Portable Measurement Methods Combined With Artificial Intelligence in Detection of Atrial Fibrillation

Overview

In Western countries, every sixth person in their lifetime and 15,000 people in Finland have a new stroke each year. About every fourth stroke is based on cardiac embolism. Atrial fibrillation (AF) is the most common arrhythmia that increases the risk of thromboembolic complications, such as stroke. It may cause formation of thrombi in the left atrium with ensuing embolization in the cerebral and peripheral circulation. AF is often asymptomatic and paroxysmal. Thus, the diagnosis of AF is often challenging. A new onset AF is usually treated with cardioversion (CV), in which the abnormal rhythm is converted back to sinus rhythm (SR). However, a long-lasting AF (\>48 hours) is associated with risk of stroke. Therefore, the duration of AF needs to be known before a CV can be performed. This study evaluates the ability of novel customer-targeted heart measuring devices to detect rhythm change and short AF episodes. Moreover, novel biomarkers will be analyzed from the blood samples of AF patients and their suitability to estimate the duration of AF will be evaluated. The research will be accomplished in cooperation with the Kuopio University Hospital Emergency Department, the Heart Center, the Department of Applied Physics of the University of Eastern Finland and Heart2Save Ltd. The results of the research project will be published in the scientific journals of medicine and medical technology and will be presented at scientific conferences of the respective fields. The research results of the project can be utilized by all companies in the medical technology industry, in particular companies that produce ECG measuring instruments and companies that produce rhythm recognition software.

