mm-Wave Radar-Based Vital Signs Monitoring and Arrhythmia Detection Using Machine Learning

Area of Science:

• Biomedical Engineering
• Signal Processing
• Artificial Intelligence

Background:

• Non-contact vital sign monitoring is crucial for remote patient care.
• Existing methods may be invasive or limited in real-world scenarios.
• Radar technology offers potential for unobtrusive physiological monitoring.

Purpose of the Study:

• To develop and validate a non-contact, non-invasive system for measuring human heart and breathing rates using FMCW mm-wave radar.
• To propose a novel diagnostic system employing artificial neural networks for arrhythmia detection.
• To analyze the impact of various factors (orientation, distance, movement) on measurement accuracy.

Main Methods:

• Utilized a 77 GHz FMCW mm-wave radar system for data acquisition.
• Reconstructed heartbeat phase signals using Fourier series analysis.
• Developed and trained a three-layer artificial neural network (ANN) using the MIT-BIH database.
• Evaluated system performance against a reference device using statistical metrics (MSE, MAE, SD, R-squared).

Main Results:

• Optimal measurement performance was observed with the individual positioned directly in front of the radar at distances of 90 cm and 120 cm.
• Lowest standard deviation and mean squared error were achieved for heart rate during walking and breathing rate while motionless.
• The trained ANN achieved 93.9% training accuracy and an R-squared value of 0.876.
• The diagnostic tool demonstrated a mean test accuracy of 75% across 15 subjects.

Conclusions:

• The FMCW mm-wave radar system is a viable tool for non-contact vital sign monitoring.
• The proposed ANN-based diagnostic system shows promise for arrhythmia detection.
• Further research can optimize the system for diverse conditions and improve diagnostic accuracy.