Continuous Blood Pressure Estimation Through Optimized Echo State Networks

Technology is pushing cardiology towards non-invasive recording of continuous blood pressure, with methods that do not require the insertion of a pressure transducer in the Aorta. Although novel analyses based on the Electrocardiogram (ECG) and Photoplethysmography (PPG) provided an elegant model of the interaction between the heart and blood vessels necessary estimate systolic/diastolic points, these methods lack long-term stability and require intermittent re-calibrations. On the other hand, time-series centric algorithms are potentially able to map this coupling in a continuous way. Recurrent Neural Networks (RNN) in particular shown high accuracy and stability. We propose here a system that automatically optimizes, among other hyper-parameters, the input scaling of ECG and PPG signals individually, exploiting an efficient Bayesian Optimization process, on a Echo State Networks (ESN), to quantify continuous blood pressure. Compared to other RNN architectures, the ESN provide a faster training time and lower computational constraints that may allow the deployment on an embedded monitoring device. Preliminary results showed an accuracy of 80% and 98% in terms of Normalized Root Mean Squared Error and Median Symmetric Accuracy, respectively. Considering peaks’ estimation, the system achieves grade A (British Hypertension Society) for both Systolic and Diastolic points, making it comparable with clinical recommended devices.