Auditory Brainstem Response Detection Using Machine Learning: A Comparison With Statistical Detection Methods.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Auditory Brainstem Response (ABR) detection from EEG is crucial for hearing assessment.
• Current statistical methods for ABR detection can be limited in accuracy and efficiency.
• Machine learning offers potential for improved automated analysis of neurophysiological signals.

Purpose of the Study:

• To train and evaluate machine learning algorithms for accurate ABR detection in EEG data.
• To compare the performance of the best machine learning model against established statistical detection methods.
• To identify suitable machine learning approaches for ABR analysis.

Main Methods:

• Four machine learning algorithms (Random Forest, CNN-LSTM, Stacked Ensemble, MLP) were trained using nested k-fold cross-validation.
• Simulated EEG data based on recorded ABRs and no-stimulus data were used for model training and testing.
• The best-performing model was compared to conventional statistical methods (Fsp, Fmp, q-sample, Hotelling's T2).

Main Results:

• The stacked ensemble model demonstrated significantly higher sensitivity compared to conventional statistical methods.
• The stacked ensemble achieved high and stable specificity across different ensemble sizes.
• Machine learning models showed superior performance in detecting ABRs from EEG data.

Conclusions:

• The stacked ensemble model is more effective than conventional statistical and alternative machine learning methods for ABR detection.
• This AI-driven method has potential for automated ABR screening devices and evoked potential software.
• Further validation with diverse patient data is recommended to assess generalizability.