CardioTML: ESP32-Based TinyML Wearable System for Early Identification of Cardiac Risk to Enhance Well-Being and Support Healthcare

Abstract

Cardio vascular disease (CVDs) are a leading cause of mortality worldwide, emphasizing the need for early detection and continuous monitoring to reduce emergency outcomes. This paper proposes an ESP32 hybrid edge-cloud wearable platform for real-time cardiac risk analysis using ECG and vital signs. The proposed CardioTML system combines an AD8232 ECG sensor and a MAX30102 PPG sensor for ECG, heart rate (BPM), and oxygen saturation (SpO2) measurements. A digital low-pass filter and 187-point morphological segmentation are also designed for real-time processing. The proposed system was tested on the MIT-BIH Arrhythmia Database, while maintaining the official training and testing splits, three classifiers: Logistic Regression, KNN, and Random Forest, Random Forest reported the highest Macro F1-score of 0.9589 for multi-class risk classification. Inference is performed using a cloud-based Flask API in a hybrid edge-cloud setup, and medical safety limits from WHO and AHA guidelines are incorporated to send emergency notifications through Telegram bot. The proposed framework offers a cost-effective and scalable approach for preventive cardiac surveillance and can be extended to a TinyML-based fully embedded system in the future. The system provides a low-cost, scalable solution supporting preventive healthcare, contributing to SDG 3 (Good Health and Well-Being) and promoting accessible remote monitoring.