| On 2022-09-13 at 15:00:00 (Brussels Time) |
Abstract
Heart diseases are one of the leading causes of death in the world, representing about 32% of all deaths globally. Atrial fibrillation (AF) is the most common heart arrhythmia. It is estimated to affect 37 million patients worldwide and 150 000 in Belgium. The prevalence increases with the age of the patient, it is estimated between 1% and 2% for the world population but rise to 20% for the population above 80 years old. AF is linked to an increased risk of stroke, heart failure and death. The evolution of the diseases occurs in three main states: paroxysmal, persistent, and permanent. During paroxysmal AF, crisis starts and stops with no known warning sign.
In our work, we are focusing on two tasks: AF detection and AF onset forecast to answer two questions: “Could ML model be used to accurately detect ongoing AF episodes in ECG recordings to assist the disease screening?” and “Is there any information in the heartbeat series preceding the AF onset that would forecast this onset?”. During this presentation, I will present the current state of our research, including the review and the reproducibility status of the state-of-the-art, the construction of a new database for AF detection and onset forecast. I will go also through the heart rate variability features used in the literature and in our work. Finally, I will present our last results with ML and DL models.
