Tech Innovations Mark Halfway Point of Yao's $1.1M Grant

Atrial fibrillation (AF) is an irregular heartbeat disorder caused by misfiring cardiac nerves that can increase the risk of stroke, heart failure, and even death. AF affects an estimated six million adults in the United States, and its prevalence increases with age, leading to overwhelming impacts on the American healthcare system.

AF can be treated with a type of surgery called ablation, wherein surgeons carefully scar heart tissue in multiple places to block the erroneous nerve signals. Unfortunately, the difficult and time-intensive procedure has a success rate just above 50 percent. Nearly 30 percent of patients still suffer from AF after multiple ablation treatments.

"Traditional 'one-size-fits-all' ablation procedures often fail to deliver reliable outcomes because the region causing the chaotic AF signals is not known in advance," said Bing Yao, the Dan Doulet Early Career Assistant Professor in the Department of Industrial and Systems Engineering. "Historically, there has not been a good strategy to search for individualized ablation targets in each patient."

Yao, an expert in physics-informed machine learning (ML), believes that next-generation technologies are the key to better patient outcomes. In 2023, she and Pennsylvania State University Professor Hui Yang were awarded a four-year, $1.1-million grant from the National Science Foundation–National Institutes of Health Smart and Connected Health program to develop an ML-enabled system that would turn individual patient data into personalized AF ablation recommendations.

"Computer simulation offers a powerful approach to addressing the challenges of AF ablation," Yao said. "It provides an effective framework for integrating diverse experimental findings and testing multiple surgical paths, which is simply not possible with animal or human subjects."

Now halfway through their grant, the researchers have created a physics-based, 3D simulation model of the heart that can accurately predict how electrical signals spread through cardiac tissue. They have also developed several software innovations allowing their model to make recommendations for sequential ablations, adjusting to changes the heart undergoes throughout treatment.

Electrical Waves in 3D

Over the last two years, Yang and members of his lab at Penn State built simulations of heart tissue based on publicly available medical imaging data provided by the Visible Human Project at the US National Library of Medicine.

Yao and her students at the University of Tennessee then joined Yang's team in transforming the imaging data and a network-based geometric model of the heart into a detailed, physics-based 3D mesh that enables reliable computer simulations of how electrical signals spread through the heart.

"A model that relies only on data can sometimes make inaccurate predictions—for instance, by suggesting signals move in ways that are physiologically impossible," Yao said. "Incorporating physical rules of how electrical signals naturally spread through the heart makes our approach more accurate in predicting cardiac electrical activity, helps us better understand irregular heart rhythms, and ultimately supports medical doctors in making more reliable decisions."

The team then had to overcome an even bigger challenge: teaching their model to simulate and optimize sequential treatments (ablations that are made in a particular order).

"Cardiac response and electrical signals evolve over time as clinicians take actions during the operation," Yao explained. "Traditional decision-making and optimization ML methods struggle with complex geometries, like the irregular shapes and curved surfaces of cardiac atria, and with spatiotemporal dynamics that evolve over time."

Yao's team overcame that challenge by developing new simulation and optimization tools that integrate advanced ML techniques with physics-based simulations. The new suite includes a model that can simulate a range of future surgical possibilities on complex 3D structures as well as a decision tree that simulates the effect of each possible ablation on both the heart itself and downstream surgical recommendations.

"Together, these innovations enable us to systematically study cardiac responses to various surgical procedures, thereby informing and optimizing sequential decision-making strategies in AF ablation to restore a normal rhythm," Yao said.

Ensuring Clinical Relevance

Throughout the long process of software development, Yao and Yang have worked closely with their clinical collaborators, Chief Medical Information Officer Christopher DeFlitch and Associate Professor Soraya Samii from Penn State Health, and Fabio Leonelli, a staff cardiologist at James A. Haley Veterans' Hospital in Tampa, Florida.

Regular meetings with DeFlitch, Samii, and Leonelli help the engineers ensure that their models address real clinical needs and cover as many patients as possible.

For example, Yao and Yang's initial model only included arrhythmias triggered by extra electrical activity in the left atrium of the heart, which is the most common cause of AF.

The medical professionals reminded them of another important form of AF, arising when heterogeneous (uneven) scar tissue disrupts normal conduction pathways, which the engineers are currently working to add to their simulation.

In a recent meeting, the medical team also expressed concerns that the models' complex outputs might be difficult for clinicians to interpret. As a result, Yao and Yang are devoting significant effort over the next year to visualization improvements that will make their models' output and recommendations clearer to clinicians.

"Our medical collaborators' perspective has been instrumental, ultimately making our predictions more realistic and clinically relevant," said Yao. "We plan to keep working closely with our medical team, actively soliciting their feedback and iteratively refining how the results are presented. This will make our tools more accessible and practical for real-world use."

Provided by University of Tennessee at Knoxville