Early detection of paroxysmal atrial fibrillation from non-episodic ECG data using cardiac dynamics features and different classification models

Atrial fibrillation is one of the major risk factors for ischemic stroke. Left atrial fibrosis is considered to be a hallmark of the structural remodeling in atrial fibrillation and can be detected and quantified non-invasively using modern, high-resolution 3D late gadolinium enhancement cardiac magnetic resonance imaging. According to the current state of knowledge this is the first study to deal with the detection and quantification of left atrial fibrosis in a study collective with cerebral ischemia but without known atrial fibrillation. It was investigated whether left atrial fibrosis can be linked to a potential detection of paroxysmal atrial fibrillation within the prospective, randomized and controlled Find-AFRANDOMISED study as well as to the cerebral ischemia in the medical history of our study participants. 
\nA total of 31 subjects (14 men, 17 women, 70.5 ± 6.2 years) of the Find-AFRANDOMISED study were examined on a 3 Tesla MR scanner of the University Medical Center Göttingen. In 29 subjects the quantification of fibrosis was successful. The median percentage of left atrial fibrosis was 1.6% (minimum: 0.1%, maximum: 3.7%). The obtained extent of left atrial fibrosis is comparable to the extent of left atrial fibrosis of healthy volunteers and also significantly lower than the extent of left atrial fibrosis of patients with atrial fibrillation in comparative studies. The low extent of fibrosis in the present study is thus in agreement with the absence of detection of paroxysmal atrial fibrillation in our study collective within the Find-AFRANDOMISED study. As a consequence, it can be concluded that a slight extent of left atrial fibrosis excludes (paroxysmal) atrial fibrillation with high probability. Furthermore, the additional volumetric and functional analysis of the left atrium revealed normal values for our study collective. Left atrial dilatation and reduction of left atrial function parameters are also characteristic features of atrial fibrillation. The normal values for left atrial volume and phasic function obtained in the present study are therefore also consistent with the absence of detection of (paroxysmal) atrial fibrillation within the Find-AFRANDOMISED study. Neither the extent of left atrial fibrosis nor the results of the left atrial volumetric and functional analysis can provide an explanation for the cerebral ischemia in the medical history of our study participants. In summary, it can be stated that the examined study collective presented with healthy left atria at the time of the MRI examination and that there is no indication of an underlying atrial disease like an “fibrotic atrial cardiomyopathy”. Further studies are needed to ascertain the pathophysiological link between left atrial fibrosis and ischemic stroke in both patients with known atrial fibrillation and patients without atrial fibrillation.

Diagnosis of paroxysmal atrial fibrillation is difficult but highly relevant in patients presenting with cerebral ischemia yet free from atrial fibrillation on admission. Early initiation and prolongation of continuous Holter monitoring may improve diagnostic yield compared with the standard of care including a 24-hour Holter recording. In the observational Find-AF trial (ISRCTN 46104198), consecutive patients presenting with symptoms of cerebral ischemia were included. Patients free from atrial fibrillation at presentation received 7-day Holter monitoring. Two hundred eighty-one patients were prospectively included. Forty-four (15.7%) had atrial fibrillation documented by routine electrocardiogram on admission. All remaining patients received Holter monitors at a median of 5.5 hours after presentation. In those 224 patients who received Holter monitors but had no previously known paroxysmal atrial fibrillation, the detection rate with early and prolonged (7 days) Holter monitoring (12.5%) was significantly higher than for any 24-hour (mean of 7 intervals: 4.8%, P = 0.015) or any 48-hour monitoring interval (mean of 6 intervals: 6.4%, P = 0.023). Of those 28 patients with new atrial fibrillation on Holter monitoring, 15 (6.7%) had been discharged without therapeutic anticoagulation after routine clinical care (ie, with data from 24-hour Holter monitoring only). Detection rates were 43.8% or 6.3% for short supraventricular runs of ≥ 10 beats or prolonged episodes (> 5 hours) of atrial fibrillation, respectively. Diagnostic yield appeared to be only slightly and not significantly increased during the first 3 days after the index event. Prolongation of Holter monitoring in patients with symptoms of cerebral ischemic events increases the rate of detection of paroxysmal atrial fibrillation up to Day 7, leading to a relevant change in therapy in a substantial number of patients. Early initiation of monitoring does not appear to be crucial. Hence, prolonged Holter monitoring (≥ 7 days) should be considered for all patients with unexplained cerebral ischemia.

