OCViTnet: Pulse Wave Learning for Cuffless Blood Pressure Estimation of Medical Service Robots

Blood pressure (BP) estimation through cuffless approaches is gaining prominence in precision healthcare, and the emergence of robotic medical services adds impetus to this trend with enhanced diagnostic capabilities. In our endeavor to enable medical service robots for cuffless BP self-estimation, we propose OCViTnet, a deep-learning network model grounded in pulse wave analysis (PWA). The model introduces a novel waveform processing architecture: an omni-scale convolution subnet captures single-period waveform characteristics, the Vision Transformer subnet extracts multi-period waveform features, and a regressor fuses phenotypic information with pulse wave features and performs BP regression. Experimental results perform exceptionally well in non-contact BP estimation from forehead imaging photoplethysmography. Additionally, this paper pioneers the utilization of the radial artery pulse wave for BP estimation. Leveraging the British Hypertension Society standard, our approach yields notable results, assessing the pulse diagnosis robot’s BP estimation performance at grade C/B (systolic/diastolic BP). This effort advances the potential of cuffless BP estimation in pulse diagnosis robotics and underscores the practical viability of our proposed intelligent methodology based on PWA.