University of British Columbia
Auditory brainstem response (ABR) detection supported by machine learning
Abstract
dc:descriptionThe auditory brainstem response (ABR) is clinically used to estimate thresholds and hearing status in infants and complex populations. In the current study, I aimed to determine the possibility that a machine learning (ML) model (CNN autoencoder) could be trained on individual participants’ data and used to detect their ABRs. The CNN autoencoder was trained on individual participants’ data to produce an objective statistic, which was used to detect ABRs from a simulated dataset (n=30) and a real human adult dataset (n=30). For the simulated data, I found the CNN autoencoder performed better than human detectors and conventional statistical detection methods (Fsp, Fmp, SDR, CCR, and Hotelling’s T2). For the real dataset, agreement in response detection was greatest among human-to-human detections (Cohen’s κ = 0.86–0.88) but low-to-moderate among human-to-CNN detections (Cohen’s κ = 0.23-0.41) and human-to-statistical detections (Cohen’s κ = 0.06–0.56). Together, these findings indicated that a personally-trained CNN autoencoder’s ABR detection performance was comparable to conventional statistical and human detectors. The CNN autoencoder appears to be a promising tool for further investigation in clinical settings.
Degree
thesis:*- Name thesis:degree_name
- Master of Science - MSc
- Level thesis:degree_level
- master's
- Discipline thesis:degree_discipline
- Audiology and Speech Sciences
- Grantor dc:publisher
- University of British Columbia
- Year dc:date
- 2026
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- McMillan, Kailey
- Contributors dc:contributor
-
- Herdman, Anthony
Rights
dc:rights- Statement dc:rights
-
- Attribution-NonCommercial-NoDerivatives 4.0 International
- Language dc:language
- eng
Identifiers
dc:identifier.*- Handle dc:identifier
- http://hdl.handle.net/2429/95647