Evaluation of the performance of machine learning algorithms applied to voice parameters in prioritizing candidate for office laryngoscopy: Automated triaging in focus
1 Department of Otorhinolaryngology, Obafemi Awolowo University, Ile-Ife, Osun State, Nigeria.
2 African Institute for Science Policy and Innovation, Obafemi Awolowo University, Ile-Ife, Nigeria.
Research Article
World Journal of Advanced Research and Reviews, 2021, 11(01), 169–175
Article DOI: 10.30574/wjarr.2021.11.1.0286
Publication history:
Received on 18 May 2021; revised on 19 July 2021; accepted on 21 July 2021
Abstract:
Background: We examined the performance of different Machine Learning Algorithms while also comparing two methods of voice assessment to look at the workability of automating triage of patient that will need prompt office laryngoscopy.
Methods: We recruited consecutive adult subjects excluding those with a history of being regular singers or choristers in the past one year and those with the previous history of laryngeal trauma. We then carried out the perceptual voice assessments on the GRBAS Scale and also obtained the basic acoustic parameters of the voice samples. Laryngeal examinations with 70-degree Hopkins’ Rod were then carried out by another examiner for all the participants to identify gross laryngeal changes or lesions. We then evaluated each machine learning algorithm comparing the perceptual and acoustic parameters in determining how well each algorithm predicts the presence of those categorized with having lesion or not by the laryngeal examination.
Results: One hundred and twenty respondents were analyzed out of which 89(74.2%) were females. The mean age was 46.5 ± 9.2 years. The perceptual evaluation generally outperformed the acoustic evaluation. Also, the Naïve Bayes Classifier (NBC) outperformed other algorithms with a F1 score of 0.55 followed by Artificial Neural Network (ANN) with the score of 0.53. However, the ANN outperformed the other with regards to the Area-under-the-curve (AUC).
Conclusion: When these metrics are taken together, the ANN still remains the best algorithm for this dataset. We are however cognisance of the needed improvement to the various aspects of this work including a larger dataset more scientific sampling.
Keywords:
Machine Learning; Voice; Office Laryngoscopy; Larynx; Metrics; Triage
Full text article in PDF:
Copyright information:
Copyright © 2021 Author(s) retain the copyright of this article. This article is published under the terms of the Creative Commons Attribution Liscense 4.0