Abstract
Background and objective: Blood pressure (BP) is one of the crucial indicators that contains valuable medical information about cardiovascular activities. Developing photoplethysmography (PPG)-based cuffless BP estimation algorithms with enough robustness and accuracy is clinically useful in practice, due to its simplicity and noninvasiveness. In this paper, we have developed and tested two frameworks for arterial blood pressure (ABP) estimation at the central arteries using photoplethysmography and electrocardiogram. Methods: Supervised learning, as adapted by most studies regarding this topic, is introduced by comparing three machine learning algorithms. Features are extracted using semi-classical signal analysis (SCSA) tools. To further increase the accuracy of estimation, another BP estimation algorithm is presented. A single feed-forward neural network (FFNN) is utilized for BP regression with PPG features, which are extracted by SCSA and later used by FFNN as the network input. Both BP estimation algorithms perform robustly against MIMIC II database to guarantee statistical reliability. Results: We evaluated the performance against the Advancement of Medical Instrumentation (AAMI) and British Hypertension Society (BHS) standards, and we have compared the standard deviation (STD) of estimation error with current state of the arts. With the AAMI standard, the first method yields comparable performance against existing literature in the estimation of BP values. Regarding the BHS protocol, the second method achieves grade A in the estimation of BP values. Conclusion: We conclude that by using the PPG signal in combination with informative features from the Schrödinger's spectrum, the BP can be non-invasively estimated in a reliable and accurate way. Furthermore, the proposed frameworks could potentially enable applications of cuffless estimation of the BP and development of mobile healthcare device.