EarSteth: Cardiac Auscultation Audio Reconstruction Using Earbuds.

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Medicine
• Telehealth Technologies

Background:

• Cardiac auscultation is crucial for diagnosing heart conditions.
• Traditional stethoscopes are impractical for remote telehealth consultations.
• A need exists for remote cardiac sound acquisition methods.

Purpose of the Study:

• To develop a system (EarSteth) for reconstructing cardiac auscultation audio using consumer earbuds.
• To adapt and evaluate machine learning models for heart sound reconstruction.
• To enable remote cardiac auscultation in telehealth settings.

Main Methods:

• Utilized consumer-grade active noise-cancelling earbuds with inner microphones.
• Developed EarStethNet, an adapted convolutional neural network (CNN) for audio super-resolution and heart sound reconstruction.
• Trained models using synchronous audio data from 15 healthy adults, comparing earbud and digital stethoscope recordings.

Main Results:

• EarStethNet accurately estimated interbeat intervals with a mean absolute error of 36.6 ± 51.1 ms.
• Reconstructed cardiac auscultation audio achieved a mean log spectral distance of 1.22 dB, indicating high fidelity.
• Demonstrated the feasibility of using earbuds for remote cardiac sound analysis.

Conclusions:

• EarSteth provides a viable solution for remote cardiac auscultation in telehealth.
• The developed AI model shows promise for accurate heart sound reconstruction from earbud audio.
• This technology can significantly enhance the capabilities of remote patient monitoring and virtual healthcare.