Measuring cardiac stroke volume through in-ear audio sensing

Area of Science:

• Cardiovascular physiology
• Biomedical engineering
• Artificial intelligence in healthcare

Background:

• Stroke volume (SV) is a critical indicator of cardiovascular health.
• Current SV measurement methods require specialized clinical equipment or dedicated wearables.
• There is a need for non-invasive, accessible methods for continuous SV monitoring.

Purpose of the Study:

• To develop and validate a deep learning system for estimating stroke volume (SV) using in-ear audio from commodity earbuds.
• To assess the system's accuracy and reliability compared to clinically validated devices.
• To explore the potential for scalable, affordable, out-of-clinic cardiovascular monitoring.

Main Methods:

• A deep learning system combining generative self-supervised/transfer learning and a transformer-based autoencoder was developed.
• The system processed in-ear audio signals to predict average stroke volume (SV).
• Data from 23 healthy participants were used, comparing estimations against clinically validated devices.

Main Results:

• The system achieved a mean absolute error of 5.24 ml for stroke volume (SV) estimation.
• A high Pearson correlation (r=0.94) was observed between predicted and true average SV.
• Percentage error within the limits of agreement was 11.05%, meeting clinical standards for SV measurement devices.

Conclusions:

• In-ear audio analysis using deep learning offers a viable method for accurate stroke volume (SV) estimation.
• This technology enables longitudinal, scalable, and affordable cardiovascular health monitoring outside clinical settings.
• The findings pave the way for widespread use of commodity devices for cardiovascular assessment.