Enhancing Auditory Brainstem Response Classification Based On Vision Transformer

Abstract

Abstract A method for testing the health of ear’s peripheral auditory nerve and its connection to the brainstem is called an auditory brainstem response (ABR). Manual quantification of ABR tests by an audiologist is not only costly but also time-consuming and susceptible to errors. Recently in machine learning have prompted a resurgence of research into ABR classification. This study presents an automated ABR recognition model. The initial step in our design process involves collecting a dataset by extracting ABR test images from sample test reports. Subsequently, we employ an elastic distortion approach to generate new images from the originals, effectively expanding the dataset while preserving the fundamental structure and morphology of the original ABR content. Finally, the Vision Transformer method was exploited to train and develop our model. In the testing phase, the incorporation of both the newly generated and original images yields an impressive accuracy rate of 97.83%. This result is noteworthy when benchmarked against the latest research in the field, underscoring the substantial performance enhancement achieved through the utilization of generated data.