Accuracy Of Atrial Fibrillation Detection Research Articles

A novel atrial fibrillation automatic detection algorithm based on ensemble learning and multi-feature discrimination.

Atrial fibrillation (AF) is a prevalent cardiac arrhythmia disorder that necessitates long-time electrocardiogram (ECG) data for clinical diagnosis, leading to low detection efficiency. Automatic detection of AF signals within short-time ECG recordings is challenging. To address these issues, this paper proposes a novel algorithm called Ensemble Learning and Multi-Feature Discrimination (ELMD) for the identification and detection of AF signals. Firstly, a robust classifier, BSK-Model, is constructed using ensemble learning. Subsequently, the ECG R-waves are detected, and the ECG signals are segmented into consecutive RR intervals. Time domain, frequency domain, and nonlinear features are extracted from these intervals. Finally, these features are fed into the BSK-Model to discriminate AF. The proposed methodology is evaluated using the MIT-BIH AF database. The results demonstrate that when RR intervals are employed as classification units, the specificity and accuracy of AF detection in long-time ECG data exceed 99%, showcasing a significant improvement over traditional single-model classification. Additionally, the sensitivity and accuracy achieved by testing cardiac segments are both above 96%. With a minimum requirement of only four cardiac segments, AF events can be accurately identified, thereby enabling rapid discrimination of short-time single-lead ECG AF events. Consequently, this approach is suitable for real-time and accurate AF detection using low-computational-power ECG diagnostic analysis devices, such as wearable devices.

A Real-Time Contact-Free Atrial Fibrillation Detection System for Mobile Devices.

As the global population ages, the death and prevalence of atrial fibrillation (AF) continue to rise, posing significant concerns due to its strong association with stroke-related disabilities. Detecting AF early before a stroke occurs has become paramount. However, existing methods face challenges in achieving quick, easy, and affordable detection in complex environments characterized by motion interference and varying light conditions. To address these challenges, we propose a system that is employable for edge computing devices like smartphones, tablets, or laptops. Meanwhile, to ensure that the dataset reflects real-world scenarios, we collect 7,216 30-second segments from 452 subjects, categorized into Atrial Fibrillation (AF), Normal Sinus Rhythm (NSR), and Other Arrhythmias (Others), with a subject ratio of 105:116:231. Our lightweight non-contact facial rPPG atrial fibrillation detection system utilizes a Convolution Neural Network (CNN) with a large receptive field and a bidirectional spatial mapping augmented attention module (BiSME-ATT) coupled with a bidirectional feature pyramid network layer (BiFPN), optimized for deployment on mobile devices by reducing model parameters and floating-point operations per second (FLOPs). Our approach significantly improves AF detection accuracy, sensitivity, specificity, positive predictive value, and negative predictive value to 94.39%, 91.57%, 95.44%, 88.06%, and 96.93%, respectively, in AF vs. Non-AF scenarios. Furthermore, the results demonstrate notable enhancements in AF detection across various motion and light intensity levels.

A generalizable and robust deep learning method for atrial fibrillation detection from long-term electrocardiogram

The reliable detection of atrial fibrillation (AF) is important for diagnosing the disease, tracking its progression, and developing individualized care strategies. However, models based on limited data are prone to data dependency due to differences in data feature distribution, which generally degrades their performance on unseen external datasets. In this work, we propose a multi-input fusion AF detection network (MIF-AFNet), which cascades residual convolutional neural networks and bidirectional long short-term memory networks to capture the deep features of electrocardiogram (ECG) and RR intervals (RRIs), respectively. Additionally, the ECG signals use a data augmentation method to alleviate the morphological imbalance. MIF-AFNet learns a robust feature representation for accurate AF detection by fusing the available information from RRIs and ECG. The proposed method was developed and evaluated using 5 long-term ECG datasets (CPSC2021, AFDB, LTAF, MITDB, and NSRDB) from PhysioNet. The subject-wise five-fold cross-validation was performed on CPSC2021, and the proposed method achieved an AF detection accuracy of 98.63%. The generalization performance is further evaluated on four external independent datasets (AFDB, LTAF, MITDB, and NSRDB), achieving accuracies of 98.63%, 97.04%, 98.07%, and 100%, respectively. The results show that the proposed method can accurately detect AF from long-term ECG recordings. In addition, the low complexity of the model makes it less demanding on computing resources. Therefore, it has the potential to improve the automatic diagnosis and management of AF in wearable device-based long-term home monitoring.

Accuracy, Usability, and Adherence of Smartwatches for Atrial Fibrillation Detection in Older Adults After Stroke: Randomized Controlled Trial.

