CEI Seminar Series - Archive
CEI Seminar Series - Archive
CEI 2022 Seminar Series
Moderator: Rob MacLeod
Presenters:
Dr. Natalia Trayanova, John Hopkins University
Title: AI-powered computational cardiology
Dr. Steven Niederer, King’s College London and Guys’ and St Thomas’ Trust
Title: Clinical Applications of Cardiac Models
Advanced imaging, modeling and simulation are increasingly being used to interpret clinical data and guide therapies. The resources required to make these models needs to be reduced to allow more people to engage in this field and to allow these models to be translated into clinical practice. I will give a brief overview of our translational projects at King’s, the tools that we are creating to support their development, and the public modeling resources we are creating for the community.
Dr. Marina Strocchi, King's College London
Title: ECGI Data Analysis in Brugada Syndrome
This talk will present results from three ECGi studies in collaboration with the clinical team at King's College London and at the University of Sydney. In these studies, we looked at left bundle branch block patients and Brugada syndrome patients. We used ECGi to quantify local activation and repolarisation times, to improve our understanding of mechanisms underlying the pathology and how patients respond to treatment.
Moderator: Dr. Linwei Wang, Rochester Institute of Technology
Presenters:
Rutger van de Leur, University Medical Center Utrecht -- Explainable AI for the ECG
Deep neural networks (DNNs) show excellent performance in interpreting electrocardiograms (ECGs), both for conventional ECG interpretation and for novel applications such as detection of reduced ejection fraction. Despite these promising developments, clinical implementation is severely hampered by the lack of trustworthy techniques to explain the decisions of the deep learning algorithm to clinicians. Especially, currently employed heatmap-based methods have shown to be inaccurate. We will discuss a novel approach that is inherently explainable and uses an unsupervised variational auto-encoder (VAE) to learn the underlying factors of variation of the ECG (the FactorECG) in a database with 1.1 million ECG recordings.
Dr. Maxime Sermesant, Inria IHU Liryc Université Côte d’Azur -- Learning to improve cardiac models
Physics-based and data-driven approaches both progressed tremendously during the last decade. There is now increasing research activity at the interface between these two scientific areas. In particular, deep learning can be used to efficiently predict the solution to a given PDE. However, in most of these approaches, the learned model can at best achieve similar accuracy as the physics-based model used to train it. Therefore, the benefit can be reduced to a better computational efficiency, with the risk of a loss of realism for conditions far from the training data. I will present an approach to learn cardiac electrophysiology models and recent results on a new framework to learn model error from data. This has the potential to produce a final dynamical model combining simulation and learning that is closer to the reality than the original biophysical model.