Considerations on Performance Evaluation of Atrial Fibrillation Detectors

Photoplethysmography based atrial fibrillation detection: a continually growing field

Objective. Atrial fibrillation (AF) is a prevalent cardiac arrhythmia associated with significant health ramifications, including an elevated susceptibility to ischemic stroke, heart disease, and heightened mortality. Photoplethysmography (PPG) has emerged as a promising technology for continuous AF monitoring for its cost-effectiveness and widespread integration into wearable devices. Our team previously conducted an exhaustive review on PPG-based AF detection before June 2019. However, since then, more advanced technologies have emerged in this field.Approach. This paper offers a comprehensive review of the latest advancements in PPG-based AF detection, utilizing digital health and artificial intelligence (AI) solutions, within the timeframe spanning from July 2019 to December 2022. Through extensive exploration of scientific databases, we have identified 57 pertinent studies.Significance. Our comprehensive review encompasses an in-depth assessment of the statistical methodologies, traditional machine learning techniques, and deep learning approaches employed in these studies. In addition, we address the challenges encountered in the domain of PPG-based AF detection. Furthermore, we maintain a dedicated website to curate the latest research in this area, with regular updates on a regular basis.

Intelligent assessment of atrial fibrillation gradation based on sinus rhythm electrocardiogram and baseline information

BACKGROUND

Atrial fibrillation (AF) is a progressive arrhythmia that significantly affects a patient's quality of life. The 4S-AF scheme is clinically recommended for AF management; however, the evaluation process is complex and time-consuming. This renders its promotion in primary medical institutions challenging. This retrospective study aimed to simplify the evaluation process and present an objective assessment model for AF gradation.

METHODS

In total, 189 12-lead electrocardiogram (ECG) recordings from 64 patients were included in this study. The data were annotated into two groups (mild and severe) according to the 4S-AF scheme. Using a preprocessed ECG during the sinus rhythm (SR), we obtained a synthesized vectorcardiogram (VCG). Subsequently, various features were calculated from both signals, and age, sex, and medical history were included as baseline characteristics. Different machine learning models, including support vector machines, random forests (RF), and logistic regression, were finally tested with a combination of feature selection techniques.

RESULTS

The proposed method demonstrated excellent performance in the classification of AF gradation. With an optimized feature set of VCG and baseline features, the RF model achieved accuracy, sensitivity, and specificity of 83.02 %, 80.56 %, and 88.24 %, respectively, under the inter-patient paradigm.

CONCLUSION

Our results demonstrate the value of physiological signals in AF gradation evaluation, and VCG signals were effective in identifying mild and severe AF. Considering its low computational complexity and high assessment performance, the proposed model is expected to serve as a useful prognostic tool for clinical AF management.

Detection of Paroxysmal Atrial Fibrillation from Dynamic ECG Recordings Based on a Deep Learning Model

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.

Statistical and Diagnostic Properties of pRRx Parameters in Atrial Fibrillation Detection

Background: We studied the diagnostic properties of the percentage of successive RR intervals differing by at least x ms (pRRx) as functions of the threshold value x in a range of 7 to 195 ms for the differentiation of atrial fibrillation (AF) from sinus rhythm (SR). Methods: RR intervals were measured in 60-s electrocardiogram (ECG) segments with either AF (32,141 segments) or SR (32,769 segments) from the publicly available Physionet Long-Term Atrial Fibrillation Database (LTAFDB). For validation, we have used ECGs from the Massachusetts Institute of Technology–Beth Israel Hospital (MIT–BIH) Atrial Fibrillation Database. The pRRx distributions in AF and SR in relation to x were studied by histograms, along with the mutual association by the nonparametric Spearman correlations for all pairs of pRRx, and separately for AF or SR. The optimal cutoff values for all pRRx were determined using the receiver operator curve characteristic. A nonparametric bootstrap with 5000 samples was used to calculate a 95% confidence interval for several classification metrics. Results: The distributions of pRRx for x in the 7–195 ms range are significantly different in AF than in SR. The sensitivity, specificity, accuracy, and diagnostic odds ratios differ for pRRx, with the highest values for x = 31 ms (pRR31) rather than x = 50 (pRR50), which is most commonly applied in studies on heart rate variability. For the optimal cutoff of pRR31 (68.79%), the sensitivity is 90.42%, specificity 95.37%, and the diagnostic odds ratio is 194.11. Validation with the ECGs from the MIT–BIH Atrial Fibrillation Database confirmed our findings. Conclusions: We demonstrate that the diagnostic properties of pRRx depend on x, and pRR31 outperforms pRR50, at least for ECGs of 60-s duration.