BackgroundLong‐time monitoring of electrocardiogram with an implantable loop recorder is useful for detection of paroxysmal atrial fibrillation (PAF) in ESUS patients. However, usefulness of noninvasive 7‐day Holter monitoring has not been established. Furthermore, the incidence of other arrhythmias such as sinus pause (SP) remains unknown.Aim(1) The aim of this study was to examine the usefulness of 7‐day Holter monitoring to detect PAF in both acute and chronic periods since index stroke and (2) to clarify the incidence of SP in ESUS patients.MethodsProspectively, a total of 738 patients including 241 consecutive ESUS patients were enrolled, from April 2016 to March 2021. One hundred forty patients among 241 consecutive ESUS patients underwent chest strap‐style 7‐day Holter monitor. We examined PAF and SP (≧3 s).ResultsThe 7‐day Holter monitoring started since index stroke at a median of 17 days. Among 80 patients within 1 month since index stroke, PAF was detected in three patients (3.8%) and SP was in two patients (2.5%). By contrast, among 52 patients over 1 month since index stroke, PAF was found in only one patients (1.9%).ConclusionAlthough the incidence of PAF detected with this monitoring was low (3.8%) in ESUS patients, SP was also found in 2.5% within 1 month since index stroke. Within 1 month since stroke onset in ESUS patients, 7‐day Holter monitoring would be useful for detection of PAF and SP that would be potentially a cardiac source of embolism and might require urgent intervention.

New insights in prevention, diagnosis and treatment of stroke: its relation with atrial fibrillation

Introduction: Detection of paroxysmal atrial fibrillation (PAF) after an ischemic cerebrovascular event is of imminent interest, because oral anticoagulation as a highly effective secondary preventive treatment is available. Whereas permanent atrial fibrillation (AF) can be detected during routine electrocardiogram (ECG), longer detection duration will detect more PAF but might be resource consuming. Methods: Patients with acute ischemic stroke were prospectively investigated with an intensified algorithm to detect PAF (7 day Holter ECG, follow-up investigations after 90 days and one year, ISRCTN 46104198). Results: 281 patients were prospectively included in the study, 44 of which had to be excluded since they presented with permanent AF as diagnosed by ECG during admission and another 13 patients due other causes leaving 224 patients (mean age: 68.5 years, 58.5% male) as the final study population. 29 (12.9%) patients could be identified to have PAF during prolonged (median: 6.7 days IQR: 4.4-7.0) Holter monitoring, an additional 8 (3.6%) and 5 (2.2%) patients after follow-up investigations after 90 days and 1 year, respectively. In 182 patients no AF could be detected. Multivariate analysis identified advanced age (Odds ratio (OR) 1.04, 95% confidence interval (CI): 1.01-1.08), a higher initial NIH-SS (OR 1.10, 95% CI: 1.00-1.21) as well as clinical symptoms >24h (OR 3.45, 95% CI: 1.06-11.19) and a history of coronary artery disease (OR 2.74, 95% CI: 1.16-6.50) to be predictive for the detection of PAF. Receiver operation characteristics demonstrated an area under the curve of 0.783 for this model. Neither the results of blood analysis, duplex sonography, cerebral imaging nor routine echocardiography could be found to be independently associated with the detection of PAF. Conclusion: In acute stroke patients with advanced age, history of coronary artery disease, higher NIH-SS scores and ischemic stroke as the presenting event, a prolonged Holter ECG monitoring and follow-up is warranted to identify PAF. This could increase the detection rate of patients requiring anticoagulation and would be expected to reduce the risk of recurrent stroke in case of consequent anticoagulation of these patients.

