Significant Improvement in Precision of Real-Time Blood Pressure Prediction Based on Complete Cycles of Measured PPGs

A novel approach is presented for accurate and continuous monitoring of blood pressure (BP) using photoplethysmography (PPG) signals. The limitations of previous methodologies in accurately distinguishing between qualified and unqualified PPG waveforms, particularly in terms of complete cycles, have undermined the accuracy of BP estimations. To address this, a two-stage deep learning model combining 1D-CNN and LSTM for PPG quality assessment and 1D-CNN and GRU for BP estimation is proposed. Experimental results show that the 1D-CNN model achieves a high classification accuracy of 98.39% for PPG signal quality assessment. Without PPG quality assessment, the mean error (ME) ± standard deviation (SD) for systolic blood pressure (SBP) and diastolic blood pressure (DBP) is 1.31pm 9.84 mmHg and 0.19pm 5.75 mmHg, respectively. However, with the implementation of PPG quality assessment, the ME pm SD for SBP is reduced to 0.47pm 6.23 mmHg, and for DBP to 0.23pm 3.93 mmHg, respectively. These results highlight the effectiveness of the proposed method, providing a promising strategy for accurate, real-time, and continuous blood pressure monitoring based on complete cycles of measured PPGs.