Iterative Demographic Attentional Feature Fusion-based CNN and Transformer Network for Accurate Cuffless Blood Pressure Estimation

Continuous cuffless blood pressure (BP) estimation is essential for cardiovascular disease monitoring. Recently, the use of deep learning models to automatically extract features and combine them with demographic features for continuous cuffless BP estimation has gained interest. Based on the observation that demographic features are highly correlated with BP estimation, this work proposes a new iterative demographic attentional feature fusion (AFF)-based CNN and Transformer network for better fusing the demographic features with the electrocardiogram (ECG) and photoplethysmography (PPG) features, as well as accurate BP estimation. This work tested model performance using a large open BP dataset, i.e., PulseDB. The BP estimation performance of the proposed model on PluseDB dataset meets the standards of the Association for the Advancement of Medical Instrumentation (AAMI) and achieves Grade A at the British Hypertension Society (BHS) standard in the estimate of systolic blood pressure (SBP) and diastolic blood pressure (DBP). The estimations of SBP and DBP have mean absolute errors (MAE) of 3.79 mmHg and 2.37 mmHg, respectively.