Atrial fibrillation can be treated using low voltage (LV) (amplitude of intracardiac electrogram < 0.5mV) targeted ablation. However, catheter characteristics can alter the voltage leading to changes in identified LV areas. This study evaluates the impact electrode size has on the voltage in healthy and diseased tissue. A realistic setup was generated of tissue, bath and two high conductivity electrodes, with centre to centre spacing of 2mm, placed in contact to the tissue and perpendicular to the planar wavefront. Simulations were performed varying the dimensions of the cubic electrodes from 0.2 to 1.6mm in healthy tissue and including fibrosis in different locations. An inverse relationship was found between the electrode size and the voltage. When including epicardial fibrosis, a voltage decrease of 1 mV was found in electrodes. When fibrosis was placed closer to the electrodes, a morphological signal change was seen and a 9 mV drop in voltage for small electrodes. Large electrodes deliver smaller voltages. A fibrotic area on the epicardial side has a small influence on the voltage, which was not amplified by increasing electrode size. Endocardial fibrosis delivers significantly smaller voltages than healthy tissue. Little difference in the voltage was seen between large electrodes (>1 mm) in diseased tissue. Electrode size needs to be accounted for when determining LV areas using different catheters.
Due to the cost-efficiency of the ECG, the interest in noninvasive techniques to assess atrial fibrillation (AF) electrophysiological complexity is increasingly high.Still, ECG-based methods to measure AF complexity are limited in clinical practice and need estimation of the atrial activity (AA) signal from sufficiently long ECG recordings.The present work proposes an algorithm for tensor decomposition called constrained alternating group lasso (CAGL) as a noninvasive tool to quantify AF complexity.Experiments with a database of 59 ECG recordings from 20 patients suffering from persistent AF show that CAGL is able to both extract the AA and quantify its complexity from very short ECG recordings (1.06 ± 0.20 s).All the patients had undergone step-wise catheter ablation (CA) that ended in procedural AF termination.CAGL is applied on the ECG recording before CA and at each step procedure, measuring the rank of the tensor that provides the AA signal.It is observed that such rank decreases at each step of the CA procedure, showing a less complex AA signal as the ablation is performed.A statistical correlation between AA complexity measured by the new index and AF recurrence after CA is observed.The proposed index is a potential tool to guide CA procedures in real time.
We recently proposed a Sympathetic Activity Index (SAI) through an orthonormal Laguerre expansion of linear RR-interval autoregressive kernels. The resulting heart rate variability (HRV) instantaneous indices may be used for the effective estimation of cardiac sympathetic outflow because the model is independent from the overlapping dynamics of the sympathetic and vagus nerves in the low frequency (LF) band (0.04-0.15Hz). In this study, we perform a preliminary validation of the SAI performance through concurrent estimates from efferent muscle sympathetic nerve activity (MSNA) recordings. ECG and MSNA were simultaneously recorded in 12 hypertensive patients during a 10min resting state in the supine position and up to 30min sodium nitroprusside (SNP, 0.4 µg/kg per minute) administration. Results show a characteristic increase of the MSNA during SNP intake with respect to the resting state (p< 0.001). While SAI was associated with a significant increase during SNP (p<0.001), LF power did not show significant changes between sessions (p>0.05). Spearman analysis highlighted a significant correlation between SAI and MSNA (r=0.47; p<0.02) and a nonsignificant correlation between LF power and MSNA (r=-0.09; p>0.05). This study provides a further validation step of the SAI and supports the use of HRV parameters as a reliable proxy of cardiac sympathetic outflow dynamics.
Cardiac abnormalities are a leading cause of death and their early diagnosis are of importance for providing timely interventions. The goal of 2020 PhysioNetlCinC challenge was to develop algorithms to diagnose multiple cardiac abnormalities using 12-lead ECG data. In this work, we develop a wide and deep transformer neural network to classify each 12-lead ECG sequence into 27 cardiac abnormality classes. Our approach combines handcrafted ECG features, which were determined to be important by a random forest model, and discriminative feature representations that are automatically learned from a transformer neural network. Our entry to the 2020 Phys-ioN etlCinC challenge placed 1st out of 41 official ranking teams (team name = prna). Using the official generalized weighted accuracy metric for evaluation, we achieved a validation score of 0.587 and top score of 0.533 on the full held-out test set.