Non-Contact Atrial Fibrillation Detection From Face Videos by Learning Systolic Peaks

OBJECTIVE

We propose a non-contact approach for atrial fibrillation (AF) detection from face videos.

METHODS

Face videos, electrocardiography (ECG), and contact photoplethysmography (PPG) from 100 healthy subjects and 100 AF patients are recorded. Data recordings from healthy subjects are all labeled as healthy. Two cardiologists evaluated ECG recordings of patients and labeled each recording as AF, sinus rhythm (SR), or atrial flutter (AFL). We use the 3D convolutional neural network for remote PPG monitoring and propose a novel loss function (Wasserstein distance) to use the timing of systolic peaks from contact PPG as the label for our model training. Then a set of heart rate variability (HRV) features are calculated from the inter-beat intervals, and a support vector machine (SVM) classifier is trained with HRV features.

RESULTS

Our proposed method can accurately extract systolic peaks from face videos for AF detection. The proposed method is trained with subject-independent 10-fold cross-validation with 30s video clips and tested on two tasks. 1) Classification of healthy versus AF: the accuracy, sensitivity, and specificity are 96.00%, 95.36%, and 96.12%. 2) Classification of SR versus AF: the accuracy, sensitivity, and specificity are 95.23%, 98.53%, and 91.12%. In addition, we also demonstrate the feasibility of non-contact AFL detection.

CONCLUSION

We achieve good performance of non-contact AF detection by learning systolic peaks.

SIGNIFICANCE

non-contact AF detection can be used for self-screening of AF symptoms for suspectable populations at home or self-monitoring of AF recurrence after treatment for chronic patients.

Detection of Brief Episodes of Atrial Fibrillation Based on Electrocardiomatrix and Convolutional Neural Network

Background: Brief episodes of atrial fibrillation (AF) may evolve into longer AF episodes increasing the chances of thrombus formation, stroke, and death. Classical methods for AF detection investigate rhythm irregularity or P-wave absence in the ECG, while deep learning approaches profit from the availability of annotated ECG databases to learn discriminatory features linked to different diagnosis. However, some deep learning approaches do not provide analysis of the features used for classification. This paper introduces a convolutional neural network (CNN) approach for automatic detection of brief AF episodes based on electrocardiomatrix-images (ECM-images) aiming to link deep learning to features with clinical meaning. Materials and Methods: The CNN is trained using two databases: the Long-Term Atrial Fibrillation and the MIT-BIH Normal Sinus Rhythm, and tested on three databases: the MIT-BIH Atrial Fibrillation, the MIT-BIH Arrhythmia, and the Monzino-AF. Detection of AF is done using a sliding window of 10 beats plus 3 s. Performance is quantified using both standard classification metrics and the EC57 standard for arrhythmia detection. Layer-wise relevance propagation analysis was applied to link the decisions made by the CNN to clinical characteristics in the ECG. Results: For all three testing databases, episode sensitivity was greater than 80.22, 89.66, and 97.45% for AF episodes shorter than 15, 30 s, and for all episodes, respectively. Conclusions: Rhythm and morphological characteristics of the electrocardiogram can be learned by a CNN from ECM-images for the detection of brief episodes of AF.

Identification of Transient Noise to Reduce False Detections in Screening for Atrial Fibrillation

Screening for atrial fibrillation (AF) with a handheld device for recording the ECG is becoming increasingly popular. The poorer signal quality of such ECGs may lead to false detection of AF, often caused by transient noise. Consequently, the need for expert review in AF screening can become extensive. A convolutional neural network (CNN) is proposed for transient noise identification in AF detection. The network is trained using the events produced by a QRS detector, classified into either true beat detections or false detections. The CNN and a low-complexity AF detector are trained and tested using the StrokeStop I database, containing 30-s ECGs from mass screening for AF in the elderly population. Performance evaluation of the CNN-based quality control using a subset of the database resulted in sensitivity, specificity, and accuracy of 96.4, 96.9, and 96.9%, respectively. By inserting the CNN before the AF detector, the false AF detections were reduced by 22.5% without any loss in sensitivity. The results show that the number of recordings calling for expert review can be significantly reduced thanks to the identification of transient noise. The reduction of false AF detections is directly linked to the time and cost spent on expert review.