Integrating Generative AI with Biomechanics Modeling to Advance Ventricular Disease Diagnosis and Personalized Treatment - Abstract: Ventricular-related diseases, including hypertrophic cardiomyopathy, heart failure, and dilated cardiomyopathy, significantly contribute to the global burden of cardiovascular disease (CVD), which claims around 17.9 million lives annually. This underscores the urgent need for enhanced prevention, diagnosis, and personalized treatment strategies. Personalized computer-based heart modeling offers a powerful, non-invasive approach to understanding cardiac function by revealing complex structure-function relationships, are widely used in cardiac research, disease analysis, device design, and treatment planning. Accurately predicting heart behavior is challenging due to the difficulty in measuring cardiac tissue material properties and determining the stress-free configuration from stressed myocardium in medical images. Current methods, such as finite element analysis (FEA) for predicting biomechanical outputs and inverse FEA for identifying material properties, are robust but often slow, limiting their clinical application and the advancement of personalized cardiovascular care. This proposal aims to develop and validate an end-to-end generative AI framework that efficiently and accurately predicts the biomechanical outputs (e.g., motion and stress) of the passively deforming bi-ventricular myocardium while identifying material properties and determining the stress-free configuration using patients' 3D imaging data and clinical measurements. We hypothesize that generative AI models trained on this data can serve as accelerated emulators for both forward and inverse FEA tasks, enhancing personalized diagnosis and treatment of ventricular diseases. This 3-year project focuses on passive mechanics related to the late-diastolic stage, using 3D heart imaging and clinical data from normal patients to train the generative AI framework before extending it to incorporate various diseases. Aim 1 will develop a generative AI-based FEA emulator to efficiently predict passive bi-ventricular biomechanical outputs, enabling faster patient-specific heart function analysis. Aim 2 will create an end-to-end generative AI framework to determine stress-free configurations and optimal material parameters from 3D imaging and clinical measurements, incorporating uncertainty quantification for robust parameter estimation. Aim 3 will establish a fine-tuning mechanism to adapt generative AI models for accurately predicting biomechanical outputs in diseased hearts, integrating patient- specific disease characteristics. Besides, this project will enhance interdisciplinary student involvement in medical research through hands-on learning and collaboration with top medical schools, fostering interest in health-related careers. This project develops a framework for personalized heart modeling that enhances clinical decision-making and has the potential to save lives. It innovatively integrates generative AI with biomechanical modeling, using a single neural network to address both forward and inverse FEA problems for rapid predictions that outperform traditional methods. Our close collaboration among experts in cardiovascular modeling, medical imaging, and clinical care ensures project success.
