A Review of Photoplethysmography-Based Blood Pressure Monitoring: From Cloud-Based Machine Learning to TinyML Edge Deployment

Cuffless continuous noninvasive blood pressure (cNIBP) monitoring based on photoplethysmography (PPG) has enjoyed great success through a wealth of high-performing machine learning (ML) algorithms. Currently, this field is targeting AI-enabled real-time continuous blood pressure (BP) monitoring wearables. However, the superior state-of-the-art cloud-based models proposed in the literature incur heavy computational load and mandate a large memory footprint and energy requirements. Such factors hinder the deployment of these complex models on resource-constrained edge devices. Tiny machine learning (TinyML) has revolutionized the wave of wearable technology and enabled ML models to run inference on edge and tiny devices. However, limited research has explored the challenges and trade-offs in adapting PPG-based BP models to TinyML platforms. In this study, we attempt to fill this gap by examining the key limitations of existing cloud-based approaches, reviewing the limited attempts at edge-based implementation and the challenges encountered, and analyzing how recent cloud advancements can inform efficient TinyML edge solutions. This work provides a comprehensive reference for researchers, outlining key constraints, optimization strategies, and pathways for adapting cloud-based advancements to TinyML-powered PPG-based BP monitoring wearables.