Atrial fibrillation (AF) is a common cause of stroke, and timely diagnosis is critical for secondary prevention. Little is known about smartwatches for AF detection among stroke survivors. We aimed to examine accuracy, usability, and adherence to a smartwatch-based AF monitoring system designed by older stroke survivors and their caregivers. This study aims to examine the feasibility of smartwatches for AF detection in older stroke survivors. Pulsewatch is a randomized controlled trial (RCT) in which stroke survivors received either a smartwatch-smartphone dyad for AF detection (Pulsewatch system) plus an electrocardiogram patch or the patch alone for 14 days to assess the accuracy and usability of the system (phase 1). Participants were subsequently rerandomized to potentially 30 additional days of system use to examine adherence to watch wear (phase 2). Participants were aged 50 years or older, had survived an ischemic stroke, and had no major contraindications to oral anticoagulants. The accuracy for AF detection was determined by comparing it to cardiologist-overread electrocardiogram patch, and the usability was assessed with the System Usability Scale (SUS). Adherence was operationalized as daily watch wear time over the 30-day monitoring period. A total of 120 participants were enrolled (mean age 65 years; 50/120, 41% female; 106/120, 88% White). The Pulsewatch system demonstrated 92.9% (95% CI 85.3%-97.4%) accuracy for AF detection. Mean usability score was 65 out of 100, and on average, participants wore the watch for 21.2 (SD 8.3) of the 30 days. Our findings demonstrate that a smartwatch system designed by and for stroke survivors is a viable option for long-term arrhythmia detection among older adults at risk for AF, though it may benefit from strategies to enhance adherence to watch wear. ClinicalTrials.gov NCT03761394; https://clinicaltrials.gov/study/NCT03761394. RR2-10.1016/j.cvdhj.2021.07.002.

Accuracy in detecting atrial fibrillation in single-lead ECGs: an online survey comparing the influence of clinical expertise and smart devices.

Manual interpretation of single-lead ECGs (SL-ECGs) is often required to confirm a diagnosis of atrial fibrillation. However accuracy in detecting atrial fibrillation via SL-ECGs may vary according to clinical expertise and choice of smart device. To compare the accuracy of cardiologists, internal medicine residents and medical students in detecting atrial fibrillation via SL-ECGs from five different smart devices (Apple Watch, Fitbit Sense, KardiaMobile, Samsung Galaxy Watch, Withings ScanWatch). Participants were also asked to assess the quality and readability of SL-ECGs. In this prospective study (BaselWearableStudy, NCT04809922), electronic invitations to participate in an online survey were sent to physicians at major Swiss hospitals and to medical students at Swiss universities. Participants were asked to classify up to 50 SL-ECGs (from ten patients and five devices) into three categories: sinus rhythm, atrial fibrillation or inconclusive. This classification was compared to the diagnosis via a near-simultaneous 12-lead ECG recording interpreted by two independent cardiologists. In addition, participants were asked their preference of each manufacturer's SL-ECG. Overall, 450 participants interpreted 10,865 SL-ECGs. Sensitivity and specificity for the detection of atrial fibrillation via SL-ECG were 72% and 92% for cardiologists, 68% and 86% for internal medicine residents, 54% and 65% for medical students in year 4-6 and 44% and 58% for medical students in year 1-3; p <0.001. Participants who stated prior experience in interpreting SL-ECGs demonstrated a sensitivity and specificity of 63% and 81% compared to a sensitivity and specificity of 54% and 67% for participants with no prior experience in interpreting SL-ECGs (p <0.001). Of all participants, 107 interpreted all 50 SL-ECGs. Diagnostic accuracy for the first five interpreted SL-ECGs was 60% (IQR 40-80%) and diagnostic accuracy for the last five interpreted SL-ECGs was 80% (IQR 60-90%); p <0.001. No significant difference in the accuracy of atrial fibrillation detection was seen between the five smart devices; p = 0.33. SL-ECGs from the Apple Watch were considered as having the best quality and readability by 203 (45%) and 226 (50%) participants, respectively. SL-ECGs can be challenging to interpret. Accuracy in correctly identifying atrial fibrillation depends on clinical expertise, while the choice of smart device seems to have no impact.

Impact of clinical expertise and choice of wearable device on accuracy to detect atrial fibrillation via single-lead ECGs

