Combining algorithms for the automated detection of auditory brainstem responses in newborns

Congenital hearing loss is a significant health problem, with a worldwide incidence of >6 per 1000 live births. Late diagnosis will delay appropriate treatment, leading to potential neurodevelopment problems. Early diagnosis requires neonatal hearing screening, where one of the most used techniques is automated Auditory Brainstem Responses (aABR). Most aABR methods utilize statistical approaches to analyze the signal's temporal or spectral parameters. While both approaches are widely used, the former is susceptible to noise/artifacts, and the latter lack of analysis of the latencies of the different waves. This work aims to develop, by combining existing methods, an aABR detection algorithm that analyzes the signal in the time domain and improves the performance of the single methods, even in the presence of long latencies of wave V. The development of the algorithm involved evaluating three methods and their combinations in a pilot study. Finally, the best variant was validated in a clinical trial with 300 neonates. The validation results confirmed a specificity of 94.11 % and a sensitivity of 100 %, similar to other studies reported in the literature. These results demonstrated that the proposed algorithm is an effective tool for detecting hearing loss in neonates.