Neural operator learning of heterogeneous mechanobiological insults contributing to aortic aneurysms

Abstract

Thoracic aortic aneurysm (TAA) is a localized dilatation of the aorta resulting from compromised wall composition, structure, and function, which can lead to life-threatening dissection or rupture. Several genetic mutations and predisposing factors that contribute to TAA have been studied in mouse models to characterize specific changes in aortic microstructure and material properties that result from a wide range of mechanobiological insults. By contrast, assessments of TAA progression in vivo are largely limited to measurements of aneurysm size and growth rate. It has been shown, however, that aortic geometry alone is not sufficient to predict the patient-specific progression of TAA but computational modeling of the evolving biomechanics of the aorta could predict future geometry and properties from initiating insults. In this work, we present an integrated framework to train a deep operator network (DeepONet)-based surrogate model to identify contributing factors for TAA by using synthetic finite element-based datasets of aortic growth and remodeling (G&R) resulting from prescribed mechanobiological insults. For training data, we investigate multiple types of TAA risk factors and spatial distributions within a computationally efficient constrained mixture model to generate axial-azimuthal maps of aortic dilatation and distensibility. The trained network is then capable of predicting the initial distribution and extent of the insult from a given set of dilatation and distensibility information, which in turn can be used to determine subsequent aortic geometry and mechanical properties. Two DeepONet frameworks are proposed, one trained on sparse information and one on full-field grayscale images, to gain insight into a preferred neural operator-based approach. Performance of the surrogate models is evaluated through multiple simulations carried out on insult distributions varying from fusiform (analytically defined) to complex (randomly generated). We show that this integrated continuous learning modeling approach can predict the patient-specific mechanobiological insult profile associated with any given dilatation and distensibility map with a high accuracy, particularly when based on full-field images. Our findings demonstrate the feasibility of

Physics-Informed Deep Neural Operator Networks

“…-The application of neural operators in life sciences is endless. For example, approaches developed in [58,24] show the application of DeepONet for accu-rate prediction of aortic dissection and aneurysm, which is patient-specific and hence could provide clinicians sufficient time for planning a surgery. -DeepONet can be used for accelerating climate modeling by adding learned high-order corrections to the low resolution (e.g., 100 Km) climate simulations.…”

“…A schematic representation of this architecture is shown in Figure 2. In [24], a DeepONet framework based on multiple input functions is proposed that encompasses two different DNNs (CNN and FNN) as branch networks. The architecture uses grayscale images of systolic and diastolic geometry (in CNNs) along with patientspecific information (such as hypertension in FNN) to predict the initial distribution and extent of the mechanobiologial insult in a patient with thoracic aortic aneurysm.…”

High‐Precision and Fast Prediction of Regional Wind Fields in Near Space Using Neural‐Network Approximation of Operators

“…Lu et al, 2021 proposed a neural operator network, DeepONet, which achieved a leap from the general approximation theory of neural operators to practice. Subsequently, DeepONet has been applied in classical fields such as life science (Goswami et al, 2022), material science (Oommen et al, 2022)and physics (Cai et al, 2020;Leoni et al, 2021;Yin et al, 2022). DeepONet is composed of the branch net and the trunk net.…”

Refined Short‐Term Forecasting Atmospheric Temperature Profiles in the Stratosphere Based on Operators Learning of Neural Networks

“…Fortunately, scientific machine learning is a new research paradigm emerging in recent years, and it is also an important frontier cross field between AI and basic science. It has accelerated the knowledge discovery and efficient scientific computing of complex physical scenes, and has developed rapidly in physics, chemistry, life science and other fields (Goswami et al, 2022;Leoni et al, 2021;Yin et al, 2022). Lu et al (2021) proposed a deep operator network (DeepONet) framework, introduced in the seminal paper (Lu et al, 2021), with utilizing the universal approximation theorem for operator (Chen & Chen, 1995).…”