Abstract Funding Acknowledgements Type of funding sources: None. Background Manual interpretation of single-lead ECGs (SL-ECG) is often required to confirm a diagnosis of atrial fibrillation (AF). However, accuracy to detect AF via SL-ECGs may vary according to clinical expertise and choice of the wearable device. Aims To compare the accuracy to detect AF via single-lead ECG from five different wearable devices (Apple Watch, Fitbit Sense, KardiaMobile, Samsung Galaxy Watch, Withings ScanWatch) between cardiologists, internal medicine (IM) residents, and medical students. Methods In this prospective study, invitations to an online survey were distributed via digital invitations among physicians from major Swiss hospitals and medical students from Swiss universities. Participants needed to classify 50 SL-ECGs (from 10 patients and 5 different devices) into three categories: sinus rhythm, AF, or inconclusive. This classification was compared to the diagnosis from an almost simultaneously recorded 12-lead ECG interpreted by two independent cardiologists. In addition, participants were asked to choose the best/worst quality/readability of each manufacturer’s SL-ECG. (Figure A) Results Overall 450 participants rated 10’865 SL-ECGs. Sensitivity and specificity for the detection of AF via SL-ECG was 75% and 92% for cardiologists, 70% and 86% for IM residents, 55% and 65% for master medical students (year 4-6) and 45% and 58% for bachelor medical students (year 1-3), p&amp;lt;0.001, (Figure A). Participants which stated prior experience in interpreting SL-ECGs demonstrated a sensitivity and specificity of 65% and 81% compared to a sensitivity and specificity of 56% and 67% for participants with no prior experience in interpreting SL-ECGs, p&amp;lt;0.001. Of all participants, 107 rated all 50 ECGs. Diagnostic accuracy of the first five interpreted SL-ECGs was 60% (IQR 40-80%) and diagnostic accuracy of the last five interpreted SL- ECGs was 80% (IQR 60-90%), p&amp;lt;0.001. No significant difference in the accuracy of AF detection was seen between the 5 different wearable devices, p=0.58. 203 participants (45%) and 226 participants (50 %) ranked SL- ECGs from the Apple Watch as the best quality and readability, respectively. Conclusion SL-ECG can be difficult to interpret. Accuracy to correctly identify AF depends on clinical expertise, while the choice of the wearable device seems to have no impact.

Automated atrial fibrillation detection with a smartwatch and smart-ring in individuals with cardiovascular disease

Abstract Funding Acknowledgements Type of funding sources: None. Introduction Wearable technology is increasingly used to identify episodes of atrial fibrillation (AF). However, most validation studies using single-lead electrocardiograms (ECGs) only included patients with either sinus rhythm or AF and, therefore, likely overestimate the sensivities and specificities of AF detection algorithms. Purpose We sought to evaluate the diagnostic accuracy of the Apple Watch and CART-I ring AF algorithms in a broad cohort of patients with pre-existing cardiovascular disease. Methods In this single-centre prospective study, individuals with known cardiovascular disease underwent simultaneous 12-lead ECG recordings with a 30-second single-lead ECG from an Apple Watch Series 6 and a CART-I Ring. The order of devices was randomly assigned. Two cardiologists independently evaluated the 12-lead ECGs for the presence of AF; any disagreements were resolved by a third cardiologist. The wearable devices’ algorithm labelled ECGs as "AF", "Not AF" or "Unclassified". Diagnostic accuracy was determined by calculating the sensitivity and specificity for each device and comparing them using McNemar’s test. The primary analysis included Unclassified ECGs, which categorised as false-positive (no AF on the 12-lead ECG) or false-negative (AF on the 12-lead ECG). A separate analysis excluded all Unclassified ECGs. Results We enrolled 400 consecutive patients. Seven patients had suboptimal 12-lead ECGs and were excluded from the analysis. The median age was 63 (58-75) years, and 73% of the population was male. Baseline demographics are depicted in Table 1. A total of 1573 ECGs were analysed. Forty-two percent of patients were in AF, 9.9% were in an atrial tachycardia, 3.3% had sinus rhythm with frequent ectopy. If all ECGs were included, the CART-I ring's AF sensitivity was significantly higher than the Apple Watch, at 86.6% (95% CI: 80.6%-91.1%) and 67.4% (95%CI: 60.0%-74.1%), respectively (Fig 1). Similarly, AF specificity was higher with CART-I ring (89.4% [95%CI: 84.7%-92.8%]) when compared to the Apple Watch (74.9% [95%CI 68.9%-80.1%]). The Apple Watch had more Unclassified ECGs than CART-I ring (78 vs. 2, respectively). They were divided into three categories: 14.1% as HR 50 bpm, 10.21% as HR &amp;gt; 120 bpm, and 76.9% as Inconclusive (Table 2). When Uclassified ECGs from the Apple Watch were excluded, AF sensitivity and specificity improved by 16.33% (p=0.002) and 13.38% (p&amp;lt;0.001), respectively. Conclusion In patients with cardiovascular disease, the diagnostic accuracy of AF detection from a single-lead ECG was lower than previously reported. The CART-I ring's AF sensitivity and specificity outperformed those of the Apple Watch. One in three patients with AF went undetected by the Apple Watch.

