Deep Learning Methods on Blood Pressure Estimation Using Photoplethysmography

2019 · Research, Deep Learning, CNN, Hypertension

The Project

Blood pressure is an important bio-signal and useful physiological parameter that provides information regarding cardiovascular health. Current methods call for regular blood pressure monitoring using a cuff-based technique. However, the physical cuff can be cumbersome and inconvenient to use, making it a suboptimal solution that lead to inconsistent blood pressure monitoring. There are also other cuff-less methods of measuring blood pressure currently being researched, which involve collecting photoplethysmographic (PPG) sensor data from the index finger to obtain features that are correlated with blood pressure. In this project, we explored the possibility of predicting blood pressure on a mobile platform by approximating a PPG signal with a camera unit. PPG and arterial blood pressure from students at the University of Waterloo was collected in addition to the data recordings from the Multiparameter Intelligent Monitoring in Intensive Care (MIMIC) Database to train our model. A convolutional neural network (CNN) was used to evaluate scalograms produced by the PPG signals and output a prediction of the user’s blood pressure