Detection and Predictors of Paroxysmal Atrial Fibrillation in Acute Ischemic Stroke and Transient Ischemic Attack Patients in Singapore

BackgroundThe detection of paroxysmal atrial fibrillation (AF) is of importance in stroke care. The method used is continuous electrocardiogram (ECG)-monitoring or multiple short ECG-recordings during an extended period. Their relative efficiency is a matter of discussion. In a retrospective cohort study on 994 patients with an ischemic stroke or transient ischemic attack (TIA), we have compared continuous 7-day monitoring to intermittent recording 60 sec three times daily with a handheld device during 3 weeks. We related the result to subsequent occurrence of AF as detected in 12-lead ECG recordings.MethodsThe patients were identified in the local database of cardiovascular investigations. Their clinical profile and vital status during the follow-up were obtained from the Swedish Stroke Register and the Swedish general population registry. For comparison, we used an age- and sex-matched population with no known cerebrovascular event and a population with a cerebrovascular event that was not screened.ResultsAF was detected in 7.1% by continuous screening and in 5.1% by intermittent screening (P = 0.3). During follow-up of 32 months, AF in 12-lead ECG was found in 7.0%. In the subgroup with positive screening, 46.3% had AF compared with 6.7% in the subgroup with negative screening (P < 0.0001).ConclusionsThe two screening approaches had a similar yield of arrhythmia, in spite of the group with intermittent monitoring having a more favorable clinical profile. A positive screening was highly predictive of AF in ECG during the follow-up.

Detection of atrial fibrillation (AF), a type of cardiac arrhythmia, is difficult since many cases of AF are usually clinically silent and undiagnosed. In particular paroxysmal AF is a form of AF that occurs occasionally, and has a higher probability of being undetected. In this work, we present an attention based deep learning framework for detection of paroxysmal AF episodes from a sequence of windows. Time-frequency representation of 30 seconds recording windows, over a 10 minute data segment, are fed sequentially into a deep convolutional neural network for image-based feature extraction, which are then presented to a bidirectional recurrent neural network with an attention layer for AF detection. To demonstrate the effectiveness of the proposed framework for transient AF detection, we use a database of 24 hour Holter Electrocardiogram (ECG) recordings acquired from 2850 patients at the University of Virginia heart station. The algorithm achieves an AUC of 0.94 on the testing set, which exceeds the performance of baseline models. We also demonstrate the cross-domain generalizablity of the approach by adapting the learned model parameters from one recording modality (ECG) to another (photoplethysmogram) with improved AF detection performance. The proposed high accuracy, low false alarm algorithm for detecting paroxysmal AF has potential applications in long-term monitoring using wearable sensors.

Background: Detection of paroxysmal atrial fibrillation (PAF) after cryptogenic stroke has a major impact on secondary stroke prevention. Identifying clinical predictors to obtain a higher detection rate is necessary. There is an unclear association between left atrial enlargement (LAE) and PAF. Our goal was to determine the yield of 30-day Holter monitoring in patients with left atrial enlargement following acute ischemic stroke. Methods: We retrospectively identified patients with acute ischemic stroke from November 2012 to June 2014 who had an echocardiogram done and a 30-day holter monitor on discharge. Demographics, medical history, clinical/laboratory data, echocardiographic features and holter monitor detected events were collected. Moderate to severe LAE was defined as left atrial volume index (LAVI) &gt;35 cc/m2 on trans-thoracic echocardiogram. Patients were divided into two groups based on LAVI &gt;35 cc/m2; thirty day holter monitoring results were retrieved in each group. Along with paroxysmal atrial fibrillation, detection of other supraventricular arrhythmias associated with subsequent development of PAF (i.e., supraventricular tachycardia (SVT) and atrial ectopy) was compared between the groups. Categorical variables were analyzed with Fischer’s exact test and continuous variables using independent sample T-test. Results: 136 patients met our inclusion criteria, of which 76 (55.9%) completed 30 days of monitoring. Moderate to severe LAE was identified in 30 of 76 patients (39.5%). PAF was detected in 4 patients (5.3%). However, PAF was more likely in patients with LAE (13.3% vs 0%, p= 0.21). Detection of any PAF, SVT and atrial ectopy was more likely in patients with LAE (30% vs 6.5%, p= 0.009). Demographics, medical history, clinical and laboratory variables did not differ between patients with and without LAE. Conclusions: In our cohort, the presence of LAE increased the yield of PAF on 30-day holter monitoring. Supraventricular arrhythmias historically associated with subsequent development of atrial fibrillation were more likely in ischemic stroke patients with LAE.

