Abstract:Background and Objective: Coronary artery disease (CAD) and heart failure are the most common cardiovascular diseases. Non-invasive diagnostic testing for CAD requires radiation, heart rate acceleration, and imaging infrastructure. Early detection of left ventricular dysfunction is critical in heart failure management, the best measure of which is an elevated left ventricular end-diastolic pressure (LVEDP) that can only be measured using invasive cardiac catheterization. There exists a need for non-invasive, safe, and fast diagnostic testing for CAD and elevated LVEDP. This research employs nonlinear dynamics to assess for significant CAD and elevated LVEDP using non-invasively acquired photoplethysmographic (PPG) and three-dimensional orthogonal voltage gradient (OVG) signals. PPG (variations of the blood volume perfusing the tissue) and OVG (mechano-electrical activity of the heart) signals represent the dynamics of the cardiovascular system.Methods: PPG and OVG were simultaneously acquired from two cohorts, (i) symptomatic subjects that underwent invasive cardiac catheterization, the gold standard test (408 CAD positive with stenosis≥ 70% and 186 with LVEDP≥ 20mmHg) and (ii) asymptomatic healthy controls (676). A set of Poincaré-based synchrony features were developed to characterize the interactions between the OVG and PPG signals. The extracted features were employed to train machine learning models for CAD and LVEDP. Five-fold cross-validation was used and the best model was selected based on the average area under the receiver operating characteristic curve (AUC) across 100 runs, then assessed using a held-out test set.Results: The Elastic Net model developed on the synchrony features can effectively classify CAD positive subjects from healthy controls with an average validation AUC=0.90±0.03 and an AUC= 0.89 on the test set. The developed model for LVEDP can discriminate subjects with elevated LVEDP from healthy controls with an average validation AUC=0.89±0.03 and an AUC=0.89 on the test set. The feature contributions results showed that the selection of a proper registration point for Poincaré analysis is essential for the development of predictive models for different disease targets.Conclusions: Nonlinear features from simultaneously-acquired signals used as inputs to machine learning can assess CAD and LVEDP safely and accurately with an easy-to-use, portable device, utilized at the point-of-care without radiation, contrast, or patient preparation.

A Signal Processing Approach for Detection of Hemodynamic Instability before Decompensation

Predicting patient decompensation from continuous physiologic monitoring in the emergency department

AI-Enabled Advanced Development for Assessing Low Circulating Blood Volume for Emergency Medical Care: Comparison of Compensatory Reserve Machine-Learning Algorithms

A Signal Processing Framework for the Detection of Abnormal Cardiac Episodes

Non-invasive Waveform Analysis for Emergency Triage via Simulated Hemorrhage: An Experimental Study using Novel Dynamic Lower Body Negative Pressure Model

Classification of Blood Volume Decompensation State via Machine Learning Analysis of Multi-Modal Wearable-Compatible Physiological Signals

Synchronized wearables for the detection of haemodynamic states via electrocardiography and multispectral photoplethysmography

Automated Recognition of Hypertension Through Overnight Continuous HRV Monitoring

Non-invasive biomarkers for detecting progression toward hypovolemic cardiovascular instability in a lower body negative pressure model

Defining Postinduction Hemodynamic Instability With an Automated Classification Model

Detecting Patterns of Physiological Response to Hemodynamic Stress via Unsupervised Deep Learning

A novel technique for the detection of myocardial dysfunction using ECG signals based on CEEMD, DWT, PSR and neural networks

Non-Invasive Heart Failure Evaluation Using Machine Learning Algorithms

Efficient detection of cardiac abnormalities via a simplified score-based analysis of the ECG signal

Multimodal Variational Autoencoder for Low-cost Cardiac Hemodynamics Instability Detection

Cardiac Bioengineering Analysis of Electrophysiological Signals Driven by Deep Learning

Robust algorithm to locate heart beats from multiple physiological waveforms by individual signal detector voting

Predictive Modeling of Biomedical Signals Using Controlled Spatial Transformation

Technology Development for Simultaneous Wearable Monitoring of Cerebral Hemodynamics and Blood Pressure

Deep Metric Learning for the Hemodynamics Inference with Electrocardiogram Signals

Predicting cardiac disease from interactions of simultaneously-acquired hemodynamic and cardiac signals