N. Kumar, S. Leo, J. Zhao, S. Lalande, D. Akinwande
University of Texas at Austin,
United States
Keywords: radio frequency, blood pressure, machine learning
Summary:
Current wearable blood pressure monitoring devices that use photoplethysmography (PPG) technology in smartwatches face significant accuracy limitations, particularly when used on individuals with darker skin tones. This bias stems from reduced optical penetration and weak light-artery interaction in PPG sensors, potentially leading to inaccurate data collection and biased AI algorithms for blood pressure estimation. This research presents an innovative solution: a smartwatch-integrable, low-power wearable antenna biosensor that measures bioimpedance (BioZ) changes from radial artery blood flow using radio frequencies (RF) for continuous blood pressure estimation. The proposed RF BioZ-based approach offers several advantages over traditional PPG methods. By utilizing RF signal reflection at the tissue-artery interface rather than optical measurements, the system provides more equitable monitoring across diverse demographics, as it is not affected by skin tone variations. The device leverages existing smartwatch WiFi antennas for RF BioZ measurements, reducing integration complexity while providing a more stable platform for data collection. The research team conducted a study with six normotensive individuals aged 20-29 years under IRB approval. The system utilized an inverted F-type WiFi antenna mounted on a 3D-printed watch strap, with measurements taken directly above the radial artery. Data collection involved various activities to capture different blood pressure ranges, including rest periods, handgrip exercises, ice bath exposure, and Valsalva maneuvers. Signal processing included alignment and normalization based on heartbeat intervals, with features extracted from both frequency and time domains. Results demonstrated strong performance, with mean absolute errors (MAE) of 5.67 ± 6.05 mmHg for RF BioZ measurements. The system achieved British Hypertension Society (BHS) Grade A and met the Association for the Advancement of Medical Instrumentation (AAMI) requirements for non-cuff-based blood pressure measuring technology. More than 60% of the estimated blood pressure readings fell within the defined threshold of 5mmHg, and the overall MAE remained under 5mmHg for both systolic and diastolic measurements. The system showed comparable performance to medical-grade pulse oximeter sensors while offering additional benefits such as enhanced wearability through non-contact format, lower power consumption, and immunity to skin color-influenced data corruption. This research represents a significant step toward developing more equitable health monitoring solutions, potentially enabling the creation of unbiased AI algorithms for blood pressure estimation across diverse populations. Future work will focus on expanding the participant pool, testing different antenna orientations to improve motion artifact resistance, and developing large-scale generalized blood pressure estimation models.