Sara M K Madsen1, James M Harte1,2, Claus Elberling, Torsten Dau1. 1. a Department of Electrical Engineering, Hearing Systems Group , Technical University of Denmark , Lyngby , Denmark and. 2. b Interacoustics A/S , Interacoustics Research Unit , Lyngby , Denmark.
Abstract
OBJECTIVE: The aims were to 1) establish which of the four algorithms for estimating residual noise level and signal-to-noise ratio (SNR) in auditory brainstem responses (ABRs) perform better in terms of post-average wave-V peak latency and amplitude errors and 2) determine whether SNR or noise floor is a better stop criterion where the outcome measure is peak latency or amplitude. DESIGN: The performance of the algorithms was evaluated by numerical simulations using an ABR template combined with electroencephalographic (EEG) recordings obtained without sound stimulus. The suitability of a fixed SNR versus a fixed noise floor stop criterion was assessed when variations in the wave-V waveform shape reflecting inter-subject variation was introduced. STUDY SAMPLE: Over 100 hours of raw EEG noise was recorded from 17 adult subjects, under different conditions (e.g. sleep or movement). RESULTS: ABR feature accuracy was similar for the four algorithms. However, it was shown that a fixed noise floor leads to higher ABR wave-V amplitude accuracy; conversely, a fixed SNR yields higher wave-V latency accuracy. CONCLUSION: Similar performance suggests the use of the less computationally complex algorithms. Different stop criteria are recommended if the ABR peak latency or the amplitude is the outcome measure of interest.
OBJECTIVE: The aims were to 1) establish which of the four algorithms for estimating residual noise level and signal-to-noise ratio (SNR) in auditory brainstem responses (ABRs) perform better in terms of post-average wave-V peak latency and amplitude errors and 2) determine whether SNR or noise floor is a better stop criterion where the outcome measure is peak latency or amplitude. DESIGN: The performance of the algorithms was evaluated by numerical simulations using an ABR template combined with electroencephalographic (EEG) recordings obtained without sound stimulus. The suitability of a fixed SNR versus a fixed noise floor stop criterion was assessed when variations in the wave-V waveform shape reflecting inter-subject variation was introduced. STUDY SAMPLE: Over 100 hours of raw EEG noise was recorded from 17 adult subjects, under different conditions (e.g. sleep or movement). RESULTS: ABR feature accuracy was similar for the four algorithms. However, it was shown that a fixed noise floor leads to higher ABR wave-V amplitude accuracy; conversely, a fixed SNR yields higher wave-V latency accuracy. CONCLUSION: Similar performance suggests the use of the less computationally complex algorithms. Different stop criteria are recommended if the ABR peak latency or the amplitude is the outcome measure of interest.
Keywords:
ABR latency and amplitude errors; Auditory brainstem response, signal quality estimators; post-averaged noise estimators