PPG-AFNet: a lightweight and intelligible network for atrial fibrillation identification using photoplethysmography signals

Photoplethysmography (PPG) signal is potentially suitable in atrial fibrillation (AF) detection for its convenience in use and similarity in physiological origin to electrocardiogram (ECG). There are a few preceding studies that have shown the possibility of using the peak-to-peak interval of the PPG signal (PPIp) in AF detection. However, as a generalized model, the accuracy of an AF detector should be pursued on the one hand; on the other hand, its generalizability should be paid attention to in view of the individual differences in PPG manifestation of even the same arrhythmia and the existence of sub-types. Moreover, a binary classifier for atrial fibrillation and normal sinus rhythm is not convincing enough for the similarity between AF and ectopic beats. In this study, we project the atrial fibrillation detection as a multiple-class classification and try to propose a training pipeline that is advantageous both to the accuracy and generalizability of the classifier by designing and determining the configurable options of the pipeline, in terms of input format, deep learning model (with hyperparameter optimization), and scheme of transfer learning. With a rigorous comparison of the possible combinations of the configurable components in the pipeline, we confirmed that first-order difference of heartbeat sequence as the input format, a 2-layer CNN–1-layer Transformer hybridR model as the learning model and the whole model fine-tuning as the implementing scheme of transfer learning is the best combination for the pipeline (F1 value: 0.80, overall accuracy: 0.87)R.

A correct and early diagnosis of cardiac arrhythmias could improve patients' quality of life. The aim of this study is to classify the cardiac rhythm (atrial fibrillation, AF, or normal sinus rhythm NSR) from the photoplethysmographic (PPG) signal and assess the effect of the observation window length. Simulated signals are generated with a PPG simulator previously proposed. The different window lengths taken into account are 20, 30, 40, 50, 100, 150, 200, 250 and 300 beats. After systolic peak detection algorithm, 10 features are computed on the inter-systolic interval series, assessing variability and irregularity of the series. Then, feature selection was performed (using the sequential forward floating search algorithm) which identified two variability parameters (Mean and rMSSD) as the best selection. Finally, the classification by linear support vector machine was performed. Using only two features, accuracy was very high for all the analyzed observation window lengths, going from 0.913±0.055 for length equal to 20 to 0.995±0.011 for length equal to 300 beats.Clinical relevance These preliminary results show that short PPG signals (20 beats) can be used to correctly detect AF.

Atrial Fibrillation is the most common cardiac arrhythmia in adults and is associated with an increased risk of stroke and other cardiovascular diseases. Early detection of atrial fibrillation is crucial for timely intervention and improved patient outcomes. While electrocardiography is the clinical gold standard for atrial fibrillation detection, it is not suitable for long-term and home monitoring due to its requirement for electrode-skin contact. Photoplethysmography, on the other hand, offers a more convenient alternative for continuous monitoring using wearable devices. This study investigates the reliability of the pulse deficit in photoplethysmography signals as an indicator for atrial fibrillation. We employ a wavelet transformation to enhance the visibility of the pulse deficit in photoplethysmography signals and train a deep neural network to classify atrial fibrillation based on these transformed signals. A five-fold cross validation revealed an average AUC of 0.975 and an F1 score of 0.935, indicating a high level of accuracy and reliability. The networks's predictions are further investigated using the gradient-weighted class activation mapping approach to validate, whether the classification is based on the pulse deficit. Our work succefully proves that the pulse deficit in photoplethysmography signals can serve as a robust indicator for atrial fibrillation. Nevertheless, the impact of other heart rate-related characteristics on the classification could not be entirely excluded.

BackgroundStudies have reported the use of photoplethysmography signals to detect atrial fibrillation; however, the use of photoplethysmography signals in classifying multiclass arrhythmias has rarely been reported. Our study investigated the feasibility of using photoplethysmography signals and a deep convolutional neural network to classify multiclass arrhythmia types.Methods and ResultsECG and photoplethysmography signals were collected simultaneously from a group of patients who underwent radiofrequency ablation for arrhythmias. A deep convolutional neural network was developed to classify multiple rhythms based on 10‐second photoplethysmography waveforms. Classification performance was evaluated by calculating the area under the microaverage receiver operating characteristic curve, overall accuracy, sensitivity, specificity, and positive and negative predictive values against annotations on the rhythm of arrhythmias provided by 2 cardiologists consulting the ECG results. A total of 228 patients were included; 118 217 pairs of 10‐second photoplethysmography and ECG waveforms were used. When validated against an independent test data set (23 384 photoplethysmography waveforms from 45 patients), the DCNN achieved an overall accuracy of 85.0% for 6 rhythm types (sinus rhythm, premature ventricular contraction, premature atrial contraction, ventricular tachycardia, supraventricular tachycardia, and atrial fibrillation); the microaverage area under the microaverage receiver operating characteristic curve was 0.978; the average sensitivity, specificity, and positive and negative predictive values were 75.8%, 96.9%, 75.2%, and 97.0%, respectively.ConclusionsThis study demonstrated the feasibility of classifying multiclass arrhythmias from photoplethysmography signals using deep learning techniques. The approach is attractive for population‐based screening and may hold promise for the long‐term surveillance and management of arrhythmia.RegistrationURL: www.chictr.org.cn. Identifier: ChiCTR2000031170.

