Comparison of machine learning models to classify Auditory Brainstem Responses recorded from children with Auditory
Hasitha Wimalarathna1, Sangamanatha Ankmnal-Veeranna2, Chris Allan3
1Department of Electrical & Computer Engineering, Western University, London, Ontario, Canada; National Centre for Audiology, Western University, London, Ontario, Canada.
Computer Methods and Programs in Biomedicine
|January 30, 2021
Summary
Machine learning models can automate auditory brainstem response (ABR) analysis for children with listening difficulties. The Xgboost algorithm achieved 92% accuracy, offering a potential clinical tool for audiologists.
Area of Science:
- Neuroscience
- Biomedical Engineering
- Machine Learning
Background:
- Auditory brainstem responses (ABRs) assess auditory nervous system integrity in individuals with listening difficulties.
- Manual ABR interpretation by audiologists requires expertise and can be prone to error.
- Machine learning (ML) offers a promising approach to automate ABR analysis and enhance diagnostic accuracy.
Purpose of the Study:
- To identify a suitable ML technique for automating ABR analysis within the Auditory Processing Disorder (APD) clinical test battery.
- To reduce human error and improve the efficiency of ABR interpretation in clinical settings.
Main Methods:
- Analyzed ABR responses from 136 children evaluated for auditory processing difficulties.
- Applied various ML algorithms including SVM, RF, DT, GB, Xgboost, and NN.
- Utilized signal feature extraction techniques and statistical testing to determine the most robust model.
Main Results:
- Identified clinically significant time-frequency signal features.
- The Xgboost ML model demonstrated the highest robustness and accuracy at 92%.
- This model effectively identified neurological abnormalities in ABR waveforms.
Conclusions:
- Accurate ML models can automate suprathreshold ABR waveform analysis.
- This research is expected to lead to an ML-based screening tool for audiologists.
- The findings may guide future development of ML paradigms to enhance clinical audiology test batteries.
