Prediction of ECG fiducial parameters from PPG signals for the analysis of cardiovascular diseases: A novel Gaussian process regression-based approach

Abstract

PurposeCardiologists use 12-lead electrocardiograph (ECG) and prolonged continuous Holter monitoring to detect abnormalities in heart rhythm, which could diagnose or hint at different cardiac diseases. However, the availability of a costly ECG machine and the expertise to interpret ECG is suboptimal in our country. The current state of the art proposes mitigating the cost and easy diagnosis by physicians with the use of low cost photoplethysmograph (PPG), which will deliver ECG information in terms of ECG fiducial parameters. MethodsThe MIMIC-III matched paired datasets of PPG/Lead-II ECG signals with the six classes (right bundle branch block (RBBB), hypertension (HYPER), arrhythmia-atrial fibrillation (ARYTH-ATRFAB), dilated cardiomyopathy (DCM), complete heart block (CHB) and NORMAL) associated cardiovascular diseases (CVDs) syndromes of 30 patients with 5 minutes duration each were taken from Physionet. The proposed work tested by predicting ECG fiducial parameters (P, Q, R, S and T) through Gaussian process regression(GPR)-based predictor by using the PPG fiducial parameters with a training and testing ratio of 70:30. The performance was analyzed by different performance evaluation parameters e.g., RMSE, rRMSE, etc. ResultsAlthough the shape of the ECG may differ in various ways from that of the normal ECG, the methodology can predict the ECG fiducials from PPG for a variety of CVD patients with a performance accuracy of 95–99% by evaluating through seven performance metrics. ConclusionThe ECG fiducial parameters can be used to analyze CVDs captured from a low-cost, handy, skin-compatible continuous and non-invasive optical recording device well known as a PPG sensor.