Feature Extraction of Photoplethysmography Waveforms for the Development of Machine Learning Based Blood Pressure Measurements

High blood pressure poses a significant health risk of strokes, heart attacks, heart failure, and increases risks of complications from other illnesses. Blood pressure is predominantly taken manually with a pressure cuff by a medical professional which is a single measurement, usually months apart. Continuous blood pressure monitoring could provide a wealth of information about how an individual’s blood pressure actually behaves in their daily life. Accessible and accurate blood pressure monitoring device is a critical medical device as changes in blood pressure are often symptomless. We developed continuous blood pressure sensing using photoplethysmography, an optical method, on an artificial circulatory system integrating sensor data using machine learning. In order to get sufficient data points for our machine learning system to accurately measure blood pressure we created a model system of a human arm using a series of diaphragm pumps, check valves, silicon tubing, silicon to simulate optical properties of skin, and a blood mimicking fluid. I also developed an algorithm to detect chosen features with biological significance from the photoplethysmography waveform to input into our machine learning algorithm.