Digital Design of Photoplethysmogram (PPG) based Cardiovascular Disease Classifiers

Abstract

Photoplethysmogram is an optically obtained signal working based on the volumetric change of blood. As heart diseases are correlated with the pumping of blood, PPG can be studied for detecting cardiovascular diseases. Researchers have already analyzed PPG signals for various disease detection, including hypertension, coronary artery disease, diabetes, and others. Also, two important health parameters: heart rate and blood pressure have been predicted from PPG signals in several studies. However, most of the work has been done at the software level without any hardware implementation. In addition, two important cardiovascular diseases related to blood flow in the brain: cerebral infarction and cerebrovascular disease are yet to be explored based on PPG signal. Hence, this study aims to develop a hardware-based system that can detect several cardiovascular diseases - hypertension, cerebral infarction, cerebrovascular disease, diabetes, and a few combinations of them. The study checks the feasibility of detecting these diseases individually in a binary classification system and also in a multiclass classification system. A system is also implemented for predicting heart rate and blood pressure from PPG signals. The systems are developed in Xilinx system generator targeting Zedboard zynq 7000 and zynq ultrascale+ FPGA board. The binary classification system uses 11 features and applied SVM classifier to get the accuracy of 96.37%, 93.48%, 96.43%, and 88.46% for detecting hypertension, cerebral infarction, cerebrovascular disease, and diabetes, respectively, consuming a total of 0.693 W power. The multi-class classification system utilizes a total of 1.403 W of power, providing an accuracy of 79.83% for detecting 7 classes of diseases. Also, the heart rate and blood pressure estimation system utilizes 0.353 W of power. The heart rate is predicted with 4.04% error while systolic and diastolic blood pressure are estimated with 3.77% and 4.8% error, respectively. The designed prototype can be further extended to develop wearable devices, and smartwatches and can be useful for medical treatment and analysis.