Hypertension Detection Based on Photoplethysmography Signal Morphology and Machine Learning Techniques

Abstract

In our modern digitalized world, hypertension detection represents a key feature that enables self-monitoring of cardiovascular parameters, using a wide range of smart devices. Heart rate and blood oxygen saturation rate are some of the most important ones, easily computed by wearable products that are provided by the photoplethysmography (PPG) technique. Therefore, this low-cost technology has opened a new horizon for health monitoring in the last decade. Another important parameter is blood pressure, a major predictor for cardiovascular characterization and health related events. Analyzing only PPG signal morphology and combining the medical observation with machine learning (ML) techniques, this paper develops a hypertension diagnosis tool, named the ANC Test™. During the development process, distinguishable characteristics have been observed among certain waveforms and certain types of patients that leads to an increased confidence level of the algorithm. The test was enchanted by machine learning models to improve blood pressure class detection between systolic normotensive and hypertensive patients. A total of 359 individual recordings were manually selected to build reference signals using open-source available databases. During the development and testing phases, different ML models accuracy of detecting systolic hypertension scored in many cases around 70% with a maximum value of 72.9%. This was resulted from original waveform classification into four main classes with an easy-to-understand nomenclature. An important limitation during the recording processing phase was given by a different PPG acquisition standard among the consulted free available databases.