Young Min park, MD: No relevant relationships to disclose.
Introduction: Although various acoustic metrics and subjetive measures have been used to evaluate patient's voice, they have not support enough information for diagnosing voice disease. Artificial intelligence technique has been used to diagnose and predict patient's prognosis in head and neck cancer and also has a potential to be used for diagnosing voice disease. The purpose of study is to improve the classification accuracy by using voice-based artificial intelligence algorithm for the diagnosis of voice diseases.
Methods: Severance Hospital dataset were used to establish classifiers and to verify the classifier's performance in the generated model. For the diagnosis of voice disorder, deep learning-based CNN models were established and classified only using patient's voice data. Classification accuracy was obtained by performing ensemble learning using CNN classification algorithm, transformer, spectrum anlaysis, and so on.
Results: More than 12,000 voice samples were collected from patients with voice disorder, In data preprocessing, we focused on a few target disease (laryngeal cancer, vocal nodule, spasmodic dysphonia) with the enough number of patients. We obtained classification accuracy of more than 84% in the established voice classification models using our dataset. To enhance the performance of our model, we used ensemble model to combine various artificial algorithm including CNN, transformer, and spectrum analaysis algorithm. The ultimate classification accuracy was more than 84 % for the voice based artificial intelligence model.
Conclusions: The results of our study suggest that ensemble learning aimed at training multiple classifiers is useful to obtain an increased classification accuracy. Although a large data amount is essential for artificial intelligence analysis, when an integrated approach is taken by combining various models, high diagnostic classification accuracy can be expected.