BackgroundWe aimed to evaluate a novel method of atrial fibrillation (AF) screening using an iPhone camera to detect and analyze photoplethysmographic signals from the face without physical contact by extracting subtle beat‐to‐beat variations of skin color that reflect the cardiac pulsatile signal.Methods and ResultsPatients admitted to the cardiology ward of the hospital for clinical reasons were recruited. Simultaneous facial and fingertip photoplethysmographic measurements were obtained from 217 hospital inpatients (mean age, 70.3±13.9 years; 71.4% men) facing the front camera and with an index finger covering the back camera of 2 independent iPhones before a 12‐lead ECG was recorded. Backdrop and background light intensity was monitored during signal acquisition. Three successive 20‐second (total, 60 seconds) recordings were acquired per patient and analyzed for heart rate regularity by Cardiio Rhythm (Cardiio Inc, Cambridge, MA) smartphone application. Pulse irregularity in ≥1 photoplethysmographic readings or 3 uninterpretable photoplethysmographic readings were considered a positive AF screening result. AF was present on 12‐lead ECG in 34.6% (n=75/217) patients. The Cardiio Rhythm facial photoplethysmographic application demonstrated high sensitivity (95%; 95% confidence interval, 87%–98%) and specificity (96%; 95% confidence interval, 91%–98%) in discriminating AF from sinus rhythm compared with 12‐lead ECG. The positive and negative predictive values were 92% (95% confidence interval, 84%–96%) and 97% (95% confidence interval, 93%–99%), respectively.ConclusionsDetection of a facial photoplethysmographic signal to determine pulse irregularity attributable to AF is feasible. The Cardiio Rhythm smartphone application showed high sensitivity and specificity, with low negative likelihood ratio for AF from facial photoplethysmographic signals. The convenience of a contact‐free approach is attractive for community screening and has the potential to be useful for distant AF screening.

BackgroundWearable devices have evolved as screening tools for atrial fibrillation (AF). A photoplethysmographic (PPG) AF detection algorithm was developed and applied to a convenient smartphone-based device with good accuracy. However, patients with paroxysmal AF frequently exhibit premature atrial complexes (PACs), which result in poor unmanned AF detection, mainly because of rule-based or handcrafted machine learning techniques that are limited in terms of diagnostic accuracy and reliability.ObjectiveThis study aimed to develop deep learning (DL) classifiers using PPG data to detect AF from the sinus rhythm (SR) in the presence of PACs after successful cardioversion.MethodsWe examined 75 patients with AF who underwent successful elective direct-current cardioversion (DCC). Electrocardiogram and pulse oximetry data over a 15-min period were obtained before and after DCC and labeled as AF or SR. A 1-dimensional convolutional neural network (1D-CNN) and recurrent neural network (RNN) were chosen as the 2 DL architectures. The PAC indicator estimated the burden of PACs on the PPG dataset. We defined a metric called the confidence level (CL) of AF or SR diagnosis and compared the CLs of true and false diagnoses. We also compared the diagnostic performance of 1D-CNN and RNN with previously developed AF detectors (support vector machine with root-mean-square of successive difference of RR intervals and Shannon entropy, autocorrelation, and ensemble by combining 2 previous methods) using 10 5-fold cross-validation processes.ResultsAmong the 14,298 training samples containing PPG data, 7157 samples were obtained during the post-DCC period. The PAC indicator estimated 29.79% (2132/7157) of post-DCC samples had PACs. The diagnostic accuracy of AF versus SR was 99.32% (70,925/71,410) versus 95.85% (68,602/71,570) in 1D-CNN and 98.27% (70,176/71,410) versus 96.04% (68,736/71,570) in RNN methods. The area under receiver operating characteristic curves of the 2 DL classifiers was 0.998 (95% CI 0.995-1.000) for 1D-CNN and 0.996 (95% CI 0.993-0.998) for RNN, which were significantly higher than other AF detectors (P<.001). If we assumed that the dataset could emulate a sufficient number of patients in training, both DL classifiers improved their diagnostic performances even further especially for the samples with a high burden of PACs. The average CLs for true versus false classification were 98.56% versus 78.75% for 1D-CNN and 98.37% versus 82.57% for RNN (P<.001 for all cases).ConclusionsNew DL classifiers could detect AF using PPG monitoring signals with high diagnostic accuracy even with frequent PACs and could outperform previously developed AF detectors. Although diagnostic performance decreased as the burden of PACs increased, performance improved when samples from more patients were trained. Moreover, the reliability of the diagnosis could be indicated by the CL. Wearable devices sensing PPG signals with DL classifiers should be validated as tools to screen for AF.

