Real-time Hearing Threshold Determination of Auditory Brainstem Responses by Cross-correlation Analysis

Abstract

<jats:p>Auditory brainstem response (ABR) is widely employed to evaluate the hearing function, both in clinics and basic research. Despite many attempts for automation over decades, reliable determination of threshold stimulus level still relies on human visual identification of waveform, which oftentimes is subjective. Here, we report a robust procedure for automatic and accurate threshold determination in both mouse and human ABR. Contrary to prior approaches, in our new threshold determination algorithm, the on-going averaging is stopped once the waveform is confirmed by a cross-correlation time shift approach. The flexible ending sweep numbers for different stimuli is used to inform the threshold determination. We found a good match of the threshold readings between the algorithm and the human judges. Moreover, in the algorithm, smaller sweep number is required for strong response from supra-threshold level, and thus a considerable portion of sweeps can be saved in comparison to the case with level averaging of a fix number. These features are attractive and implementation of this method in commercial devices will make the ABR test procedure more objective and efficient.</jats:p>

Objective hearing threshold identification from auditory brainstem response measurements using supervised and self-supervised approaches

“…Over the years, ABR has been discussed in literature as Auditory Evoked Potentials (AEP) [ 12 ], Cortical Auditory Evoked Potentials (CAEP) [ 13 ], Brainstem auditory evoked potential (BAEP) [ 14 , 15 ], Brainstem Evoked Response Audiometry (BERA) [ 16 ], and Auditory Evoked Potential (EAP) [ 17 ] . Many approaches applied and combined methods from different fields of statistics [ 9 , 13 , 17 – 30 ], often involving feature extraction from the time and/or the frequency domain. Some approaches also involved bootstrapping [ 31 ], comparison to templates [ 15 , 23 ], or deep learning [ 12 , 14 , 32 – 34 ].…”

“…While most of the published methods for automated threshold identification use averaged response data, a recently published method [ 30 ] processes individual sweep responses with good results. Unfortunately, although always generated during ABR, individual sweep response time curves are not always easily accessible.…”

“…and Auditory Evoked Potential (EAP) [17]. Many approaches applied and combined methods from different fields of statistics [9,13,[17][18][19][20][21][22][23][24][25][26][27][28][29][30], often involving feature extraction from the time and/or the frequency domain. Some approaches also involved bootstrapping [31], comparison to templates [15,23], or deep learning [12,14,[32][33][34].…”