An Effective Atrial Fibrillation Detection from Short Single-Lead Electrocardiogram Recordings Using MCNN-BLSTM Network

Atrial fibrillation (AF) is an arrhythmia that may cause blood clots and increase the risk of stroke and heart failure. Traditional 12-lead electrocardiogram (ECG) acquisition equipment is complex and difficult to carry. Short single-lead ECG recordings based on wearable devices can remedy these shortcomings. However, reliable and accurate atrial fibrillation detection is still an issue because of the limited information on the short single-lead ECG recordings. In this paper, we propose a novel multi-branch convolutional neural network and bidirectional long short-term memory network (MCNN-BLSTM) to deal with the reliability and accuracy of AF detection in short single-lead ECG recordings. Firstly, to fuller extract the feature information of short single-lead ECG recordings, the MCNN module is designed to dynamically set several corresponding branches according to the number of slices of short single-lead ECG recordings. Then, the BLSTM module is designed to further enhance the feature information learned from each branch. We validated the model on the PhysioNet/CinC Challenge 2017 (CinC2017) database and verified the generalization of the model on the China Physiological Signal Challenge 2018 (CPSC2018) database. The results show that the accuracy of the model on the CinC 2017 database reaches 87.57%, and the average F1 score reaches 84.56%. The accuracy of the model on the CPSC 2018 database reaches 87.50%, and the average F1 score reaches 82.01%. Compared with other advanced methods, our model shows better performance and can meet the daily needs of atrial fibrillation detection with short ECG wearable devices.

Continuous mHealth Patch Monitoring for the Algorithm-Based Detection of Atrial Fibrillation: Feasibility and Diagnostic Accuracy Study.

BackgroundThe detection of atrial fibrillation (AF) is a major clinical challenge as AF is often paroxysmal and asymptomatic. Novel mobile health (mHealth) technologies could provide a cost-effective and reliable solution for AF screening. However, many of these techniques have not been clinically validated.ObjectiveThe purpose of this study is to evaluate the feasibility and reliability of artificial intelligence (AI) arrhythmia analysis for AF detection with an mHealth patch device designed for personal well-being.MethodsPatients (N=178) with an AF (n=79, 44%) or sinus rhythm (n=99, 56%) were recruited from the emergency care department. A single-lead, 24-hour, electrocardiogram-based heart rate variability (HRV) measurement was recorded with the mHealth patch device and analyzed with a novel AI arrhythmia analysis software. Simultaneously registered 3-lead electrocardiograms (Holter) served as the gold standard for the final rhythm diagnostics.ResultsOf the HRV data produced by the single-lead mHealth patch, 81.5% (3099/3802 hours) were interpretable, and the subject-based median for interpretable HRV data was 99% (25th percentile=77% and 75th percentile=100%). The AI arrhythmia detection algorithm detected AF correctly in all patients in the AF group and suggested the presence of AF in 5 patients in the control group, resulting in a subject-based AF detection accuracy of 97.2%, a sensitivity of 100%, and a specificity of 94.9%. The time-based AF detection accuracy, sensitivity, and specificity of the AI arrhythmia detection algorithm were 98.7%, 99.6%, and 98.0%, respectively.ConclusionsThe 24-hour HRV monitoring by the mHealth patch device enabled accurate automatic AF detection. Thus, the wearable mHealth patch device with AI arrhythmia analysis is a novel method for AF screening.Trial RegistrationClinicalTrials.gov NCT03507335; https://clinicaltrials.gov/ct2/show/NCT03507335

Head-to-head comparison of single-lead and six-lead electrocardiography of a wrist-worn device for atrial fibrillation detection

Abstract Funding Acknowledgements Type of funding sources: Public grant(s) – EU funding. Main funding source(s): Funded by the European Regional Development Fund under agreement with the Research Council of Lithuania. Background The increasing numbers of available mHealth tools for electrocardiography (ECG)-based atrial fibrillation (AF) detection promote long-term screening. A common feature of smartwatches is lead-I-like ECG. However, only limited data exists directly comparing the performance of single-lead and six-lead wearable-recorded ECGs. Purpose To compare the accuracy of single-lead and six-lead ECGs of the same wrist-worn device for AF detection. Methods We included patients with AF which represent the main group for testing the diagnostic ability of wearable. In addition, authors selected control groups of stable sinus rhythm (SR) and SR with frequent premature contractions. Cardiac rhythm was monitored using an investigational wrist-worn device which provides six-lead ECG, similar to standard limb leads. To display a single-lead wearable-recorded ECGs, the same six-lead ECG tracings were trimmed to a width of lead-I-like ECGs. A validated Holter ECG device constituted a gold standard test for rhythm verification. Two independent diagnosis-blinded cardiologists evaluated reference, six-lead and single-lead ECGs as "AF", "SR", or "Inconclusive". A third cardiologist evaluated ECGs only in cases of physician disagreement. Results A total of 344 adult patients were enrolled in this study including AF group (121 patients) and control group of SR with or without frequent premature contractions (223 patients). Patients with missing (11/420; 2.62%) or insufficient quality (43/420; 10.24%) of wearable-recorded ECGs were excluded. AF detection based on single-lead and six-lead ECGs yielded sensitivity of 95.73% (95% CI 90.31–98.6%) and 99.16% (95% CI 95.41–99.98%), respectively. Specificity was 100% (95% CI 96.19–100%) for both single-lead and six-lead ECGs when differentiating between AF and stable SR. If patients with frequent premature beats were included in the control group, the specificity of single-lead and six-lead ECGs dropped to 95.81% (95% CI 92.31–98.07%) and 99.1% (95% CI 96.78–99.89%), respectively. False positive cases were more common for single-lead ECG (9/332) compared to six-lead ECG (2/341) (P=0.02). There was a strong association between reference ECGs and wearable-recorded ECGs (P&amp;lt;0.001): Cramer‘s V 0.91, (95% CI 0.82–1.0) for single-lead ECG and 0.98 (95% CI 0.89–1.0) for six-lead ECG. After including a control group of frequent premature contractions, single-lead ECG (12/344) was more frequently labelled "Inconclusive" than six-lead ECG (3/344) (P=0.01). Inter-rater agreement, measured as Cohen’s kappa, indicated great concordance in both methods but was higher for six-lead ECG (0.945, P&amp;lt;0.001) than single-lead ECG (0.887, P&amp;lt;0.001). Conclusions Six-lead ECG of a wearable device demonstrated higher diagnostic accuracy of AF detection than single-lead ECG when controlled by patients with frequent premature contractions. The performance of both methods was equivalent when controlled by patients with stable SR.

