A Novel Convolutional Neural Network-Based Algorithm for Heart Rate Measurement From Ballistocardiography Signals in...

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Healthcare
• Physiological Monitoring

Background:

• Continuous vital sign monitoring is crucial for early detection and improved patient outcomes.
• Traditional heart rate (HR) monitoring often requires skin contact, causing patient discomfort.
• Ballistocardiography (BCG) offers a promising noncontact alternative for continuous monitoring.

Purpose of the Study:

• Develop and validate a novel algorithm for accurate, noncontact, continuous HR monitoring using BCG signals and convolutional neural networks (CNNs).
• Enhance HR measurement accuracy across diverse healthcare settings by integrating time-domain peak detection, short-time Fourier transform, and CNN models.
• Ensure the algorithm's robustness, generalizability, and clinical applicability following FDA's Good Machine Learning Practice guidelines.

Main Methods:

• A CNN model was developed, trained on over 129,000 data points from 373 participants and tuned on over 75,000 data points from 192 participants.
• The algorithm integrates time-domain peak detection with short-time Fourier transform and CNNs for HR measurement from BCG signals.
• The algorithm was rigorously tested on over 70,000 data points from 205 participants across 5 independent studies, including ICU patients, to ensure performance in diverse conditions.

Main Results:

• The algorithm achieved a mean absolute error of under 3 bpm and a detection rate exceeding 80%.
• Bland-Altman analysis showed high accuracy with a minimal bias of 0.25 and limits of agreement within 8.59 bpm.
• Deming regression yielded a Pearson correlation coefficient of 0.97, demonstrating strong alignment with reference HR measurements.

Conclusions:

• The CNN-based algorithm provides a robust solution for contactless HR monitoring, overcoming limitations of previous methods in noise management and adaptability.
• Demonstrated accuracy in noisy, real-world clinical environments highlights its potential for widespread patient monitoring.
• The technology promises improved patient comfort and facilitates early detection of clinical deterioration.