Objective auditory brainstem response classification using machine learning

Abstract

OBJECTIVE: To use machine learning in the form of a deep neural network to objectively classify paired auditory brainstem response waveforms into either: 'clear response', 'inconclusive' or 'response absent'. DESIGN: A deep convolutional neural network was trained and fine-tuned using stratified 10-fold cross-validation on 190 paired ABR waveforms. The final model was evaluated on a test set of 42 paired waveforms. STUDY SAMPLE: The full dataset comprised 232 paired ABR waveforms recorded from eight normal-hearing individuals. The dataset was obtained from the PhysioBank database. The paired waveforms were independently labelled by two audiological scientists in order to train and evaluate the network's performance. RESULTS: The trained neural network was able to classify paired ABR waveforms with 92.9% accuracy. The sensitivity and specificity were 92.9% and 96.4% respectively. CONCLUSIONS: This neural network may have clinical utility in assisting clinicians with waveform classification for the purpose of hearing threshold estimation. Further evaluation on a large clinically-obtained dataset would provide further validation with regards to the clinical potential of the neural network in diagnostic adult testing, newborn testing and in automated newborn hearing screening.

Automatic Recognition of Auditory Brainstem Response Characteristic Waveform Based on Bidirectional Long Short-Term Memory

“…The reduction in recording time was expected to promote the application of this measurement technique in clinical practice. McKearney and MacKinnon (19) divided ABR data into clear response, uncertain, or no response. In their work, they constructed a deep convolutional neural network and fine-tuned it to realize ABR classification.…”

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

“…First, it provides minimal quality control for unambiguous waveform recognition for both humans and algorithms. Such standardized data will benefit machine-learning-based approaches by minimizing annotation discrepancy in the training data (McKearney and MacKinnon, 2019). Second, when to stop averaging is an important decision during ABR recording (Don and Elberling, 1996; Madsen et al, 2018), the new method makes the ABR test more efficient by avoiding prolonged acquisition and redundant recordings.…”

“…As precise and objective measurement of small hearing threshold elevation became critical for diagnosis of progressive hearing loss (Barreira-Nielsen et al, 2016), hidden hearing loss (Kujawa and Liberman, 2009; Mehraei et al, 2016; Ridley et al, 2018; Sergeyenko et al, 2013), age-related hearing loss (Gates and Mills, 2005; Sergeyenko et al, 2013) and tinnitus (Bramhall et al, 2018; Castaneda et al, 2019), automated approaches with high precision and reliability are in demand to objectify the ABR threshold determination. Over decades, many attempts were made including: (1) quantification of the waveform similarity by comparison to existing templates (Davey et al, 2007; Elberling, 1979; Valderrama et al, 2014) as well as based on features learned by artificial neural network (Alpsan and Ozdamar, 1991; McKearney and MacKinnon, 2019) from human annotated datasets; (2) quantification of the waveform stability by cross-correlation function between single-sweeps (Bershad and Rockmore, 1974; Weber and Fletcher, 1980), interleaved responses (Berninger et al, 2014; Xu et al, 1995) or responses at adjacent stimulus levels (Suthakar and Liberman, 2019); (3) the ‘signal quality’ through scoring procedures like F-ratios (Cebulla et al, 2000; Don and Elberling, 1994; Elberling and Don, 1984; Sininger, 1993). Due to inconsistencies in waveform and signal-to-noise-ratio (SNR) introduced by differences in test subject conditions, electrode placement and impedance, as well as acquisition settings, the accurate threshold determination is only possible under a narrow range of experimental settings, hampering direct comparisons of ABR data and results across laboratories.…”

Contributions and limitations of using machine learning to predict noise-induced hearing loss

“…Machine learning (ML) has been widely applied to automatically identify inter-correlations between data that would normally require a great deal of manpower and be difficult to define manually (McKearney and MacKinnon 2019 ). The application of ML to the field of Audiology has shown promise, because of its effectiveness in analyzing non-linear relationships between data such as predicting hearing thresholds of patients who are exposed to specific risk factors (Chang et al 2019 ).…”