Accuracy of atrial fibrillation detection by an insertable cardiac monitor in patients undergoing catheter ablation: Results of the BioVAD study.

BackgroundInsertable cardiac monitors (ICMs) are increasingly used to evaluate the atrial fibrillation (AF) burden after catheter ablation of AF. BioMonitor III (BM3) is an ICM with a long sensing vector, which enhances sensing capabilities. The AF detection algorithm of the BM3 is based on R–R interval variability.ObjectiveTo evaluate the performance of the AF detection algorithm of BM3 in patients before and after catheter ablation of AF using simultaneous Holter recordings.MethodsIn this prospective study, we enrolled patients scheduled for catheter ablation of paroxysmal or persistent AF. After BM3 implantation, patients had a 4 days Holter registration before and 3 months after ablation. All true AF episodes ≥2 min on the Holter were annotated and matched with BM3 detected AF detections.ResultsThirty‐one patients were enrolled (mean age 60 ± 8, 74% male, 68% paroxysmal AF). Fifty‐six Holter registrations were performed in 30 patients. Twelve patients demonstrated at least one true AF episode with a total AF duration of 570 h. The AF burden accuracy of BM3 before catheter ablation was 99.6%, with a duration sensitivity of 98.6% and a duration specificity of 99.9%. The AF burden accuracy of BM3 after catheter ablation was 99.8%, with a duration sensitivity of 90.2% and a duration specificity of 99.9%. Overall, the AF burden detected on the Holter and BM3 demonstrated a high Pearson correlation coefficient of 0.996.ConclusionBM3 accurately detects AF burden in patients before and after catheter ablation of AF.

Abstract 9886: Use of a Smartwatch and App Designed by Stroke Survivors for Atrial Fibrillation Detection in Older Adults After Stroke/Transient Ischemic Event: Preliminary Findings from an Ongoing Randomized Clinical Trial

Background: Atrial fibrillation (AF) is a common cause of ischemic stroke that, though critical to diagnose and treat, often goes undetected. Smartwatches are capable of AF detection, but little is known about their use among stroke survivors. In collaboration with stroke survivors and clinicians, we designed a smartphone-smartwatch system for AF detection. In our ongoing clinical trial, we are examining the performance and user experience of the system in older stroke survivors. Hypothesis: The system will be accurate for AF detection among stroke survivors compared to 14-day ECG patch monitoring; participants will find the system usable and will adhere to daily use over 30 days. Methods: The Pulsewatch study is an ongoing two-phased randomized controlled trial with randomization at two timepoints into control or intervention groups. Phase I (14 days) assesses accuracy and usability of the smartwatch and study app. Phase II (30 days) examines watch wear adherence. The intervention group receives a smartwatch with embedded AF detection and a smartphone app for rhythm and symptom tracking. Eligibility criteria are: &gt; 50 years old, ischemic stroke in the past decade, and no major contraindications to anticoagulation therapy. System accuracy for AF detection is compared to an ECG patch monitor, and usability is assessed via the System Usability Scale. Adherence is defined as daily watch wear time over the monitoring period. Results: To date, 69 participants have been randomized into phase I and 36 into phase II (mean age 65, 41% female, 87% White). The smartwatch system demonstrated a sensitivity of 66.7% (95% CI: 9.4 - 99.2) and specificity of 92.6% (95% CI: 82.1 - 97.9) for detection of 3 AF cases over the 14 day period (91.2% accuracy) compared to the ECG patch. The mean usability score was 65/100 and on average, participants wore the watch on 12/30 days for an average of 1.8 hours (SD 0.7) daily. Conclusion: A smartwatch and app dyad demonstrated reasonable accuracy for AF detection and high usability, though adherence was modest. Use of commercial wearables for AF detection among stroke patients shows promise, but additional data from the ongoing Pulsewatch clinical trial will continue to inform strategies to maximize adherence to these devices among older adults.

