Quantifying cardiometabolic risks by AI based assay of photoplethysmogram data

The National Heart, Lung, and Blood Institute awarded East Carolina University $561,100 under the Cardiovascular Diseases Research program (CFDA 93.837) on August 15, 2026 to develop a portable photoplethysmogram (PPG) prototype device and establish a platform method for noninvasive measurement of cardiometabolic disease biomarkers.

The research develops a 2-site PPG data acquisition system to derive blood pressure waveform signals and extract biomarker panels including systolic and diastolic blood pressures, blood glucose, and pulse wave velocity. The project applies machine learning models integrated into physics-informed neural networks to reduce noise artifacts such as motion artifacts in acquired PPG data and extract reliable biomarkers for quantifying cardiometabolic disease risk. Work is performed in Greenville, North Carolina, with a period of performance through April 30, 2029. This is a Project Grant, an assistance type supporting investigator-initiated research at academic institutions.

Abstract

Cardiometabolic diseases (cmds) are major causes of morbidity and mortality worldwide in the face of the escalation of obesity in populations of mid-to-high income countries despite the significant progress in prevention and treatment of cardiovascular diseases (cvds). The long-term goal is to noninvasively measure the blood pressure wave (BPW) propagating in artery by an optical method of photoplethysmogram (PPG) and extract a panels of biomarkers for continuous and quantitative assessing CMD risks. The objective of the proposed research is to develop a portable PPG prototype device and establish and validate a platform method by noninvasive acquisition of 2-site PPG data to derive the BPW waveform signals. The panel of CMD biomarkers obtained from the BPW waveform signals consists of systolic and diastolic blood pressures (BPS), blood glucose (BG) and pulse wave velocity (PWV) as the core metrics. Our hypothesis is that noninvasive 2- site acquisition of PPG data allows reliable measurement of BPS, BG and PWV values as core biomarkers to quantify CMD risks. A novel research strategy for development of PPG prototype devices and a platform method of data assay will be applied to significantly reduce noises such as motion artifacts in acquired PPG data, derive BPW waveform signals from the denoised PPG data and integrate 3 models with machine learning into physics-informed-neural networks (pinns) for obtaining the biomarker panel. The rational for the research strategy is that significant improvement of prognostic values of the CMD biomarkers derived from PPG data can be achieved by removing the "black-box" nature of deep learning using application of quantitative modeling of the basic physiological process underlying the PPG data. The objective of this project will be accomplished by three specific aims: (1) we will design and construct highly portable PPG prototype devices for acquisition of PPG data at the top and base of the same proper digital artery arising from palmar arches on an index finger of participant. (2) we will develop a simulation tool by integrating 3 models developed in our previous works for simulation of BPW propagating in a proper digital artery and scattered light distribution on skin surface, which will be used to train physics-informed neural networks (pinns) for obtaining the CMD biomarkers from the PPG data of the study participants. (3) we will conduct a human subject study of 250 participants to acquire PPG data in collaboration with the CVP rehab clinic of ECU health. Accuracy of the extracted CMD biomarkers with pinns will be determined to improve the data assay platform and test our hypothesis. The proposed research is innovative because IT (a) employs a 2-site measurement scheme to acquire PPG data and effective denoising algorithms for derivation BPW waveform signals of high signals-to-noise ratios, and (b) provides a systematic strategy for noninvasive determination of BPS, BG and PWV as the core CMD biomarkers by pinns that combines AI based data mining with first-principle based accurate modeling of hemodynamics and light-tissue interaction.