Diagnostic auditory brainstem response analysis : evaluation of signal-to-noise ratio criteria using signal detection theory

This study evaluated an online measure of signal-to-noise ratio (SNR) as a response-detection tool for threshold auditory brainstem response (ABR) testing. Threshold-ABR data were analysed for 98 infants and young children tested at BC Children's Hospital, and results were validated on an additional 10 patients. Using signal detection theory, it was possible to assess test performance for the SNR measure, with expert-clinician judgement as the gold standard. In addition, a range of SNR criteria were assessed in terms of sensitivity (the ability to accurately identify a response) and specificity (the ability to accurately reject waveforms that do not contain a response). The effect of residual noise (RN) exclusion criteria on SNR test performance, sensitivity, and specificity was also investigated. Waveforms to 500-, 2000-, and 4000-Hz air-conducted brief-tone stimuli were included in this study. Overall, SNR was found to have a test performance of A=.91, with improved performance (A=.93) when high residual-noise waveforms (RN>0.08 μV) were excluded. When low-RN data were separated by frequency, test performance for each frequency was A=.94. Results suggest that the optimal SNR criterion is slightly lower for 500-Hz recordings than for 2000- or 4000-Hz recordings. However, when high-RN recordings were excluded, a SNR criterion of 0.98 achieved a minimum specificity of 95% for each stimulus frequency, with sensitivity values ranging from 64%(for 500 Hz) to 79% (for 4000 Hz). Findings confirm the hypotheses that SNR accurately distinguishes response-present from response-absent waveform, and that quiet recordings are more easily interpreted than noisy recordings using SNR. Guidelines are provided for the clinical use of SNR as an objective response-detection tool.Medicine, Faculty ofAudiology and Speech Sciences, School ofGraduat