Abstract
Heart rate (HR) is a critical indicator of cardiovascular health, traditionally measured via electrocardiograph (ECG). However, ECG's practical limitations in daily settings have led to the rise of smartphone-based alternatives like photoplethysmography (PPG). This scoping review investigates gyrocardiography (GCG)—an emerging method that uses a smartphone’s built-in gyroscope to detect heart rate through chest vibrations. Analyzing seven key studies from the last decade, this review evaluates data collection, pre-processing, and HR estimation algorithms. While GCG shows promise due to its hardware-free nature and relative immunity to gravity-induced bias, the field currently faces challenges regarding standardized performance metrics and the need for algorithms that function beyond a supine position.
Introduction
Monitoring heart rate is essential for diagnosing conditions like Atrial Fibrillation (AFib) and Coronary Artery Disease (CAD). While traditional ECGs are the gold standard, they are often impractical for long-term daily use. Smartphone-based PPG is a popular alternative but remains highly sensitive to ambient light and motion artifacts.
Gyrocardiography (GCG) offers a compelling solution by measuring the angular vibrations of the chest wall caused by cardiac activity. Unlike accelerometers, gyroscopes are theoretically more robust against device tilt relative to gravity. Despite this potential, GCG remains a relatively overlooked sensor in mobile health. This paper reviews the current state of research to determine the clinical readiness of smartphone-based GCG.
Methodology
The authors conducted a scoping review following PRISMA guidelines, filtering an initial pool of 114 papers down to seven core studies published between December 2012 and January 2023.
The selection process focused on studies that utilized smartphone gyroscope data for HR estimation, excluding review papers, posters, and conference abstracts to ensure a high standard of technical detail.
Technical Pipeline
The reviewed studies generally follow a three-stage technical pipeline:
- Data Collection: Participants typically place a smartphone on their sternum while in a supine (lying down) position.
- Pre-processing: To isolate the faint heartbeat signal from breathing noise and sensor drift, studies employ bandpass filters (typically 1–45 Hz). Advanced techniques like Singular Spectrum Analysis (SSA) and Successive Mean Quantization Transform (SMQT) are used to balance the signal's dynamic range.
- HR Estimation: Researchers primarily use two algorithmic approaches: Short-term Auto-Correlation (S-AC), which identifies repeating patterns in the signal, and Peak Detection, which counts heartbeats by identifying local maxima in the processed waveform.
Datasets and Participant Diversity
The core studies utilized datasets ranging from 11 to 435 participants. These cohorts often included individuals with pre-existing conditions such as AFib and heart failure to test the algorithms' diagnostic utility.
A notable trend in the literature is the concentration of research; five of the seven analyzed studies originated from the same research group, leading to significant reuse and expansion of the same core datasets. Validation was typically performed against "gold standards" such as clinical-grade ECGs or PPG sensors.
Key Insights and Results
The "Feature" vs. "Metric" Gap
A critical finding of this review is that five out of the seven studies did not report the raw accuracy of heart rate detection. Instead, they treated the derived heart rate as a feature for machine learning models designed to classify diseases (e.g., detecting AFib) rather than a standalone metric.
Gravity and Orientation
GCG’s primary advantage over seismocardiography (accelerometer-based) is its inherent robustness against gravity. Because gyroscopes measure angular velocity, they are less biased by the phone's orientation. Research indicates that the Y-axis (the longitudinal axis of the phone) often provides the most reliable data for HR estimation.
Quantitative Performance
Due to a lack of standardization, quantitative results are sparse. However, the studies that did report metrics showed high potential:
- Accuracy: One study reported a Median Absolute Error (MedAE) of 1.03 beats per minute (bpm).
- Reliability: Another study cited a total error rate of 1.18%, consisting of minimal missed beats and false detections.
Challenges and Future Directions
For GCG to move from the lab to clinical practice, several hurdles must be overcome:
- Posture Constraints: Current research is almost entirely limited to supine subjects. Future algorithms must account for the noise introduced when users are sitting or standing.
- Hardware Variance: There is no universal algorithm that accounts for the varying sensor sensitivities across different smartphone models.
- Standardization: The authors advocate for a standardized protocol using ECG as a reference and Mean Absolute Error (MAE) as a primary metric to allow for fair comparison between emerging algorithms.
Summary updated