Predicting blood pressure from the change of photoplethysmography features

Blood pressure (BP) is a crucial indicator for assessing cardiovascular health, and effective monitoring contributes to early treatment of cardiovascular diseases. Photoplethysmography (PPG) has shown immense potential in non-invasive BP monitoring. However, existing studies often overlook the impact of feature variations on BP changes. Therefore, this study propose a blood pressure estimation method based on PPG feature changes. Initially, high-quality PPG segments are obtained through signal processing from PPG waveforms, and 68 contributory PPG differential features (DF) for BP changes are derived through differencing. Subsequently, optimal subsets of DFs are identified using correlation analysis and stepwise forward feature selection. Finally, an interpretable model is employed to map DF to BP changes, incorporating calibration values for accurate BP estimation. The proposed method is evaluated on self-collected datasets. The root mean square error (RMSE) for systolic blood pressure (SBP) is 8.39 mmHg, and for diastolic blood pressure (DBP), it is 5.75 mmHg. The corresponding Pearson correlation coefficient (PCC) values are 0.88 and 0.83. Correlation analysis indicates a close relationship between PPG differential features and BP changes. Compared to direct use of PPG features, the RMSE of BP estimation improves by approximately 40%, highlighting the effectiveness of PPG differential features in BP estimation.