Abstract
Hearing loss is a pervasive global health challenge with profound impacts on communication, cognitive function, and quality of life. Recent studies have established age-related hearing loss as a significant risk factor for dementia, highlighting the importance of hearing loss research. Auditory brainstem responses (ABRs), which are electrophysiological recordings of acoustically evoked synchronized neural activity from the auditory nerve and brainstem, serve as in vivo correlates for sensory hair cell and synaptic function, hearing sensitivity, and other critical readouts of auditory pathway physiology, making them highly valuable for both basic neuroscience and clinical research. Despite their utility, traditional ABR analyses rely heavily on subjective manual interpretation, which may introduce variability and pose challenges for reproducibility across studies. Here, we introduce Auditory Brainstem Response Analyzer (ABRA), a novel suite of open-source ABR analysis tools powered by deep learning, which automates and standardizes ABR waveform analysis. ABRA employs convolutional neural networks trained on diverse datasets collected from multiple experimental settings, achieving rapid and unbiased extraction of key ABR metrics, including peak amplitude, latency, and auditory threshold estimates. We demonstrate that ABRA’s deep learning models provide performance comparable to expert human annotators while dramatically reducing analysis time and enhancing reproducibility across datasets from different laboratories. By bridging hearing research, sensory neuroscience, and advanced computational techniques, ABRA facilitates broader interdisciplinary insights into auditory function. An online version of the tool is available for use at no cost at https://abra.ucsd.edu .
Full Text
The Full Text of this preprint is available as a PDF (2.3 MB). The Web version will be available soon.