Auditory Attention Decoding

We developed an EEG-based model that can detect which sound a person is focusing on in real time during a two-speaker listening task. By applying a sliding-window linear decoder, our system tracks attention shifts every second, showing strong potential for use in smart hearing aids and auditory neurofeedback.

We developed and validated a cost-efficient EEG system that accurately tracks auditory attention using a minimal setup. By combining OpenBCI, Arduino, and a real-time decoder model, our system achieved high performance (90% offline, 78% real-time) even in a non-soundproof room. This study demonstrates the feasibility of auditory attention decoding (AAD) in realistic environments, paving the way for portable neuro-steered hearing aids and clinical neural tracking applications.

A deep learning model was developed to classify tympanic membrane images into four categories: normal, OME, COM, and cholesteatoma. The model achieved 97.2% accuracy and helped junior residents improve their diagnosis by up to 18%.This study shows the potential of AI as a clinical support and educational tool in otology.

In partnership with AIDOT Inc., we are developing an AI-powered digital otoscope that can analyze tympanic membrane images in real time.This device aims to improve the accuracy and efficiency of ear disease diagnosis, making it a valuable tool in both clinical and primary care settings.

A deep learning-based model was constructed to predict hearing recovery in over 1,100 patients with sudden sensorineural hearing loss (SSNHL). Using clinical and audiological data from the initial visit, the two-layered neural network achieved high prediction accuracy (AUC 0.92) in identifying patients likely to fully recover. This study highlights the potential of AI-driven prognosis tools to support personalized treatment planning and decision-making in SSNHL care.

A smartphone-based audiometry app was evaluated against conventional hearing tests in 70 patients. The app showed high reliability in detecting normal to mild hearing loss, but lower accuracy in moderate to severe and asymmetric cases.These findings highlight the potential of mobile audiometry as a convenient and accessible tool for hearing screening.