BPWave: An advanced Deep Learning Framework for Continuous Non-Invasive Cuffless Blood Pressure Estimation and Hypertension Detection

The precise measurement of blood pressure is essential for evaluating cardiovascular health and managing hypertension, a significant contributor to global morbidity and mortality. This study introduces two empirical formulae derived from normalized Photoplethysmography (PPG) values, facilitating accurate estimation of systolic (SBP) and diastolic (DBP) blood pressure levels. Leveraging a novel attention mechanism-based Artificial Neural Network (ANN), a robust model is developed capable of categorizing individuals into distinct hypertension stages i.e. Normal, Pre-Hypertension, Hypertension Stage-1 (HS1), and Hypertension Stage-2 (HS-2) using heart rate, SBP, and DBP data from publicly available datasets. The obtained root-mean-squared error (RMSE) values for SBP and DBP are 2.623 and 3.481, respectively, and mean absolute error (MAE) values for SBP and DBP are 3.049 and 3.852, respectively. Notably, the attention mechanism-based ANN model achieves a classification accuracy of 98.30%. Implemented on an Arduino UNO, the framework utilizes a PPG sensor for real-time testing, accurately predicting SBP and DBP and enabling hypertension classification. Additionally, an Android app has been developed for continuous monitoring of health parameters. Realtime continuous predictions on 11 volunteers using the proposed hardware framework yielded average (MAE ± SD) error values of 3.625 ± 2.58 mmHg for SBP and 4.085 ± 2.74 mmHg for DBP. Additionaly, mean arterial pressure (MAP) values were computed to contrast the performance of BPWave with existing cuff-based sphygmomanometer device.