Definition of a sensitivity profile for drug treatment and identification on clinical biomarkers in atrial fibrillation

Résumé

Atrial Fibrillation (AF) is one of the most common arrhythmias in clinical practice and thus far, the electrophysiological mechanisms underlying its initiation and maintenance are not fully understood. The study of such mechanism including clinical information, computational models and artificial intelligence (AI) algorithms that enable the identification of new patterns for the personalization of the treatments is key to unveil the characteristics of the arrhythmia. At the present, the treatment of choice for AF patients with higher effectiveness has proved to be cardiac ablation. This invasive procedure uses a catheter to ablate or burn the area of the cardiac tissue that is responsible for the maintenance of the arrhythmia. In order to find this specific area, it is indispensable to perform an electrophysiological study to evaluate the intracavitary electrical signals. In the computational field, several studies have presented personalized approaches that aim to stablish a complimentary platform for ablation planning. In this area, electrocardiographic imaging has also been used for the stratification and prior characterization of patients before the ablation procedure. Finally, observational studies enable the characterization of the AF population, enabling to collect, not only electrophysiological data but clinical biomarkers that can be related with the prognosis of the patients. Due to all the information produced during this type of studies, AI has been recently incorporated into these studies, with the main objective of identifying patterns or biomarkers that are able to characterize these patients including all the collected information. In addition, prediction algorithms, that allow to estimate the success of the treatment and prognosis of the patient have also been developed. For this purpose, these three fields of study were explored in this thesis. First, computational simulations using a population of models were performed to evaluate arrhythmia inducibility and maintenance under different scenarios. Due to the variability introduced in the population of models in combination with different drugs, AI algorithms were applied to extract patterns that identified the most proarrhythmic profiles. Secondly, personalized simulations were performed in a cohort of patients including their anatomical cardiac geometries and considering different arrhythmic scenarios. These experiments were achieved with a lowered computational costs and included the identification of a biomarker extracted from the simulation analysis that characterized the activity in the pulmonary vein area and evaluating it with the 12-month ablation outcome. Finally, the STRATIFY-AF observational study was analyzed, using the ECGi information from the patients combined with clinical infor-mation. As a results, a stratification score was obtained to predict the most successful treatment for each of the patients. The results presented in this thesis illustrate that the combination of in silico technologies with clinical data and processing algorithms can be of great utility to further investigate the arrhythmic mechanisms.