AI-Based Early Cardiac Risk Detection and Emergency Alert Using Multi- Parameter Wearable Physiological Monitoring

Abstract

Cardiovascular diseases (CVDs) remain the leading cause of global mortality, with acute events such as myocardial infarction and sudden cardiac arrest often occurring without sufficient prehospital warning. Advances in wearable sensor technology and edgecomputing Artificial Intelligence (AI) have enabled non-invasive, continuous monitoring of multi-parameter physiological signals, including Electrocardiogram (ECG), Photoplethysmogram (PPG), Heart Rate Variability (HRV), Blood Oxygen Saturation (SpO2), and Skin Temperature. This review evaluates existing paradigms for automated cardiac arrhythmia detection, ischemic risk profiling, and emergency response dispatch. The primary foundational methodology examined leverages a hybrid Convolutional Neural Network and Long Short-Term Memory Network (1D-CNN-LSTM) paired with SHapley Additive exPlanations (SHAI / SHAP) for model interpretability. Evaluated across benchmark datasets including PhysioNet MIT-BIH Arrhythmia and the PTB-XL ECG Database, the hybrid deep learning architecture achieves a classification accuracy of 98.42% for multiclass arrhythmia detection and ischemic episode classification. Beyond single-lead ECG analysis, this paper reviews self-supervised representation learning and multi-sensor fusion models designed to mitigate motion artifacts and signal loss in free-living environments. Based on the reviewed literature, a critical research gap is identified: existing wearable solutions treat risk detection and emergency escalation as isolated processes, lacking real-time localized edge inference paired with explainable clinical decision-support. To address this, we propose an end-to-end framework combining multi-parameter wearable sensor data, edge-deployed Lightweight MobileNet-LSTM networks, Explainable AI (XAI) feature attribution, automated dynamic risk scoring, and automated emergency alert escalation with GPS location tracking.