative performance evaluation; (ii) a single channel ASSR detector in which the traditional sweep-by-sweep (SBS) averaging approach performance was compared to the proposed epoch-by-epoch (EBE) averaging aproach; and (iii) a multiple channel ASSR detector in which three Independent Component Analysis (ICA) algorithms – JADE, SOBI, and WASOBI – were applied for comparative performance evaluation from the reference method, in which ICA was not applied. The performance evaluation method was the Receiver Operating Characteristic (ROC) analysis, in which the Area Under the Curve (AUC) metric was used to estimate the detector’s accuracy levels for ASSR detection. From the results of the first test scenario it was concluded that the T2 and MSC tests presented the best overall performance, specially at lower SNR conditions. Results from scenario 2 indicated that the SBS approach resulted in higher accuracy levels after the transitory period of the AUC curve related to test time duration, while the EBE resulted in the steepest AUC curves for the first seconds of test time duration. From the results of scenario 3 it was not observed significant ASSR’s detection time reduction when ICA algorithms were applied in situations closely related to hearing threshold estimation.