On-device Prior Knowledge Incorporated Learning for Personalized Atrial Fibrillation Detection

Atrial Fibrillation (AF), one of the most prevalent arrhythmias, is an irregular heart-rate rhythm causing serious health problems such as stroke and heart failure. Deep learning based methods have been exploited to provide an end-to-end AF detection by automatically extracting features from Electrocardiogram (ECG) signal and achieve state-of-the-art results. However, the pre-trained models cannot adapt to each patient’s rhythm due to the high variability of rhythm characteristics among different patients. Furthermore, the deep models are prone to overfitting when fine-tuned on the limited ECG of the specific patient for personalization. In this work, we propose a prior knowledge incorporated learning method to effectively personalize the model for patient-specific AF detection and alleviate the overfitting problems. To be more specific, a prior-incorporated portion importance mechanism is proposed to enforce the network to learn to focus on the targeted portion of the ECG, following the cardiologists’ domain knowledge in recognizing AF. A prior-incorporated regularization mechanism is further devised to alleviate model overfitting during personalization by regularizing the fine-tuning process with feature priors on typical AF rhythms of the general population. The proposed personalization method embeds the well-defined prior knowledge in diagnosing AF rhythm into the personalization procedure, which improves the personalized deep model and eliminates the workload of manually adjusting parameters in conventional AF detection method. The prior knowledge incorporated personalization is feasibly and semi-automatically conducted on the edge, device of the cardiac monitoring system. We report an average AF detection accuracy of 95.3% of three deep models over patients, surpassing the pre-trained model by a large margin of 11.5% and the fine-tuning strategy by 8.6%.

Estimation of PQ distance dispersion for atrial fibrillation detection

Background and objective: Atrial fibrillation (AF) is the most common cardiac arrhythmia in the world. It is associated with significantly increased morbidity and mortality. Diagnosis of the disease can be based on the analysis of the electrical atrial activity, on quantification of the heart rate irregularity or on a mixture of the both approaches. Since the amplitude of the atrial waves is small, their analysis can lead to false results. On the other hand, the heart rate based analysis usually leads to many unnecessary warnings. Therefore, our goal is to develop a new method for effective AF detection based on the analysis of the electrical atrial waves.Methods: The proposed method employs the fact that there is a lack of repeatable P waves preceding QRS complexes during AF. We apply the operation of spatio-temporal filtering (STF) to magnify and detect the prominent spatio-temporal patterns (STP) within the P waves in multi-channel ECG recordings. Later we measure their distances (PQ) to the succeeding QRS complexes, and we estimate dispersion of the obtained PQ series. For signals with normal sinus rhythm, this dispersion is usually very low, and contrary, for AF it is much raised. This allows for effective discrimination of this cardiologic disorder.Results: Tested on an ECG database consisting of AF cases, normal rhythm cases and cases with normal rhythm restored by the use of cardioversion, the method proposed allowed for AF detection with the accuracy of 98.75% on the basis of both 8–channel and 2–channel signals of 12 s length. When the signals length was decreased to 6 s, the accuracy varied in the range of 95%−97.5% depending on the number of channels and the dispersion measure applied.Conclusions: Our approach allows for high accuracy of atrial fibrillation detection using the analysis of electrical atrial activity. The method can be applied to an early detection of the desease and can advantageously be used to decrease the number of false warnings in systems based on the analysis of the heart rate.

Wrist Band Photoplethysmography Autocorrelation Analysis Enables Detection of Atrial Fibrillation Without Pulse Detection.

Atrial fibrillation is often asymptomatic and intermittent making its detection challenging. A photoplethysmography (PPG) provides a promising option for atrial fibrillation detection. However, the shapes of pulse waves vary in atrial fibrillation decreasing pulse and atrial fibrillation detection accuracy. This study evaluated ten robust photoplethysmography features for detection of atrial fibrillation. The study was a national multi-center clinical study in Finland and the data were combined from two broader research projects (NCT03721601, URL: https://clinicaltrials.gov/ct2/show/NCT03721601 and NCT03753139, URL: https://clinicaltrials.gov/ct2/show/NCT03753139). A photoplethysmography signal was recorded with a wrist band. Five pulse interval variability, four amplitude features and a novel autocorrelation-based morphology feature were calculated and evaluated independently as predictors of atrial fibrillation. A multivariate predictor model including only the most significant features was established. The models were 10-fold cross-validated. 359 patients were included in the study (atrial fibrillation n = 169, sinus rhythm n = 190). The autocorrelation univariate predictor model detected atrial fibrillation with the highest area under receiver operating characteristic curve (AUC) value of 0.982 (sensitivity 95.1%, specificity 93.7%). Autocorrelation was also the most significant individual feature (p < 0.00001) in the multivariate predictor model, detecting atrial fibrillation with AUC of 0.993 (sensitivity 96.4%, specificity 96.3%). Our results demonstrated that the autocorrelation independently detects atrial fibrillation reliably without the need of pulse detection. Combining pulse wave morphology-based features such as autocorrelation with information from pulse-interval variability it is possible to detect atrial fibrillation with high accuracy with a commercial wrist band. Photoplethysmography wrist bands accompanied with atrial fibrillation detection algorithms utilizing autocorrelation could provide a computationally very effective and reliable wearable monitoring method in screening of atrial fibrillation.

