From Pulse to Insight: An End-To-End Explainable Deep Spatio-Temporal Fusion Network for Cuffless Blood Pressure...

Continuous blood pressure monitoring is essential for timely detection and proper management of cardiovascular diseases. Though traditional cuff-based methods are accurate and widely used, they are inconvenient for repetitive and long-term monitoring. In this study, we introduce a novel end-to-end cuffless blood pressure estimation system that only uses photoplethysmography (PPG) signal. Our approach employs a deep spatio-temporal fusion network, integrating convolutional and recurrent layers to extract robust features from the PPG waveform. The model is trained and validated on the PulseDB dataset. Furthermore, we used Grad-Cam for explainability which helps in the decision-making process. We also developed a custom PPG signal acquisition device to facilitate real-time blood pressure monitoring via a dedicated application. The proposed system achieved mean absolute errors of 3.57 mmHg for systolic and 2.20 mmHg for diastolic blood pressure, along with low bias and acceptable variability which also meets both AAMI and BHS validation standards. This work demonstrates a reliable, non-invasive, and interpretable solution for cuffless continuous blood pressure monitoring, offering significant potential to improve cardiovascular care in clinical settings.Clinical relevance- This study provides a cuffless, real-time method for monitoring systolic and diastolic blood pressures using only PPG signal. The model's explainability enhances trust, making it a reliable tool for early detection and management of hypertension and related cardiovascular risks.