3IA PhD/Postdoc Seminar #31

Flash presentation

The extremal range: a local statistic for studying spatial extremes

Abstract:
In this talk, I will introduce the extremal range, an extreme value theory tool that we developed for studying the spatial extent of extreme events of random fields over R^d. Conditioned on exceedance of a high threshold at a location s, the extremal range at s is the random variable defined as the smallest distance from s to a location where there is a non-exceedance. I will show how this notion is connected to well-known tools in stochastic geometry literature, and how it provides useful summary statistics for climate scientists.

Identifying spatiotemporal dispersion in catheter ablation of persistent atrial fibrillation: a comparative study of machine learning techniques using both real and realistic synthetic multipolar electrograms

Abstract:
Atrial fibrillation (AF) is a common heart condition affecting the elderly, making it a growing public health concern. Catheter ablation (CA) is acknowledged as the most effective long-term treatment for persistent AF. A novel CA approach centers on spatiotemporal dispersion (STD). STD areas are defined as clusters of intracardiac electrograms (EGMs) displaying interelectrode time and space dispersion. STD patterns are identified visually by interventional cardiologists using the Pentaray multielectrode mapping catheter, following established guidelines at the block. Despite a growing number of patients eligible for CA, determining the most effective ablation strategy remains elusive due to the reliance on visually interpreted STD patterns, presenting challenges in standardization because of operator variability and learning curves, leading to inconsistencies in the process. The bad quality of the data and labeling errors make machine learning (ML) techniques for STD classification difficult to apply, posing challenges to the development of a support decision system to help cardiologists in this difficult task. A clear ML methodological analysis is lacking in the state of the art. To increase the reliable data amount, we have adopted an in silico approach, whereby synthetic EGMs are generated from simplistic in silico 2D models using the OpenCARP simulation software. The work presented in this talk opens the path to new understanding on AF, using physiological signal analysis and ML on real and synthetic EGM data for automatic identification of STD patterns. We also propose a mathematical algorithm modeling the reasoning process of the interventional cardiologist during STD-based CA. This algorithm uses real data solely for testing, improving interpretability and explainability over the ML approach.

Open discussion on the two contributions