Using machine learning to assist auditory processing evaluation

Abstract

Introduction Approximately 0.2–5% of school-age children complain of listening difficulties in the absence of hearing loss. These children are often referred to an audiologist for an auditory processing disorder (APD) assessment. Adequate experience and training is necessary to arrive at an accurate diagnosis due to the heterogeneity of the disorder. Objectives The main goal of the study was to determine if machine learning (ML) can be used to analyze data from the APD clinical test battery to accurately categorize children with suspected APD into clinical sub-groups, similar to expert labels. Methods The study retrospectively collected data from 134 children referred for ADP assessment from 2015 to 2021. Labels were provided by expert audiologists for training ML models and derived features from clinical assessments. Two ensemble learning techniques, Random Forest (RF) and Xgboost, were employed, and Shapley Additive Explanations (SHAP) were used to understand the contribution of each derived feature on the model's prediction. Results The RF model was found to have higher accuracy (90%) than the Xgboost model for this dataset. The study found that features derived from behavioral tests performed better compared to physiological test features, as shown by the SHAP. Conclusion The study aimed to use machine learning (ML) algorithms to reduce subjectivity in audiological assessments used to diagnose APD in children and identify sub-groups in the clinical population for selective interventions. Significance The study suggests that this work may facilitate the future development of APD clinical diagnosis software.