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.

Atrial fibrillation (AF) is the most common cardiac arrhythmia in the world, associated with increased risk of thromboembolic events and an increased mortality rate. In addition, a significant portion of AF patients are asymptomatic. Current AF diagnostic methods, often including a body surface electrocardiogram or implantable loop recorder, are both expensive and invasive and offer limited access within the general community. Objective: We tested the feasibility of the detection of AF using a photoplethysmographic signal acquired from an inexpensive, non-invasive earlobe photoplethysmographic sensor. This technology can be implemented into wearable devices and would enable continuous cardiac monitoring capabilities, greatly improving the rate of asymptomatic AF detection. Approach: We conducted a clinical study of patients going through electrical cardioversion for AF treatment. Photoplethysmographic recordings were taken from these AF patients before and after their cardioversion procedure, along with recordings from a healthy control group. Using these recordings, cardiac beats were identified and the inter-systolic interval was calculated. The inter-systolic interval was used to calculate four parameters to quantify the heart rate variability indicative of AF. Receiver operating characteristic curves were used to calculate discriminant thresholds between the AF and non-AF cohorts. Main results: The parameter with the greatest discriminant capability resulted in a sensitivity and specificity of 90.9%. These results are comparable to expensive ECG-based and invasive implantable loop recorder AF detection methods. Significance: These results demonstrate that using a non-invasive earlobe photoplethysmographic signal is a viable and inexpensive alternative to ECG-based AF detection methods, and an alternative that could be invaluable in detecting subclinical AF.

Background Atrial fibrillation (AFIB) is a common atrial arrhythmia that affects millions of people worldwide. However, most of the time, AFIB is paroxysmal and can pass unnoticed in medical exams; therefore, regular screening is required. This paper proposes machine learning (ML) methods to detect AFIB from short-term electrocardiogram (ECG) and photoplethysmography (PPG) signals. Aim Several experiments were conducted across five different databases, with three of them containing ECG signals and the other two consisting of only PPG signals. Experiments were conducted to investigate the hypothesis that an ML model trained to predict AFIB from ECG segments could be used to predict AFIB from PPG segments. Materials and methods A random forest (RF) ML algorithm achieved the best accuracy and achieved a 90% accuracy rate on the University of Mississippi Medical Center (UMMC) dataset (216 samples) and a 97% accuracy rate on the Medical Information Mart for Intensive Care (MIMIC)-III datasets (2,134 samples). Results A total of 269,842 signal segments were analyzed across all datasets (212,266 were of normal sinus rhythm (NSR) and 57,576 corresponded to AFIB segments). Conclusions The ability to detect AFIB with significant accuracy using ML algorithms from PPG signals, which can be acquired via non-invasive contact or contactless, is a promising step forward toward the goal of achieving large-scale screening for AFIB.

Atrial fibrillation detection on reconstructed photoplethysmography signals collected from a smartwatch using a denoising autoencoder

Atrial Fibrillation (AF) is the most common cardiac arrhythmia in the world, associated with an increased risk of thromboembolic events and an increased mortality rate. Due to the frequently asymptomatic nature of AF, a significant portion of AF is subclinical. To address this issue, we tested the feasibility of detecting AF using photoplethysmographic signal acquired from a noninvasive earlobe photoplethysmographic sensor. Photoplethysmographic recordings were taken from AF patients before and after cardioversion procedures, along with recordings from a healthy control group. This signal was analyzed and multiple parameters of heart rate variability were calculated. The parameter with the highest discriminant capability resulted in a sensitivity and specificity of 90.9%. These results show that using earlobe photoplethysmographic signal is a viable, inexpensive, and non-invasive AF detection method that could be invaluable in detecting subclinical AF.