Accurate detection of atrial fibrillation using a smartphone remains uncertain: a systematic review and meta-analysis

Abstract Introduction Early diagnosis of atrial fibrillation (AF) is essential to reduce complications such as stroke, and improve patient quality of life. Novel screening techniques using smartphone camera photoplethysmography (PPG) can be used for AF detection, but their clinical applicability remains unclear. Purpose To assess the diagnostic accuracy of smartphone PPG compared to conventional ECG for AF detection. Methods We performed a systematic review of MEDLINE, EMBASE, Cochrane library, and other databases (1980-October 2019), including any study or abstract where smartphone finger-tip PPG was compared with a reference ECG (1, 3 or 12-lead). Outcomes were sensitivity (SE), specificity (SP), positive and negative predictive value (PPV; NPV) and overall accuracy. Bivariate hierarchical random effects meta-analysis was performed for studies with confidence intervals for SE and SP, and funnel plots were used to identify publication bias. Study quality was assessed using the established QUADAS-2 tool by two independent graders. Results 1350 publications were screened, of which 17 studies were included in the systematic review (7 full text publications and 10 abstracts), providing 21 comparisons of accuracy for AF detection. Most studies were based in secondary care and small (range n=33 to 1095), with a total of 5469 participants including 1384 with AF. Only 4 studies were multicentre. Smartphone applications used were Cardiio Rhythm, Fibricheck, Preventicus and Heartbeats, with 7 studies not specifying the tool used. Overall SE and SP for AF detection were high, ranging from 76 to 100%, and 85 to 100% respectively. PPV ranged from 54 to 100% and NPV from 77 to 100%, with overall accuracy between 61 and 99%. The meta-analysis included 12 comparisons from 10 studies (n=2714; 936 with AF). The pooled SE was 93% (95% CI 90–96%) and SP 97% (95% CI 95–99%); Figure 1A. QUADAS-2 assessment demonstrated poor quality of studies overall, with a high or unclear risk of bias in at least one domain for all studies. There was clear evidence of publication bias; Figure 1B. Conclusions PPG offers the potential for large scale, non-invasive, patient-led screening of AF. However, current evidence is limited to biased, low quality studies often with unrealistic results for AF detection. These are insufficient to advise clinicians on the true value of current smartphone PPG technology. Figure 1. Meta-analysis &amp; publication bias Funding Acknowledgement Type of funding source: Public grant(s) – EU funding. Main funding source(s): BigData@Heart EU/EFPIA IMI 116074.

Diagnostic Utility of Smartwatch Technology for Atrial Fibrillation Detection – A Systematic Analysis

Smartphone technologies have been recently developed to assess heart rate and rhythm, but their role in accurately detecting atrial fibrillation (AF) remains unknown. We sought to perform a meta-analysis using prospective studies comparing Smartwatch technology with current monitoring standards for AF detection (ECG, Holter, Patch Monitor, ILR). We performed a comprehensive literature search for prospective studies comparing Smartwatch technology simultaneously with current monitoring standards (ECG, Holter, and Patch monitor) for AF detection since inception to November 25th, 2019. The outcome studied was the accuracy of AF detection. Accuracy was determined with concomitant usage of ECG monitoring, Holter monitoring, loop recorder, or patch monitoring. A total of 9 observational studies were included comparing smartwatch technology, 3 using single-lead ECG monitoring, and six studies using photoplethysmography with routine AF monitoring strategies. A total of 1559 patients were enrolled (mean age 63.5 years, 39.5% had an AF history). The mean monitoring time was 75.6 days. Smartwatch was non-inferior to composite ECG monitoring strategies (OR 1.06, 95% CI 0.93 - 1.21, p=0.37), composite 12 lead ECG/Holter monitoring (OR 0.90, 95% CI 0.62 - 1.30, p=0.57) and patch monitoring (OR 1.28, 95% CI 0.84 - 1.94, p=0.24) for AF detection. The sensitivity and specificity for AF detection using a smartwatch was 95% and 94%, respectively. Smartwatch based single-lead ECG and photoplethysmography appear to be reasonable alternatives for AF monitoring.