Background and objectives: Atrial fibrillation (AF) is a common cause of cerebral ischemia. Paroxysmal AF is often underdiagnosed by standard diagnostic procedures. Patients with diagnosed AF receive a different secondary prophylactic therapy. This thesis is based on three original publications on improving the detection and prediction of paroxysmal AF in patients with acute cerebral ischemia. In “Enhanced Detection of Paroxysmal Atrial Fibrillation by Early and Prolonged Continuous Holter Monitoring in Patients with Cerebral Ischemia Presenting in Sinus Rhythm” we investigated the diagnostic yield of a 7-day Holter-ECG and the usefulness of an early application. In “Age-dependent yield of screening for undetected atrial fibrillation in stroke patients: the Find-AF study”, we studied the age distribution of paroxysmal AF to find out which age group benefits most from the applied monitoring approach. In “Excessive Supraventricular Ectopic Activity Is Indicative of Paroxysmal Atrial Fibrillation in Patients with Cerebral Ischemia“ we evaluated the predictive value of frequent premature atrial complexes (PAC) and prolonged supraventricular (SV-) runs. Methods: Patients with acute cerebral ischemia were included in the prospective single-center observational trial “Find-AF” (ISRCTN 46104198). Patients without AF at presentation received early and prolonged Holter-ECG-monitoring. To analyse the age distribution of paroxysmal AF, the detection rate was determined in 5-year age clusters from 60 to 85 years. The markers of excessive supraventricular ectopic activity (ESVEA) were analysed in a 24-hour Holter-ECG-interval without AF. The median was chosen as the cut-off level. Results: 281 patients were included. 44 (15.7 %) had AF at baseline. All others received early (median 5.5 hours after presentation) and prolonged (median 6.7 days) Holter-ECG-monitoring. 28 (12.5 %) of 224 patients showed newly diagnosed paroxysmal AF, significantly more than in any 48-hour (6.4 %, p = 0.023) or 24-hour interval (4.8 %, p = 0.015). The detection rates within the seven 24-hour intervals were not significantly different. The detection rate of paroxysmal AF significantly increased with age (p = 0.004), while the number needed to screen to find one patient with paroxysmal AF decreased from 18 (< 60 years) to 3 (≥ 85 years). Patients with frequent PAC (> 4/hour) and those with prolonged SV-runs (> 5 beats) showed significantly more paroxysmal AF: 19.6 % vs. 2.8 % for frequent PAC (p = 0.001) and 17.0 % vs. 4.9 % for prolonged SV-runs (p = 0.003). Multivariate analyses including various clinical AF-predictors confirmed the independent predictive value of both markers. Conclusions: 1.) In patients with acute cerebral ischemia, prolonging the Holter-ECG-interval significantly increases the detection rate of paroxysmal AF. The early initiation appears to be of less importance. 2.) The detection rate of paroxysmal AF increases with age, therefore prolonged Holter-ECG-monitoring is most efficient in elderly patients. 3.) Both frequent PAC and prolonged SV-runs are valid predictors of paroxysmal AF in patients with acute cerebral ischemia.

