Machine Learning-Based Pulse Wave Analysis for Early Detection of Abdominal Aortic Aneurysms Using In Silico Pulse Waves

Physics-Informed Graph Neural Networks to solve 1-D equations of blood flow

BACKGROUND AND OBJECTIVE

Computational models of hemodynamics can contribute to optimizing surgical plans, and improve our understanding of cardiovascular diseases. Recently, machine learning methods have become essential to reduce the computational cost of these models. In this study, we propose a method that integrates 1-D blood flow equations with Physics-Informed Graph Neural Networks (PIGNNs) to estimate the propagation of blood flow velocity and lumen area pulse waves along arteries.

METHODS

Our methodology involves the creation of a graph based on arterial topology, where each 1-D line represents edges and nodes in the blood flow analysis. The innovation lies in decoding the mathematical data connecting the nodes, where each node has velocity and lumen area pulse waveform outputs. The training protocol for PIGNNs involves measurement data, specifically velocity waves measured from inlet and outlet vessels and diastolic lumen area measurements from each vessel. To optimize the learning process, our approach incorporates fundamental physical principles directly into the loss function. This comprehensive training strategy not only harnesses the power of machine learning but also ensures that PIGNNs respect fundamental laws governing fluid dynamics.

RESULTS

The accuracy was validated in silico with different arterial networks, where PIGNNs achieved a coefficient of determination (R2) consistently above 0.99, comparable to numerical methods like the discontinuous Galerkin scheme. Moreover, with in vivo data, the prediction reached R2 values greater than 0.80, demonstrating the method's effectiveness in predicting flow and lumen dynamics using minimal data.

CONCLUSIONS

This study showcased the ability to calculate lumen area and blood flow rate in blood vessels within a given topology by seamlessly integrating 1-D blood flow with PIGNNs, using only blood flow velocity measurements. Moreover, this study is the first to compare the PIGNNs method with other classic Physics-Informed Neural Network (PINNs) approaches for blood flow simulation. Our findings highlight the potential to use this cost-effective and proficient tool to estimate real-time arterial pulse waves.

The effect of cardiac properties on arterial pulse waves: An in-silico study

This study investigated the effects of cardiac properties variability on arterial pulse wave morphology using blood flow modelling and pulse wave analysis. A lumped-parameter model of the left part of the heart was coupled to a one-dimensional model of the arterial network and validated using reference pulse waveforms in turn verified by comparison with in vivo measurements. A sensitivity analysis was performed to assess the effects of variations in cardiac parameters on central and peripheral pulse waveforms. Results showed that left ventricle contractility, stroke volume, cardiac cycle duration, and heart valves impairment are determinants of central waveforms morphology, pulse pressure and its amplification. Contractility of the left atrium has negligible effects on arterial pulse waves. Results also suggested that it might be possible to infer left ventricular dysfunction by analysing the timing of the dicrotic notch and cardiac function by analysing PPG signals. This study has identified cardiac properties that may be extracted from in vivo central and peripheral pulse waves to assess cardiac function.

Classification and regression of stenosis using an in-vitro pulse wave data set: Dependence on heart rate, waveform and location

BACKGROUND

Data-based approaches promise to use the information in cardiovascular signals to diagnose cardiovascular diseases. Considerable effort has been undertaken in the field of pulse-wave analysis to harness this information. However, the inverse problem, inferring arterial properties from waveform measurements, is not well understood today. Consequently, uncertainties within the estimation hinder the diagnostic application of such methods.

METHOD

This work contributes a publicly available data set measured at an in-vitro cardiovascular simulator, focusing on a set of input conditions (heart rate, waveform) and stenosis locations. Furthermore, a first attempt is undertaken to perform classification and regression on this data set using standard machine learning methods on features extracted from four peripheral pressure signals.

RESULTS

The locations of six different stenoses could be distinguished at high accuracy of 93%, where transfer function-based features outperformed features based solely on signal shape in almost all cases. Furthermore, regression on the stenosis position could be performed with a root mean square error of 2.4 cm along a 20 cm section of the arterial system using a shallow neural network. However, the performance difference between shape and transfer function features was not clear for this task.

CONCLUSION

The data set contains 800 measurements and allows investigating the influence of different heart boundary conditions, such as heart rate and waveform shape, on classification and regression tasks. Extracting features that minimise this influence is a promising way of improving the performance of these tasks.

A machine learning model for non-invasive detection of atherosclerotic coronary artery aneurysm

PURPOSE

Atherosclerosis plays a significant role in the initiation of coronary artery aneurysms (CAA). Although the treatment options for this kind of vascular disease are developing, there are challenges and limitations in both selecting and applying sufficient medical solutions. For surgical interventions, that are novel therapies, non-invasive specific patient-based studies could lead to obtaining more promising results. Despite medical and pathological tests, these pre-surgical investigations require special biomedical and computer-aided engineering techniques. In this study, a machine learning (ML) model is proposed for the non-invasive detection of atherosclerotic CAA for the first time.

