ABR-Attention: An Attention-Based Model for Precisely Localizing Auditory Brainstem Response.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Artificial Intelligence

Background:

• Auditory Brainstem Response (ABR) is crucial for assessing auditory function and diagnosing neurological conditions.
• Extracting characteristic waves, like Wave V latency, from ABR provides objective auditory loss indicators.
• Current extraction methods are often time-consuming and labor-intensive for clinicians.

Purpose of the Study:

• To develop a novel deep learning network for automated Wave V latency extraction from ABR.
• To reduce the clinical workload associated with ABR analysis.
• To provide an accurate and efficient tool for objective localization of ABR characteristic waves.

Main Methods:

• Introduction of the ABR-Attention deep learning model, incorporating self-attention and derivative attention modules.
• Utilizing a 10-fold cross-validation approach to evaluate model accuracy.
• Assessing the model's performance across varying sound pressure levels (SPLs) and error scales.

Main Results:

• The ABR-Attention model achieved a high overall accuracy of 96.76 ± 0.41% in extracting Wave V latency.
• The model demonstrated efficacy with an error scale as low as 0.1ms.
• Experimental results validated the model's performance and robustness.

Conclusions:

• ABR-Attention presents a novel and effective deep learning solution for automated Wave V latency extraction.
• The model significantly reduces the burden on clinicians by automating a critical diagnostic step.
• This approach offers a new pathway for the objective and precise localization of ABR characteristic waves.