TL;DR
This paper explores the feasibility of using "hearables" (ear-worn devices) to record electrocardiogram (ECG) signals for continuous cardiac monitoring. They demonstrate that signals captured from a single ear location can reliably reflect key cardiac rhythms, comparable to the standard Lead I ECG, even in noisy real-world conditions like simulated driving. This opens doors for unobtrusive, everyday health monitoring.
What Problem Does This Paper Address?
The increasing prevalence of wearable technology for health monitoring presents an opportunity for continuous, unobtrusive vital sign tracking. While smartwatches and chest straps are common, they have limitations: photoplethysmography (PPG) from smartwatches lacks detailed cardiac information, and chest straps can be obtrusive. "Hearables," devices worn in the ear, offer a promising alternative due to their stable position and potential for integration with audio accessories. However, the ability to capture a meaningful ECG signal from a single ear location, and its reliability in real-world scenarios, remained largely unexplored. This paper aims to bridge this gap by investigating the feasibility and quality of single-ear ECG recordings.
Methods
The researchers employed a two-pronged approach:
- Biophysics Modeling: They used a detailed 3D model of the human body to simulate ECG signal propagation from the heart to various head and ear locations. This allowed them to understand the underlying principles of how cardiac electrical activity translates to signals at these specific sites, isolating the ECG from other biological noise.
- Real-World Measurements:
- Experiment A (Resting): Ten subjects had ECG recorded from scalp, ear (helix and concha), and neck locations, alongside a standard wrist-based Lead I ECG. This experiment focused on establishing the basic feasibility and quality of signals under ideal, resting conditions.
- Experiment B (Dynamic): Five subjects performed a simulated driving task for an hour while wearing hearable devices. This dynamic scenario introduced real-world noise and movement artifacts, allowing for an assessment of the system's robustness.
Signal processing involved bandpass and notch filtering, R-peak detection, and averaging of multiple cardiac cycles to extract clear rhythms. Performance was evaluated using metrics like Pearson correlation with Lead I, amplitude ratios, wave timing accuracy, and normalized variance.
Datasets
The study utilized two main sources of data:
- Simulated Data: Generated from a VHP-Female Computational Phantom v. 2.1 and v. 2.2, incorporating detailed anatomical and electrical properties of human tissues.
- Measured Data: Collected from 10 healthy subjects during Experiment A (resting) and 5 subjects during Experiment B (simulated driving). The paper does not introduce a new benchmark dataset but rather uses existing experimental setups and subject data for evaluation.
Key Insights and Ideas
- The Ear as a Viable ECG Sensor: The study provides strong evidence that hearables can capture ECG signals that are remarkably similar in shape and timing to the standard Lead I ECG. This is a significant finding, as it suggests hearables can offer more than just basic heart rate monitoring.
- Lead I Equivalence: The single-ear ECG signals closely resemble Lead I, which is crucial for diagnosing various heart conditions like myocardial infarction and atrial fibrillation. This equivalence is attributed to the similar projection plane of the ear channels relative to the heart vector.
- Robustness in Noise: Even in a noisy, simulated real-world environment (driving), the hearable ECG signals demonstrated resilience. This suggests that with appropriate signal processing, these devices can provide reliable data for continuous monitoring.
- Cross-Ear vs. Single-Ear: While cross-ear measurements showed slightly higher signal quality, single-ear recordings also proved to be viable, especially with sufficient averaging. This is important for the development of compact, single-ear hearable devices.
- Impact of Averaging: The study highlights how averaging multiple cardiac cycles significantly improves the signal-to-noise ratio (SNR), making even lower-amplitude signals identifiable. This is key for extracting meaningful data from less ideal recording locations.
Key Results
- Signal Similarity: Cardiac rhythms recorded from the ear closely matched simulated equivalents and the reference Lead I ECG.
- Performance Metrics:
- Correlation: Left ear ECG showed a correlation of r=0.85 with Lead I during driving (N=240 cycles), and r=0.86 at rest. Cross-ear ECG achieved r=0.94 during driving.
- Wave Timing: RMSE for wave timings in the left ear was around 7 ms during driving, comparable to the 5 ms for cross-ear ECG.
- Amplitude Ratio: The RMS amplitude ratio between R-peaks and other waves was generally close to 1 for ear channels, indicating good waveform fidelity.
- Real-World Feasibility: Measurements during a one-hour simulated driving task demonstrated that single-ear and cross-ear ECG signals remained robust and identifiable, even with motion artifacts.
- Scalp ECG Limitations: The scalp ECG channel performed significantly worse, likely due to hair interference, lower signal amplitude, and higher background noise (EEG, EMG).
Limitations and Future Work
- Amplitude Discrepancy: A minor discrepancy was observed between simulated and measured R-peak amplitudes in the left and right ears, which the authors attribute to the model's sensitivity to electrode positioning. Further investigation with precisely positioned measurements is suggested.
- Amplitude Dependence: The quality of single-ear ECG might be dependent on the overall amplitude of the heart's electrical activity, as observed in participants with lower Lead I amplitudes.
- Future Work: The authors suggest integrating additional sensors like PPG and BCG into hearable platforms for comprehensive vital sign monitoring. They also aim to explore the detection of various cardiac conditions. The potential for 24/7 continuous monitoring and early alerts for heart malfunctions is a key future direction.
Summary updated