Auditory Brainstem Response Data Preprocessing Method for the Automatic Classification of Hearing Loss Patients.

Area of Science:

• Biomedical Engineering
• Neuroscience
• Medical Imaging Analysis

Background:

• Auditory Brainstem Response (ABR) testing objectively assesses hearing by measuring brainstem electrical signals via the auditory nerve.
• ABR is crucial for individuals unable to provide subjective feedback, such as infants, the elderly, and disabled patients.
• Current ABR data variability across devices poses challenges for consistent analysis and AI model training.

Purpose of the Study:

• To propose an image preprocessing pipeline for Auditory Brainstem Response (ABR) graph data.
• To standardize diverse ABR image data for improved deep learning model performance.
• To evaluate the efficacy of a deep learning model in classifying hearing loss using preprocessed ABR data.

Main Methods:

• Developed an image preprocessing technique to standardize ABR graph data from various measurement devices.
• Applied the VGG16 Convolutional Neural Network (CNN) model for classification tasks.
• Trained and tested the VGG16 model on a dataset of 10,000 preprocessed ABR images to classify hearing loss presence or absence.

Main Results:

• Standardized ABR image data improved the performance of the deep learning model.
• The VGG16 model demonstrated accuracy in classifying the presence or absence of hearing loss using the preprocessed data.
• Analysis of various weights verified the classification learning capabilities of the model.

Conclusions:

• The proposed image preprocessing method is effective for creating standardized ABR datasets for deep learning.
• This approach can enhance the accuracy of AI-driven hearing loss detection.
• The findings provide a foundation for setting criteria in preprocessing and learning medical graph data, including ABR graphs.