UPCOMING TALKS:

September 9, 2026:

Title: "LEARNING HEART DISEASE: MODELING HEART AND HEART VALVE DISEASE USING NOVEL GPU BASED FINITE ELEMENT APPROACHES"

Speaker: Prof. Michael Sacks at UT Austin

Zoom Link: https://utexas.zoom.us/j/86761043788?pwd=QLrw4lmwH4NIpKrNb7mvznQGjrHAi2.1&from=addon

Abstract: Computational models of cardiac mechanics can serve as vital predictive tools to assist our understanding of cardiac pathologies. For example, we have developed a novel pipeline to convert pre-operative clinical images of the mitral valve into predictive patient specific models capable of accurately predicting the post-repair state.  However, the slow computational speed of traditional finite element models limits their practical clinical application. We have previously demonstrated the efficacy of the neural network finite element (NNFE) approach for predicting cardiac mechanics of an idealized left ventricle within clinically relevant timeframes without compromising accuracy. In this work, we demonstrate that the NNFE approach can 1) predict the complex deformation patterns arising due to the presence of infarct tissue, 2) predict the effects of infarct occurring in varying locations and of varying sizes, and 3) capture the compensatory mechanisms of the healthy tissue that displays a higher contractile level in order to maintain stroke volume. We will next show how this approach can be connected to clinical pipelines via 0D/1D circulatory models.  These results represent an important first step towards high-speed patient specific cardiac disease modeling.

Bio: Professor Sacks is a world authority on cardiovascular modeling and simulation, particularly on developing patient-specific, simulation-based approaches for the understanding and treatment of heart and heart valve diseases. His research is based on multi-scale modeling, quantification, and simulation of the biophysical behavior of the constituent cells and tissues and translation to the organ level in health, disease, and treatment. For example, he has developed novel non-invasive methods to quantify pre- and post-surgical state of the mitral valve from pre-surgical clinical images. These methods utilize novel GPU based software technologies to create high-speed digital twin software tools for clinical decision making.  At a more fundamental level, Dr. Sacks has determined the how local stress environments of heart valve interstitial cells alter their biosynthetic responses in the context of altered heart organ-level responses. His research also includes developing novel neural network based cardiac models to simulate growth and remodeling of the heart at greatly accelerated speeds. Dr. Sacks is also currently active in modeling replacement heart valve materials and the resulting valve designs using machine learning based approaches.


 

PAST TALKS:

April 23, 2026: 

Title: "Efficient Medical Image Segmentation Across the Pipeline"

Speaker:  Radu Marculescu, Ph.D., Professor, Department Electrical and Computer Engineering, UT Austin

Abstract:  Medical image segmentation is a central problem in AI for health, but improvements in accuracy often come with growing computational and memory costs. This talk presents a recent line of work on efficient segmentation across the pipeline, showing how efficiency can be built directly into decoding, 3D network design, and multi-scale prediction fusion. Rather than treating efficiency as a post hoc constraint or compression step, I will argue that it can serve as a core design principle. By rethinking where computation is most valuable, we can build segmentation models that remain accurate while also improving scalability, inference speed, and practical deployability across a range of real-world medical imaging settings.
Bio: Radu Marculescu is a Professor and the Laura Jennings Turner Chair in Engineering in the Department of Electrical and Computer Engineering at The University of Texas at Austin. Between 2000-2019, he was a Professor in the Electrical and Computer Engineering department at Carnegie Mellon University. His current research focuses on developing AI/ML algorithms for computer vision, bioimaging, and Internet-of-Things (IoT) applications. He is an IEEE Fellow, an ACM Fellow, and an AAAS Fellow. More info available here: http://radum.ece.utexas.edu.


April 9, 2026: 

Title: "Adaptive Radiation Therapy at Scale: Clinical Necessity or Selective Application?"

Speaker: Ergys D. Subashi, Ph.D., Associate Professor, Department of Radiation Physics, MD Anderson Cancer Center

Abstract: Adaptive radiation therapy improves treatment precision by accounting for geometric and functional changes in patient anatomy, enabling better target coverage and sparing of normal tissue. However, routine adaptation is challenging due to workflow complexity, time constraints, staffing demands, and stringent quality assurance requirements. This talk reviews our work on implementation of adaptive MR-guided radiotherapy and discusses how AI-driven tools—such as auto-contouring, rapid planning, and decision support—can streamline workflows to support a safe and scalable implementation of daily adaptive radiotherapy.

Bio: Dr. Ergys D. Subashi is an associate professor in the Department of Radiation Physics at MD Anderson Cancer Center. His work is focused on MRI-guided adaptive radiation therapy, with an emphasis on developing imaging methods that improve tumor delineation, treatment planning, and real-time monitoring of anatomical changes during therapy. His research includes techniques that characterize motion and functional heterogeneity in tumors, enabling more precise and individualized radiation therapy. In his clinical role, Dr/ Subashi focuses on treatment planning, workflow optimization, and quality assurance for MR-linac systems, addressing uncertainties related to patient motion, imaging performance, and treatment delivery. His work aims to enhance the safety, accuracy, and robustness of MRI-guided radiotherapy, particularly in anatomically complex regions such as the abdomen and pelvis.



AIHealthTalk: 03/26/26 - "Trustworthy Health AI: Challenges & Lessons Learned"
 

AIHealthTalk: 03/12/26 - "Generative magnetic resonance multitasking: patient-specific AI models for high-dimensional imaging"
 

AIHealthTalk: 02/26/26 - "Real-World Data to Real-World Evidence with some AI: Successes, Challenges, and Opportunities"
 

AIHealthTalk: 02/12/26 - "Knowledge-Informed Weakly-Supervised Deep Learning Models for Cancer Applications" 
 

AIHealthTalk: 01/29/26 - "Enhancing GI Tract Cancer Diagnosis Through Generative Models and Vision-based Robotic Tactile Sensing" 
 

2025 

AIHealthTalk: 11/06/25 - Semantics in Medicine: Expert, Data, and Application Perspectives

AIHealth Talk: 10/23/25 - Predicting Long Term Mortality in COPD Using Deep Learning Imaging Markers

AIHealthTalk:10/9/25 - Using Large Language Models to Simulate Patients for Training Mental Health

 AIHealthTalk: 09/25/25 - PanEcho: Toward Complete Al-Enabled Echocardiography Interpretation

AIHealthTalk: 09/11/25 - Clinical Deployment of AI:From Single Models to Compound Agentic Systems

April 10, 2025: Na Zou, Assistant Professor, University of Houston
Exploring and Exploiting Fairness in AI/ML: Algorithms and Applications

April 24, 2025: Edison Thomaz, Associate Professor and William H. Hartwig Fellow, Electrical and Computer Engineering, UT Austin
Identifying Digital Biomarkers of Cognitive Impairment from Real World Activity Data


Past Talks: 

Fall 2024

Nov. 14: Ziyue Xu, NVIDIA Health
Flexible Modality Learning: Modeling Arbitrary Modality Combination via the Mixture-of-Experts Framework

Oct 31: Greg Durrett, Associate Professor, The University of Texas at Austin
Specializing LLMs for Factuality and Soft Reasoning

Oct 17: Akshay Chaudhari, Stanford University
Towards Multi-modal Foundation Models for 3D Medical Imaging

Oct 3: Tianlong Chen, UNC 

Sept 19: Carl Yang, Assistant Professor of Computer Science, Emory University
KG-LLM Co-Learning for Health