The research aims to solve the following medical problems: 1. To study the feasibility of wrist worn PPG devices and single-lead ECG chest band. Special interest will be the use of AI in data analysis and its impact on arrhythmia detection. 2. Develop state of the art PPG and ECG based methods for long term AF monitoring. The main research questions are: 1. Can a single-lead ECG and PPG measurement be used to detect atrial fibrillation? 2. Can artificial intelligent (AI) arrhythmia analysis reliable detect rhythm changes from customer-targeted PPG and ECG recording? 3. Are biomarkers measured from blood sample suitable for the estimation of recent-onset AF duration? Specific methodological aims are: 1. Develop and test AI-based methods for arrhythmia detection and AF diagnostics 2. AI can be utilized in the screening and diagnostics of atrial fibrillation. 3. Study the kinetics of cardiac biomarkers in recent-onset AF 4. Construct statistical model for the estimation of the duration of arrhythmia episode. The purpose of the study's method development is to evaluate reliability of heart rate measurement in single-lead ECG and pulse wave measurement with healthy and patients with heart problems. The study develops computing methods based on lightweight measurement technology to reliably identify the most common cardiac arrhythmia, atrial fibrillation. The diagnosis and treatment of atrial fibrillation are decisive factors for preventing strokes. Cardioversion (CV) is a treatment procedure used to return an abnormal AF rhythm to a normal sinus rhythm (SF) in recent-onset AF. It can be performed with electrical cardioversion or with antiarrhythmic drugs. If the patient is not on oral anticoagulant (OAC) therapy, cardioversion must be performed within 48 hours after the onset of the arrhythmia. Namely, after 48 hours the risk of stroke increases substantially. If the patient has had AF for more than 48 hours, OAC must be started and used for three weeks before CV can be performed. Research patients with new-onset atrial fibrillation (\<48h) and scheduled for CV will be recruited within the research. The study will be take place at the emergency department of Kuopio University Hospital (KUH). During the study relevant patient-related information. Research patients have already undergone a 12-channel clinical ECG registration included in the normal treatment process to diagnose recent-onset AF. In actual study measurements, a Holter-ECG device is attached on patient's chest using five wet electrodes to be used as golden standard for rhythm monitoring. The lightweight measurement methods are compared with the result of the Holter-ECG registration. In addition, photopletysmograms are placed on patient's wrist for PPG registration and a single-lead dry electrode ECG sensor of the patient's chest. Moreover, IV cannula will be attached for the blood samples taken before and afterward of CV according to the study protocol. The study compares the ability of these lightweight measurement methods to detect heart rhythms compared to the Holter registration. The devices used for the measurement are: 1. Faros 360 EKG sensor with wet electrodes (Mega Elektroniikka, http://www.megaemg.com/ Kuopio Suomi). Faros 360 Holter is CE and FDA 510(k) cleared class 2a medical device, which is attached to the patient's chest with five single-use wet electrodes. 2. Suunto Movesense one-time ECG device (Suunto Oy, http://www.suunto.com Vantaa Suomi). Movesense is CE cleared consumer device, which is used with two dry electrodes to the ECG measurement. In the previous study (Afib24h), Valvira was reported and the research received permission for the clinical device study (Movesense + chest strap combination). 3. Empatica E4 activity bracelet (Empatica Ltd http://www.empatica.com Milan Italia), which is CE cleared consumer device. Empatica E4 is also a photopletysmogram, which measures optically the amount of blood circulating in the blood vessel. The researcher attaches devices to the patient. After that, the researcher starts registration with Faros 360 (device 1) and Empatica E4 (device 3) devices. Heart rate detection by ECG measurement is most commonly done by the detection of QRS complexes. Numerous of these QRS detectors have been developed in recent decades. ECG measurement with dry electrodes involves considerably more movement disturbances, compared to the wet electrode measurements, as even the small movements of the device induce major changes to the ECG signal. In addition, especially when using thumbs as a measurement points, the EMG noise from the muscles is remarkably high compared to the wet electrode measurements. This project utilizes the methods developed in the earlier mobile-ECG-projects for noise and QRS detection to allow reliably detection of QRS complexes and heart rate irregularities in the dry electrode measurements. Moreover, the previously developed heart rate detection methods are evaluated and validated by study's measurements of atrial fibrillation and normal sinus rhythm. This study examines capability of pulse detection in detection of atrial fibrillation. The photopletysmogram measures the absorption of light in the tissue. The absorption of light into the blood is greater than the absorption into the surrounding tissue. When the heart beats, capillaries expand and contract based on blood volume changes. Photopletysmography allows the heart rate measurement by detecting changes in absorption. Photopletysmogram, like a mobile-ECG device, is particularly sensitive to motion, even the small motion of led/photodiode induce major change in light intensity. Also, physiological changes cause a disturbance in heart rate measurement, for example, as the vascular elasticity changes, the pulse time changes, resulting a disturbance in measurement. Unlike the high-frequency pierced QRS complex, the pulse wave is a low-frequency up-down variation, which causes its own challenges for accurate heart rate measurement. The atriums work insufficiently in atrial fibrillation therefore the ventricles are not completely filled with blood. In addition, atrial fibrillation causes the irregular conduction of impulses from atriums to the ventricles leading to pulse irregularity. The amount of blood pumped varies from one stroke to stroke, which makes the pulse wave detection challenging. This project develops methods for accurate heart rate measurement from a pulse wave series. The method development aims to take account of disturbances due to the motion of the meter, pulse wave irregularities typical of atrial fibrillation. The main goal of the method development is to determine the pulse so precisely that pulse irregularity due to atrial fibrillation can be distinguished from normal sinus rhythm and the rhythm change can be detected with accurate manner. In atrial fibrillation, electrical impulses conduct randomly to the ventricles, causing the heart rate to be irregular and uneven. A large campaign by the Heart Association "Feel your pulse - prevent the stroke" is based on heart rate or pulse recognition. Pulse recognition is of course the cheapest method to detect atrial fibrillation, but this method produces a large number of false positives. By ECG measurement, the detection of atrial fibrillation is much more reliable. Automated atrial fibrillation detection algorithms have been developed for this purpose. Identification of the atrium activation in long-term Holter-ECG measurements is generally very challenging due to the poor signal-noise-ratio (motion, muscle-artefacts and partly overlapping much stronger ventricular activity). For this reason, most atrial fibrillation detection algorithms are based on the identification of pulse irregularity. For parametrization of the irregularity of the heart rate (RR-interval) has been introduced several relatively simple but reliable time-level methods. As an example, A RdR-based method wherein the RR intervals (heart rate) are represented as a function of consecutive RR interval changes (heart rate change). The RdR-graph defines the fragmentation of the pattern resulting from irregular heart rate changes. In addition, there are methods that estimate RR time series internal coherence. Various nonlinear methods have also been introduced to parametrization of the heart rate variation, enabling the dynamics of the heart rate variation to be described more broadly (without limitation of the linearity assumption). One class of nonlinear methods are different entropy quantities, these are particularly interesting for the identification of atrial fibrillation and the irregular heart rate. Entropy quantities can be used to estimate the regularity and predictability of the RR time series. Typically, the reliable calculation of the entropy quantities requires a relatively long measurement time, but also entropy quantities that are suitable for the analysis of short measurements have been introduced. This research project develops new atrial fibrillation detection algorithms for the mobile-ECG measurement and pulse wave measurement on the basis of already existing methods. Algorithms must take into account atrial and ventricular premature complexes. Ignoring of these increases the irregularity of the RR time series and thus increases the number of false positive atrial fibrillation. Natriuretic peptides are hormones secreted by the heart. Atrial natriuretic peptide (ANP) is secreted from atria and brain-natriuretic peptide (BNP) from atria and ventricles. BNP and NT-pro-BNP are cleaved from their precursor by pro-BNP. Both are used in the diagnosis of heart failure. Patients with AF, without other heart disease, were found to have elevated BNP. High sensitivity cardiac troponins T (hs-cTnT) and I (hs-cTnI) are cardiac-specific biomarkers secreted by myocardial cells and their levels correlate the amount of cardiomyocyte injury. Patients with persistent AF has shown to have elevated troponins levels indicating non-specific cardiomyocyte injury. However, no significant evidence indicating correlation between TnI kinetics and other AF-related biomarkers has been associated with persistent AF. The levels of TnT seems to decrease after the successfully cardioversion. Therefore, the kinetics of cardiac troponins especially TnT may be useful to predict the duration of AF episode. The study aims to determine whether peptides can be used to assess the duration of atrial fibrillation. This can be used to plan future treatment for atrial fibrillation. Moreover, this project examines the kinetics of cardiac biomarkers and their association with the duration of AF.