Previous studies have reported a low, approximately 1% to 3%, rate of detection of occult atrial fibrillation (AF) with Holter monitor in patients with acute stroke. Furthermore, at least one study has reported that Holter monitoring could not always corroborate initial electrocardiographic (ECG) detection of AF suggesting underestimation of AF by Holter. We compare the detection of new-onset AF by serial ECG assessments and Holter after acute ischemic stroke. One hundred forty-four patients with ischemic stroke admitted to a stroke unit were studied. The number of ECGs conducted within the first 3 days up to the detection of AF as well as the time interval for Holter "hookup" and subsequent reporting of AF was documented. ECGs were performed in 143 patients with a baseline of 10 (7%) patients having a history of AF. Serial ECGs detected 15 new AF cases in <2 days of admission, thereby increasing the total number of known AF cases to 25 (17.5%), a 2.6-fold increased realization of AF (P=0.011). Holter was also completed in 12 of 15 new cases of AF but surprisingly identified AF in only 50% (6 of 12). Holter monitoring was performed in 126 cases and in this subgroup, there was no statistically significant difference in the rate of AF detection with ECG or Holter. Serial ECG assessments within the first 72 hours of an acute stroke significantly improve detection of AF. The discordance regarding the corroboration of AF by Holter in ECG-positive patients with AF supports previous observations and suggests a high incidence of paroxysmal AF as a cause of ischemic stroke.

Background and Objectives: Atrial fibrillation (AF) is one of the most common arrhythmias clinically. Aging tends to increase the risk of AF, which also increases the burden of other comorbidities, including coronary artery disease (CAD), and even heart failure (HF). The precise detection of AF is a challenge due to its intermittence and unpredictability. A method for the accurate detection of AF is still needed. Methods: A deep learning model was used to detect atrial fibrillation. Here, a distinction was not made between AF and atrial flutter (AFL), both of which manifest as a similar pattern on an electrocardiogram (ECG). This method not only discriminated AF from normal rhythm of the heart, but also detected its onset and offset. The proposed model involved residual blocks and a Transformer encoder. Results and Conclusions: The data used for training were obtained from the CPSC2021 Challenge, and were collected using dynamic ECG devices. Tests on four public datasets validated the availability of the proposed method. The best performance for AF rhythm testing attained an accuracy of 98.67%, a sensitivity of 87.69%, and a specificity of 98.56%. In onset and offset detection, it obtained a sensitivity of 95.90% and 87.70%, respectively. The algorithm with a low FPR of 0.46% was able to reduce troubling false alarms. The model had a great capability to discriminate AF from normal rhythm and to detect its onset and offset. Noise stress tests were conducted after mixing three types of noise. We visualized the model’s features using a heatmap and illustrated its interpretability. The model focused directly on the crucial ECG waveform where showed obvious characteristics of AF.

Extended cardiac monitoring immediately after acute ischemic stroke (AIS) increases paroxysmal atrial fibrillation (PAF) detection, but its reliability for detection or exclusion of longer term paroxysmal PAF is unknown. We evaluated the positive and negative predictive value (PPV and NPV) of AF detection early after AIS, for PAF confirmation 90 days later. We investigated 49 patients within 7 days of AIS for PAF according to current guidelines; 23 patients received 7 days of additional noninvasive cardiac event monitoring with an R-test device early after their stroke (ISRCTN 97412358). Ninety days after AIS, everyone underwent 7 days of cardiac event monitoring. We calculated the PPV and NPV of immediate PAF detection through extended cardiac event monitoring and through any investigative modality, for the presence of PAF on the 90-day event monitor. PAF detected by a 7-day event monitor within 2 weeks of AIS had a PPV of 100% (95% confidence interval, 72%-100%) for PAF confirmation after 90 days. NPV after 7 days of event monitoring was 64% (95% confidence interval, 35%-87%). PAF detected early through any modality had a PPV of 100% (95% confidence interval, 76%-100%). However, the NPV in the absence of R-test monitoring was only 42% (95% confidence interval, 28%-58%). AF detection through any means immediately after stroke holds strong PPV for confirmation after 90 days, justifying treatment decisions on early monitoring alone. However, failure to identify AF through early monitoring has only modest NPV even after 7 days of monitoring; repeated investigation is desirable.

