Le but de ce projet est de développer un algorithme robuste et fiable pour la détection des arythmies cardiaques durant la vie de tous les jours à l'aide d'un simple capteur optique porté au poignet (p. ex. dans une montre intelligente). Cette étude présente un grand intérêt clinique, car les montres intelligentes actuellement disponibles pour la détection d'arythmies cardiaques ne distinguent que le rythme cardiaque normal de l'anormal et ne prennent pas nécessairement en compte d'autres arythmies que la FA.
Nous proposons de nous attaquer à ce problème en créant une grande base de données de signaux ECG et optique via des études cliniques. Nous commencerons par réaliser une étude dans des conditions contrôlées sur 100 patients souffrant d'arythmies cardiaques (incluant au moins 10 différents types d’arythmies). Cette étude nous aidera à développer notre algorithme pour distinguer différents types d'arythmie en utilisant des outils de traitement de signaux avancées. Dans un second temps, nous évaluerons notre algorithme dans des conditions ambulatoires sur plus de 200 patients portant une montre intelligente au quotidien sur plusieurs jours.
Cardiac arrhythmias have a prevalence of 3.2-6.6% in elderly population (65 to 73 years). Atrial fibrillation (AF), the most common arrhythmia, affects more than 10 million individuals in Europe, and is expected to increase, up to 17.9 million people by 2060. The associated healthcare costs worldwide add up to several 100 billion USD. People suffering from cardiac arrhythmias are prone to syncope, fatigue or even sudden death from ventricular tachycardia or fibrillation. Because of the asymptomatic and intermittent nature of several arrhythmias in their early stages, they are often diagnosed too late, when a patient is hospitalized for complications such as heart failure or stroke. Long-term, continuous ambulatory monitoring of cardiac function is therefore crucial for early arrhythmias detection and classification in order to provide proper diagnosis, treatment and reduce risks of complications. However, cardiac monitoring with current ambulatory electrocardiogram (ECG) devices are cumbersome and limited to a couple of days. Recent advances in photoplethysmography (PPG) have made this technology a promising tool to detect irregular heart rate in ambulatory settings. Because of its non-invasiveness and its unobtrusiveness (e.g. measurement via a simple smartwatch), PPG is promising for long-term ambulatory screening of arrhythmias and more specifically for intermittent AF and non-sustained arrhythmias. The main goal of the present project is to develop a robust and reliable algorithm to detect and classify arrhythmias in everyday life with a PPG-based sensor. Important issues need to be addressed to efficiently diagnose arrhythmias in ambulatory settings. First, correct discrimination of different types of arrhythmias require in-depth investigations since certain can easily be confounded (e.g. intermittent AF vs frequent ectopic beats). This explains why current commercial products do not claim to diagnose AF but rather detect abnormal rhythm. For this purpose, specificities of PPG waveforms during arrhythmia occurrences will be analyzed and relevant features will be selected to optimize the classification of each type of arrhythmia. Second, PPG signals are known to be sensitive to motion artifacts or device-skin pressure, which may confound PPG-based arrhythmia detection. To address this issue, the present project will investigate methods to increase the robustness of arrhythmia detection against artifacts, which occur naturally in ambulatory recordings. This research will investigate the PPG signal morphology related to physiological and non-physiological changes as well as the relevance of PPG-based features to discriminate arrhythmias. To this end, clinical data will be recorded on patients by using a PPG wrist monitor simultaneously with a 12-lead ECG and intracardiac signals at the university hospital in Lausanne (CHUV). Then, ambulatory data will be recorded on patients using the same PPG and a Holter device at the university hospital in Bern (Inselspital). Our project is divided into four work packages (WP1 to WP4) with the following objectives: WP1) To collect comprehensive and high-quality datasets including arrhythmia episodes of at least 10 different types. To do so, we perform three clinical studies on arrhythmia patients: one study during catheter ablation procedures (controlled conditions), and two long-term studies in ambulatory settings (experimental conditions); WP2) To gain a deeper understanding of the PPG signal and improve the design of dedicated PPG wrist devices;WP3) To develop a robust signal quality index to reliably detect and reject artifacts occurring in ambulatory settings;WP4) To investigate advanced PPG signal features to better discriminate between the different types of arrhythmias, using advanced machine learning techniques. The outcome of the present use-inspired project is a novel framework to reliably detect and classify arrhythmias in ambulatory settings based on PPG. This development paves the way to an effective low-cost diagnostic tool for early screening of arrhythmias. Considering the predicted increase of arrhythmia prevalence, our solution is of crucial benefit for today’s ageing populations and aims to reduce the steadily rising of healthcare costs. Furthermore, the high availability of PPG (e.g. via a smartphone camera) makes our solution accessible to the broader public, including low-income countries.
Last updated:19.03.2022