METHODS

The database for study was collected from hemodynamic analysis and computed tomography angiography (CTA) of 80 CAAs from 61 patients, approved by the Institutional Review Board (IRB). The proposed ML model is formulated for learning by a one-class support vector machine (1SVM) that is a field of ML to provide techniques for outlier and anomaly detection.

RESULTS

The applied ML algorithms yield reasonable results with high and significant accuracy in designing a procedure for the non-invasive diagnosis of atherosclerotic aneurysms. This proposed method could be employed as a unique artificial intelligence (AI) tool for assurance in clinical decision-making procedures for surgical intervention treatment methods in the future.

CONCLUSIONS

The non-invasive diagnosis of the atherosclerotic CAAs, which is one of the vital factors in the accomplishment of endovascular surgeries, is important due to some clinical decisions. Although there is no accurate tool for managing this kind of diagnosis, an ML model that can decrease the probability of endovascular surgical failures, death risk, and post-operational complications is proposed in this study. The model is able to increase the clinical decision accuracy for low-risk selection of treatment options.

Determination of aortic pulse transit time based on waveform decomposition of radial pressure wave

Carotid-femoral pulse transit time (cfPTT) is a widely accepted measure of central arterial stiffness. The cfPTT is commonly calculated from two synchronized pressure waves. However, measurement of synchronized pressure waves is technically challenging. In this paper, a method of decomposing the radial pressure wave is proposed for estimating cfPTT. From the radial pressure wave alone, the pressure wave can be decomposed into forward and backward waves by fitting a double triangular flow wave. The first zero point of the second derivative of the radial pressure wave and the peak of the dicrotic segment of radial pressure wave are used as the peaks of the fitted double triangular flow wave. The correlation coefficient between the measured wave and the estimated forward and backward waves based on the decomposition of the radial pressure wave was 0.98 and 0.75, respectively. Then from the backward wave, cfPTT can be estimated. Because it has been verified that the time lag estimation based on of backward wave has strong correlation with the measured cfPTT. The corresponding regression function between the time lag estimation of backward wave and measured cfPTT is y = 0.96x + 5.50 (r = 0.77; p < 0.001). The estimated cfPTT using radial pressure wave decomposition based on the proposed double triangular flow wave is more accurate and convenient than the decomposition of the aortic pressure wave based on the triangular flow wave. The significance of this study is that arterial stiffness can be directly estimated from a noninvasively measured radial pressure wave.

A proof of concept study for machine learning application to stenosis detection

This proof of concept (PoC) assesses the ability of machine learning (ML) classifiers to predict the presence of a stenosis in a three vessel arterial system consisting of the abdominal aorta bifurcating into the two common iliacs. A virtual patient database (VPD) is created using one-dimensional pulse wave propagation model of haemodynamics. Four different machine learning (ML) methods are used to train and test a series of classifiers—both binary and multiclass—to distinguish between healthy and unhealthy virtual patients (VPs) using different combinations of pressure and flow-rate measurements. It is found that the ML classifiers achieve specificities larger than 80% and sensitivities ranging from 50 to 75%. The most balanced classifier also achieves an area under the receiver operative characteristic curve of 0.75, outperforming approximately 20 methods used in clinical practice, and thus placing the method as moderately accurate. Other important observations from this study are that (i) few measurements can provide similar classification accuracies compared to the case when more/all the measurements are used; (ii) some measurements are more informative than others for classification; and (iii) a modification of standard methods can result in detection of not only the presence of stenosis, but also the stenosed vessel.

Graphical Abstract

Machine learning for detection of stenoses and aneurysms: application in a physiologically realistic virtual patient database

This study presents an application of machine learning (ML) methods for detecting the presence of stenoses and aneurysms in the human arterial system. Four major forms of arterial disease-carotid artery stenosis (CAS), subclavian artery stenosis (SAS), peripheral arterial disease (PAD), and abdominal aortic aneurysms (AAA)-are considered. The ML methods are trained and tested on a physiologically realistic virtual patient database (VPD) containing 28,868 healthy subjects, adapted from the authors previous work and augmented to include disease. It is found that the tree-based methods of Random Forest and Gradient Boosting outperform other approaches. The performance of ML methods is quantified through the [Formula: see text] score and computation of sensitivities and specificities. When using six haemodynamic measurements (pressure in the common carotid, brachial, and radial arteries; and flow-rate in the common carotid, brachial, and femoral arteries), it is found that maximum [Formula: see text] scores larger than 0.9 are achieved for CAS and PAD, larger than 0.85 for SAS, and larger than 0.98 for both low- and high-severity AAAs. Corresponding sensitivities and specificities are larger than 90% for CAS and PAD, larger than 85% for SAS, and larger than 98% for both low- and high-severity AAAs. When reducing the number of measurements, performance is degraded by less than 5% when three measurements are used, and less than 10% when only two measurements are used for classification. For AAA, it is shown that [Formula: see text] scores larger than 0.85 and corresponding sensitivities and specificities larger than 85% are achievable when using only a single measurement. The results are encouraging to pursue AAA monitoring and screening through wearable devices which can reliably measure pressure or flow-rates.