| Abstract #674, Date 2/16/99, Session R6, Poster (B174) |
| Optimized methods for automated detection of auditory brainstem responses in humans |
| *S.T. Neely, M.P. Gorga, P.A. Dorn (Boys Town National Research Hospital) |
Auditory brainstem responses (ABR) were recorded from 50 normal hearing human subjects. Sufficent waveform information was saved during the recording to allow post hoc evaluation of ABR detection statistics. Each of these detection statistics is essentially a signal-to-noise variance ratio; however, details of the signal and noise variance computations differed. One of these statistics, called Fsp (Elberling and Don, 1985, Scand. Audiol. 14, 89-96), uses a "single-point" variance to estimate the noise, has been used successfully in previous studies for ABR detection, and provides a baseline for comparison in this study. The ABR data from 25 subjects were used to optimize parameters (such as time window and weighting method), while the other 25 subjects were used to evaluate detection performance. Test performance was evaluated for the task of detecting the ABR when the stimulus was a 100 ms, 60 dB pSPL click, repeated 4000 times. Compared with the Fsp, small, but consistent, improvements in test performance were obtained by (1) using between-block variance (instead of within block variance) to estimate the noise (2) using the square root of the within-block variance (instead of just the variance) to determine weights for the weighted average of the signal waveform. Alternate techniques for computing the signal variance, which involve cross-correlation with either a replicated waveform or a template waveform, did not consistently improve test performance. Detection performance obtained by these automated methods was similar to that obtained by visual detection of the ABR waveform. |