• Citer

Paroxysmal Atrial Fibrillation Onset Forecast and Risk Identification During Sinus Rhythm: A Machine Learning Approach

par Gilon, Cédric
Président du jury Decaestecker, Christine
Promoteur Bersini, Hugues
Co-Promoteur Carlier, Stéphane
Publication Non publié, 2024-02-28

Thèse de doctorat

• Accès en ligne
• Détails
• Contenu
• Statistiques

Résumé :

Atrial fibrillation (AF) is one of the most common heart rhythm disorders. Patients affected by this condition have a fivefold increased risk of stroke. During AF, irregular atrial contractions disrupt the normal cardiac cycle, resulting in irregular heartbeats. This thesis proposes a machine learning (ML) approach to predict the onset of paroxysmal AF episodes using electrocardiograms (ECG). A new database of long-term ECG recordings from patients with AF, annotated by a cardiologist, was created. ML models were trained on these data to predict the onset of AF. We showed that model performance improved as the prediction got closer to the onset of AF. Models using heart rate variability and RR intervals performed better than those using raw ECG signals. We then selected ECG recordings from healthy people and added them to the database. These additional recordings made it possible to compare the sinus rhythm of healthy people and people with AF. ML models were able to identify the signatures of AF within sinus rhythm, suggesting the possibility of improving AF screening and treatment strategies using ML techniques.