AI-enabled Application Process Interface for On-device Personalized Blood Pressure Monitoring

We present a blood pressure estimation framework that utilizes a novel application process interface (API) for end-toend automated estimation of systolic and diastolic blood pressure. The proposed API estimates a pulse propagation feature, pulse arrival time (PAT), which is used to model systolic and diastolic blood pressure. The proposed API has low computational overhead and can be deployed to run seamlessly on resourceconstrained wearable devices. The API is reconfigurable, which makes it suitable for deploying in any IoT-enabled wearable device. The blood pressure estimation framework takes advantage of collaboration between resource-constrained wearable devices and semi-resource-constrained edge devices to facilitate personalized model training. The proposed framework is validated for performance on three datasets (two datasets are open-access and publicly available, and the third is an in-house dataset). Performance evaluation shows reliable performance with average errors of 8.3,3.7, and 4.01 mmHg for SBP estimation and 4.8, 2.9, and 1.8 mmHg for DBP estimation for the three datasets. Simulation results of the proposed API on a low-cost resourceconstrained microcontroller device showed excellent promise in terms of latency, power consumption, and memory requirements. The proposed blood pressure estimation framework can enhance the state of real-world continuous blood pressure monitoring by providing an affordable and sustainable solution for the consumer market.