Automated wave labelling of the auditory brainstem response using machine learning.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Auditory Brainstem Response (ABR) analysis is crucial for diagnosing auditory pathway disorders.
• Accurate labeling of ABR waveform peaks (I, III, V) is essential for reliable interpretation.
• Current ABR interpretation can be subjective and time-consuming.

Purpose of the Study:

• To compare the performance of various machine learning algorithms for labeling key ABR waveform peaks.
• To develop an algorithm that provides confidence measures for ABR wave latency estimates.

Main Methods:

• Secondary analysis of a published ABR dataset comprising 482 suprathreshold waveforms from 81 participants.
• Comparison of five machine learning algorithms using a nested k-fold cross-validation procedure.
• Training an additional algorithm to estimate confidence in ABR wave latency.

Main Results:

• A convolutional recurrent neural network (CRNN) demonstrated superior performance over other evaluated algorithms.
• The CRNN achieved 95.9% accuracy in labeling ABR waves within ±0.1 ms of the target.
• Mean absolute error for wave latency estimation was 0.025 ms, with high confidence linked to greater accuracy.

Conclusions:

• Machine learning, particularly CRNNs, shows significant potential to aid clinicians in ABR interpretation.
• The developed algorithm offers promising results for automated ABR analysis.
• Further research involving large, diverse datasets and clinical validation is necessary for real-world application.