Validating the robustness of an internet of things based atrial fibrillation detection system

This paper describes the validation of a deep learning model for Internet of Things (IoT) based health care applications. As such, the deep learning model was created to detect episodes of Atrial Fibrillation (AF) using Heart Rate (HR) signals. The initial Long Short-Term Memory (LSTM) model was developed using 20 data sets, from distinct subjects, obtained from the AFDB database on PhysioNet. This model achieved an AF detection accuracy of 98.51% with ten fold cross validation. In this study, we validated the initial results by testing the developed deep learning model with unknown data. To be specific, we fed the data from 82 subjects to the deep learning system and compared the classification results with the diagnosis results indicated by human practitioners. The validation results show 94% accuracy with an area under the Receiver Operating Characteristic (ROC) curve of 96.58. These results indicate that the LSTM model is able to extract the feature maps from the unknown data and hence detect the AF periods accurately. With this blindfold validation testing we violated a well known design rule for learning systems which states that more data should be used for training than for testing. By doing so, we have established that our deep learning system is fit for practical deployment, because in a practical situation the diagnosis support system must apply the knowledge, extracted from a limited training data set, to a HR trace from a patient.

P5648Efficiency and accuracy of arrhythmia detection using implantable cardiac monitor: A prospective multicenter randomized clinical trial comparing Reveal LINQ and Confirm Rx

Abstract Background Implantable cardiac monitor (ICM) has been used to detect occult cardiac arrhythmias in a variety of clinical situations. The reliability and accuracy of diagnosing cardiac arrhythmia could impact patient care. However, it is not clear the efficiency and reliability of detecting cardiac arrhythmias between currently available two ICMs. Purpose To compare the efficiency and accuracy of diagnosing cardiac arrhythmias between Reveal LINQ™ and Confirm Rx™. Methods In the prospective multicenter randomized study, a total of 80 patients (age: 61±17 years, men: 41 [51%]) with cryptogenic stroke (n=52) or recurrent unexplained episodes of palpitations (n=3) or syncope (n=25) received an ICM for detection of arrhythmias. Patients were randomized one to one into receiving either Reveal LINQ™ or Confirm Rx™. Arrhythmic events are defined as pauses of ≥3.0 seconds, bradycardia with heart rate of ≤40 bpm, tachycardia with heart rate of ≥150 bpm, and atrial fibrillation (AF). The time of detected arrhythmic events and patient activated events to the time of data availability on respective website site for analysis were compared between the two ICM. All of the arrhythmia events were adjudicated by two cardiologists to assess the accuracy. Results A total of 680 arrhythmic events including 352 AFs, 109 bradycardias, 167 tachycardias and 52 pauses were transmitted over a mean follow-up of 4.2±2.5 months. The mean time to data transmission was significantly faster in Confirm Rx™ than in Reveal LINQ™ (24±103 vs 475±426 min, P&lt;0.0001). One-hundred fifty-nine of 352 AF events (45%) and 237 of 328 non-AF events (72%) were accurately detected by ICMs (P&lt;0.0001). Twenty-eight of 51 AF events (55%) were detected accurately by Reveal LINQ™ and 131 of 301 AF events (44%) were accurately detected by Confirm Rx™ (P=0.13). The reason for inaccurate AF detection was frequent ectopic beats (n=134), P-wave oversensing (n=15), T-wave oversensing (n=8), both P- and T-waves oversensing (n=18) and sinus tachycardia with R-wave undersensing (n=18). Conclusions Data transmission of arrhythmic events using Confirm Rx™ is significantly faster than Reveal LINQ™. The accuracy of AF detection with current algorithms in both ICM remains suboptimal. More patients and longer follow-up is required to confirm these findings.

Accurate detection of atrial fibrillation from 12-lead ECG using deep neural network

Atrial fibrillation (AF) is the most common heart arrhythmia, and 12-lead electrocardiogram (ECG) is regarded as the gold standard for AF diagnosis. Highly accurate diagnosis of AF based on 12-lead ECG is valuable and remains challenging. In this paper, we proposed a novel method with high accuracy for AF detection based on deep learning. The proposed method constructed a novel one-dimensional deep densely connected neural network (DDNN) to detect AF in ECG waveforms with a length of 10s. A large set of 16,557 12-lead ECG recordings collected from multiple hospitals and wearable ECG devices were used to evaluate the performance of the DDNN. In the test dataset (3312 12-lead ECG recordings), the DDNN obtained high performance with an accuracy of 99.35 ± 0.26%, a sensitivity of 99.19 ± 0.31%, and a specificity of 99.44 ± 0.17%. Its high performance and automatic nature both demonstrate that the proposed network has a great potential to be applied to clinical computer-aided diagnosis of AF or future screening of AF in wearable devices.