AI-Based Real Time Health Monitoring System for Comprehensive Vital Sign Assessment Using PPG Signals

Continuous health monitoring is essential for the early detection of physiological abnormalities and chronic medical events. However, conventional multi-parameter patient monitoring systems remain expensive and inaccessible for large-scale non-clinical deployment. This paper presents a low-cost (less than <tex-math notation="LaTeX">\}15$ </tex-math>) real-time health monitoring system that simultaneously estimates oxygen saturation (SpO2), heart rate (HR), and respiratory rate (RR) using a single photoplethysmography (PPG) signal acquired from a MAX30102 sensor interfaced with an ESP32 microcontroller. SpO2 is computed using the ratio-of-ratios method on red and infrared signals, HR is derived through adaptive peak detection on a fourth-order Butterworth bandpass-filtered signal (0.5–4 Hz), and RR is estimated using Hilbert transform-based envelope detection of respiratory-induced modulations, eliminating the need for dedicated respiratory sensors. The extracted physiological parameters are further analyzed using a multilayer perceptron (MLP) model for automated health-risk classification into Healthy, Mild Risk, and Severe Risk categories, achieving an accuracy of 91.2%. Experimental validation involving 200 participants aged 18–70 years demonstrated mean absolute errors of ±0.21% for SpO2, ±0.68 BPM for HR, and ±0.62 breaths/min for RR, with Pearson correlation coefficients of 0.980, 0.985, and 0.92, respectively, when compared with a GE Healthcare B105 Patient Monitor used as the clinical reference device. Additional ambulatory evaluations during walking and light physical activity maintained correlation coefficients above 0.91 for all parameters. The proposed system supports more than 48 hours of continuous operation on a 1000 mAh battery with an end-to-end latency of approximately 1.2 seconds. Combined with a web-based interface for real-time visualization, automated reporting, and intelligent risk assessment, the proposed solution offers a scalable and cost-effective approach for remote healthcare monitoring and early identification of physiological abnormalities.