Are you an EPFL student looking for a semester project?
Work with us on data science and visualisation projects, and deploy your project as an app on top of GraphSearch.
Publication
Are you an EPFL student looking for a semester project?
Work with us on data science and visualisation projects, and deploy your project as an app on top of GraphSearch.
Atrial fibrillation (AF) is the most common cardiac arrhythmia; it will affect one in four adults worldwide in their lifetime. AF has serious consequences, including drastically increased risk of stroke. Catheter ablation surgery is an established treatment option for persistent AF, that is, an advanced form of AF that does not terminate spontaneously. However, while estimates vary widely by study, health care center, and exact ablation strategy, a common figure for post-ablation AF recurrence for patients with persistent AF is 50%. There is therefore great interest in developing pre-ablation screening to select persistent AF patients most likely to benefit from undergoing the surgery. The clinical diagnosis of AF requires recording an electrocardiogram (ECG). This signal provides a wealth of information about the function of a patient's heart. Many studies have investigated the ECG for calculating digital biomarkers related to ablation outcomes. These studies have resulted in the development of signal processing algorithms that aim to quantify AF disease complexity from a patient's recorded ECG in an electrophysiologically interpretable way. In this thesis, we propose, test, and evaluate digital predictive biomarkers obtained through signal processing of ECG recorded pre-ablation for noninvasive selection of patients most likely to benefit long term from ablation. We additionally track the progression of the biomarkers throughout ablation to see whether they evolve differently according to procedural ablation outcome. We used standard 12-lead ECG recorded in a persistent AF patient cohort before and during catheter ablation. We applied two distinct analysis approaches. In the first, fibrillatory wave extraction was applied to the ECG recorded on several precordial leads to allow for direct analysis of the atrial component. We then calculated spatiotemporal biomarkers correlated with AF complexity over the course of catheter ablation surgery. In the second approach, RR-interval sequences were extracted from ECG lead II, and heart rate variability (HRV) analysis was performed, to quantify the relationship between HRV and procedural and clinical ablation outcomes. In both approaches, we found that higher levels of AF organization measured pre-ablation using the ECG biomarkers were associated with AF termination by catheter ablation and long-term maintenance of sinus rhythm thereafter. We then recorded body surface potential mapping (BSPM) ECG including 252 leads distributed across the torso. Atrial fibrillatory wave extraction was again applied to allow for a spatiotemporal analysis of all BSPM leads. We propose two novel BSPM biomarkers inspired by spatiotemporal signal processing methods that quantify regularity in space and time of the high-dimensional ECG and analyze their association with clinical ablation outcomes. It was found that increasing regularity in the high dimensional BSPM signal was associated with AF termination and long-term maintenance of sinus rhythm. These biomarkers were also shown to be performant for predicting ablation outcome.The results of this thesis demonstrate that ECG biomarkers were predictive of catheter ablation outcomes for the persistent AF populations tested. The use of such biomarkers could help guide treatment planning for both doctors and patients and would aid in the creation of an informed and transparent treatment pathway tailored to each patient's individual needs.