Computational Modeling of Ultrasound Impulse Induced Wave Velocities in the Fibrotic Atria

Abstract

Atrial fibrillation (AF) is the most prevalent cardiac arrhythmia. AF is closely linked to atrial fibrosis, where pathologic non-conductive fibrotic tissue disrupts electrical pathways, promoting spontaneous and reentrant arrhythmic activity. Assessing atrial fibrosis is important for better AF diagnosis and treatment. Realtime assessment during therapeutic ablation procedures could improve treatment outcomes. Fibrosis manifests in patterns including compact, interstitial, diffuse, and patchy, each affecting heart structure and electrical conduction uniquely. Current fibrosis assessment methods have limitations including low resolution, non-realtime, and accessibility issues, highlighting the need for more precise and accessible evaluation methods. This dissertation investigates the relationship between simulated atrial fibrosis and mechanical wave propagation to explore the potential of shear wave elasticity imaging (SWEI) in evaluating local fibrosis. SWEI has been used previously in the clinic and in real-time during ablation procedures to perform lesion evaluation. The study uses finite element modeling (FEM) to simulate ultrasound-induced mechanical wave propagation across all common forms of atrial fibrosis, processing the data using two methods. First, Radon sum-derived velocities were used to characterize the relationship between wave velocity and fractional fibrotic content. Second, deep learning methods were applied to displacement-time signals to predict either fractional content or fibrosis type. The work is presented in three parts. Part 1 describes FEM simulations of compact fibrosis, establishing a significant linear correlation between fibrosis content and mechanical conduction velocities. Part 2 expands the evaluation to include interstitial, diffuse, and patchy fibrosis types, confirming the efficacy of shear wave velocity in quantifying fibrosis content and exploring the importance of background stiffness level and fibrosis type. Across all types and background stiffness levels, a correlation coefficient of 0.86 between mechanical wave velocity and fractional fibrotic content, with a Mean Square Error (MSE) of 0.0272 was found with significant variability caused by the background stiffness level. Part 3 employs convolutional neural networks (CNNs) to evaluate fibrosis ratios and classify patterns using raw mechanical wave displacement data. This method demonstrated high accuracy with an MSE of 0.0002 when predicting the fibrotic fraction and an overall accuracy of 90.7% when classifying the fibrosis type. Overall, this dissertation demonstrates the potential and some limitations of using SWEI to evaluate local fibrosis, potentially benefiting clinical practices by providing real time assessments of atrial myocardial fibrosis.