Cardiac arrhythmias (CAs) are associated with critical heart-related complications such as stroke or heart failure. Because of the intermittent and asymptomatic presentation of some CAs in their early stages, the screening remains limited using traditional methods based on ECG. Recent advances in photoplethysmography (PPG) have revealed substantial potential for wearable devices to detect CAs within large populations. Although PPG has demonstrated efficiency in discriminating atrial fibrillation (AF) from normal sinus rhythm, it remains unclear whether AF detection methods remains effective in the presence of CAs other than AF. In this study, we applied a recurrent neural network on a dataset containing eight different types of CAs (Table 1). The classifier processes sequences of interbeat intervals (IBIs) as input and discriminates between normal and abnormal rhythms. The algorithm was evaluated on 64 patients (45 males and 19 females , with a mean age of 55.9 ± 16.0) undergoing a diagnostic or therapeutic electrophysiological procedure. This dataset includes simultaneous recordings of PPG signals from a wrist-bracelet and 12-lead ECG, with the latter used as the gold-standard for annotating cardiac rhythms. The classifier achieved 84% accuracy, 77% sensitivity and 88% specificity in detecting abnormal rhythms. Table 1 shows that AF (99.6%), atrial tachycardia (100%) and AVRT (96.4%) were well classified as abnormal rhythm, whereas the reliability of the detection decreased for atrial flutter (65.4%), atrial or ventricular bigeminy (72.4%) and ventricular tachycardia (80.2%). Undetected abnormal rhythms were often characterized by a rather regular rhythm (illustrated by VT in Figure 1) and CAs for which PPG heartbeat detection was initially not optimal (Bigeminy in Figure 1). In conclusion, this study shows the capability of PPG-based technology to detect various CAs extending beyond AF. It highlights the merits and limitations of IBI-based detection of abnormal rhythms, highlighting the need for a better comprehension of the peripheral hemodynamic signature of CAs.

Atrial fibrillation (AF) detection from wristwatch is important as it can lead to non-invasive, long-term and continuous monitoring of AF from photoplethysmogram (PPG) signal. In this paper, we propose a novel method not only to detect AF from wristwatch PPG, but also to automatically distinguish between clean and corrupted PPG segments. We use accelerometer data as well as variable frequency complex demodulation based time-frequency analysis of the PPG signal to detect motion and noise artifacts in the PPG signal waveform. Next, root mean square of successive differences and sample entropy are extracted from the beat-to-beat intervals of the clean detected PPG signals, which we use to separate AF from normal sinus rhythm. UMass dataset consisting of 20 subjects has been used in this study to test the efficacy of the proposed algorithm. Our method achieves sensitivity, specificity and accuracy of 96.15%, 97.37% and 97.11%, respectively, which shows the potential of a practical and reliable AF monitoring scheme.

We propose an efficient approach based on a convolutional denoising autoencoder (CDA) network to reduce motion and noise artifacts (MNA) from corrupted atrial fibrillation (AF) and non-AF photoplethysmography (PPG) data segments so that an accurate PPG-signal-derived heart rate can be obtained. Our method's main innovation is the optimization of the CDA performance for both rhythms using more AF than non-AF data for training the AF-specific CDA model and vice versa for the non-AF CDA network. To evaluate this unconventional training scheme, our proposed network was trained and tested on 25-sec PPG data segments from 48 subjects from two different databases-the Pulsewatch dataset and Stanford University's publicly available PPG dataset. In total, our dataset contains 10,773 data segments: 7,001 segments for training and 3,772 independent segments from out-of-sample subjects for testing. Using real-life corrupted PPG segments, our approach significantly reduced the average heart rate root mean square error (RMSE) of the reconstructed PPG segments by 45.74% and 23% compared to the corrupted non-AF and AF data, respectively. Further, our approach exhibited lower RMSE, and higher sensitivity and PPV for detected peaks compared to the reconstructed data produced by the alternative methods. These results show the promise of our approach as a reliable denoising method, which should be used prior to AF detection algorithms for an accurate cardiac health monitoring involving wearable devices. PPG signals collected from wearables are vulnerable to MNA, which limits their use as a reliable measurement, particularly in uncontrolled real-life environments.

Cardiovascular diseases are the primary cause of deaths in the world. Atrial fibrillation (AF) is the most common type of cardiac arrhythmia. Due to its high prevalence and associated risks, early detection of AF is an important objective for healthcare systems worldwide. The growing demand for medical assistance implies increased expenses, which could be limited by implementing ambulatory monitoring techniques based on wearable devices, thus, reducing the number of people requiring observation in hospitals. One of the main challenges in this context is related to the large amount of data from patients to be analyzed, which points to the suitability of using computational intelligence techniques for it. The selection of the features to be extracted from data plays a key role in order for any classifier of heart rhythm to provide good results in this regard. This paper demonstrates that it is possible to achieve an accurate detection of AF using a very low number of relatively simple features extracted from photoplethysmographic signals, enabling the use of affordable wearable devices (with scarce processing and data storage resources) with this purpose over long periods of time. This fact has been validated in experiments using data from real patients under medical supervision.