Physics-informed Neural Networks For Modeling Physiological Time Series For Cuffless Blood Pressure Estimation

PINN models are a type of deep neural network that are trained through transductive learning, which uses additional information from unlabeled input data. This is done by modifying the conventional loss function to include an additional loss term. The weight assignment of this additional term is not straightforward and must be carefully selected. This modified loss function is constructed based on Taylor’s approximation of known physiological dynamics that drive the translation of bioimpedance to blood pressure.