Introduction Artificial intelligence (AI), particularly deep learning (DL), has demonstrated high performance in various diagnostic problems, including the detection of paroxysmal atrial fibrillation (AF) from electrocardiograms (ECGs). In mobile applications, these approaches can help to detect AF at an early stage and thus prevent possible secondary diseases. However, due to their black-box nature, DL approaches lack explainability and interpretability. Furthermore, 12-lead ECGs are the clinical standard, but mobile devices usually only provide 1–3 leads, which are similar but not identical to Einthoven leads and differ in morphology. Purpose We apply a recently introduced explainable ECG analysis architecture (xECGArch) to single-lead ECGs to verify its performance in mobile applications. xECGArch can learn rhythmic and morphological characteristics using two parallel convolutional neural networks of different dimensions (Fig. 1). When combined with methods from explainable AI, the classification can be traced back to rhythmic and morphological characteristics. Methods We used a subset of 9 854 ECGs (n(AF) = n(non-AF) = 4 927) from 4 public databases (Chapman-Shaoxing, CPSC2018, Georgia, and PTB-XL). Noise was removed using a cascade of high-pass filtering at 0.3 Hz and discrete wavelet transform. For each lead, we trained xECGArch on single-lead ECGs of 10 s length to classify AF vs. non-AF using 90% of the ECGs with 5-fold cross-validation and tested it on the remaining randomly selected ECGs (n(AF) = 507, n(non-AF) = 479). Consistent data splitting was ensured for all leads. Model explanations were generated using deep Taylor decomposition. Results Both models achieved minimum 91.1% accuracy for each ECG lead (Fig. 1). On lead I, the morphology model outperformed the rhythm model with 92.8% vs. 91.5% accuracy, while the rhythm model slightly outperformed the morphology model on leads II and III with 94.2%–95.0% accuracy vs. 91.1%–94.0%. Both models achieved maximum accuracy of 95.0% (rhythm) and 94.0% (morphology) on lead II. By combining both models, xECGArch reached 93.7%–95.1% accuracy. The model explanations confirm that the rhythm model primarily considers QRS complexes as relevant, while the morphology model focuses on fibrillatory waves (Fig. 1). Conclusions Both rhythm and morphology models achieved reliable results across all leads, competing with the state-of-the-art for automatic AF detection from single-lead ECGs. The combination of both models in xECGArch further increased the accuracy. xECGArch is based on the medical reading of ECGs by considering rhythm and morphology characteristics and is therefore interpretable by design. The model explanations are consistent with the diagnostic criteria for AF (Fibrillatory waves and absolute arrhythmia) across leads. With high accuracy on different single-lead ECGs and improved interpretability, xECGArch provides a trustworthy method for ECG analysis in mobile applications.

The detection of Paroxysmal Atrial Fibrillation (PAF) is a fairly complex process performed manually by cardiologists or electrophysiologists by reading an electrocardiogram (ECG). Currently, computational techniques for automatic detection based on fast Fourier transform (FFT), Bayes optimal classifier (BOC), k-nearest neighbors (K-NNs), and artificial neural network (ANN) have been proposed. In this study, six features were obtained based on the morphology of the P-Wave, the QRS complex and the heart rate variability (HRV) of the ECG. The performance of this methodology was validated using clinical ECG signals from the Physionet arrhythmia database MIT-BIH. A feedforward neural network was used to detect the presence of PAF reaching a general accuracy of 97.4%. The results obtained show that the inclusion of the information of the P-Wave, HRV and QR Electrical alternans increases the accuracy to identify the PAF event compared to other works that use the information of only one or at most two of them.