AI-Enabled Wearable Hemodynamic Monitoring System for Early Identification of Thrombotic Events

Abstract

Thrombotic processes are usually silent, and the initial hemodynamic changes pass undiagnosed through the normal clinical evaluation process that was based on infrequent and symptomatic measurement. The necessity in the constant, non-invasive monitoring has led to the creation of wearable administrations that can record when changes in the cardiovascular system are dynamic and which occur before the emergence of thrombases. This paper presents an AI empowered wearable hemodynamic monitoring system based on the implementation of multi-modal sensors such as photoplethysmography, impedance plethysmography, electrocardiography, temperature, and inertial measurement units to obtain high-fidel physiological signals in real time. Effective preprocessing methods, including adaptive filtering, noise suppression, and motion-artifact elimination, were used to enhance signal quality, and time-domain, morphological, cardiac-timing, autonomic, and thermal features were extracted. The most discriminating parameters to predict early thrombotic-risk were determined through feature selection with the use of PCA and RFE. A deep-learning model that was hybrid, consisting of convolutional and long short-term memory networks, with additional support of anomaly detection algorithms was trained using both synthetic and clinically validated datasets to represent the subtle deviations in vascular resistance, venous return, autonomic balance, and waveform morphology. It was found that the developed model had significant predictive performance with 93 percent of accuracy, 91 percent sensitivity, 94 percent specificity and AUC of 0.95 to distinguish between normal and pre-thrombotic physiological states. This system also detected the latency up to 65 percent times of traditional methods, and real time inference on edge hardware took only 32 ms per segment, and continuous monitoring could be carried out. These results confirm that AI-enhanced wearable, multi sensor hemodynamic signal integration will provide a predictive, early warning system, with great potential to preventive and remote health use.