Electrocardiogram (ECG) analysis is the standard of care for the diagnosis of irregular heartbeat patterns, known as arrhythmias.This paper presents a deep learning system for the automatic detection and multilabel classification of arrhythmias in ECG recordings.Our system composes three differentiable operators: a scattering transform (ST), a depthwise separable convolutional network (DSC), and a bidirectional long short-term memory network (BiLSTM).The originality of our approach is that all three operators are implemented in Python.This is in contrast to previous publications, which precomputed ST coefficients in MATLAB.The implementation of ST on Python was made possible by using a new software library for scattering transform named Kymatio.This paper presents the first successful application of Kymatio to the analysis of biomedical signals.As part of the PhysioNet/Computing in Cardiology Challenge 2020, we trained our hybrid Scattering-LSTM model to classify 27 cardiac arrhythmias from two databases of 12-lead ECGs: CPSC2018 and PTB-XL, comprising 32k recordings in total.Our team "BitScattered" achieved a Challenge metric of 0.536±0.012over ten folds of cross-validation but this result may be over-optimistic since we were not able to rank and score on the hidden test set.
Genome-wide association studies (GWAS) have discovered hundreds of genetic loci for resting heart rate (RHR). However, the impact of intra-individual variation in RHR on GWAS results is unclear. We evaluated this impact by analyzing two RHR recordings from N ~61,000 subjects from UK Biobank. In addition, we modelled variations in RHR as independent white zero-mean Gaussian noise with a standard deviation of 0.5x, 1x, and 2x the standard deviation of the difference between the original RHR values (4,8, and 16 bpm, respectively). The two original RHR recordings were highly correlated (ρ =0.77), but results from the genetic analyses were s lightly different: the number of genome-wide significant (p < 5x10−8) variants at the locus with the strongest reported association (MYH6): n=39 vs. n=34; the p-value of the corresponding lead-variant, 3.6x10−24 vs. 2.1x10−19; and the estimated heritability 20.0% vs. 16.7%. Simulated data showed an inverse relationship between RHR variation and genetic association strength and heritability. Results formally demonstrate the impact of intra-individual RHR variability on the discovery of genetic variants in single-measurement studies.
Cardiovascular remodeling induced by intrauterine growth restriction manifests in adulthood by more globular ventricles, as evidenced by in vivo measurements. The angle between the dominant vectors of the QRS and T-wave loops has been reported to be significantly altered as a result of the induced remodeling. To investigate whether the more globular ventricular shape was a major factor contributing to such alteration, we performed electrophysiological simulations in a human biventricular model for control and in a model obtained by deforming the control one to represent a more spherical left ventricle (SLV). Transmural ventricular heterogeneities and a Purkinje network were included. 12-lead ECGs were calculated, from which spatial QRS and T-wave angles were computed. The angle between the T-wave and the XZ-plane was found to increase in the SLV model, showing a variation similar to that reported in in vivo studies. However, the angle between the dominant vectors of the QRS and T-wave loops projected onto the XY-plane was lower for control, contrary to clinical observations in IUGR adults. Other clinical results could not be reproduced in our simulations either. Our findings suggest that a more globular left ventricular shape leads to changes in the angles of QRS and T-wave loops, but further research is needed to fully understand these changes and the underlying mechanisms.
In this study we present a system for automated processing of signals from the polysomnogram (PSG) for the detection of apnoea and non-apnoea arousals. The PSG signals were divided into 15 second epochs and 59 time- and frequency-domain features were derived for each epoch. Features from adjacent 4 epochs were combined and processed with a bank of ten feed-forward neural networks each with a single hidden layer of 20 units. The system outputs a 200 Hz annotation signal containing probability estimates that each sample was associated with an apnoea or non-apnoea arousal, or no-arousal. Data from the Physionet Computing in Cardiology Challenge 2018 was used to develop and test the system. Performance of the system was assessed using three class and two class metrics. With the system classifying three classes, the volume under the receiver operator characteristic (ROC) surface was 0.75 with an optimal specificity of 0.72, a sensitivity of 0.76 for the apnoea arousals, and a sensitivity of 0.69 for the non-apnoea arousals. When the two arousal classes were combined into one arousal class, the area under the precision recall curve was 0.74, the area under the ROC curve was 0.91, with an optimal specificity and sensitivity of 0.85.
Weaning is the process of withdrawing mechanical ventilation at the Intensive Care Units. The problem is that around 20% of weaned patients were not actually ready for discontinuation. Studies suggest that vagal dysfunction is lower in patients successfully weaned. Therefore, the Baroreflex Sensitivity (BRS) and Heart Rate Variability (HRV) are estimated to see if they can provide additional information to improve the prediction of weaning outcomes. 9 successfully weaned patients (S-group) and 6 unsuccessfully weaned (F-group) were monitored in the last hour prior to the Spontaneous Breathing Trial. The BRS is estimated through spectral analysis, to obtain the a parameter in the low and high frequency bands, and through the capacity, C, estimated by the Bivariate Phase Rectified Signal Average (BPRSA) method. The current clinic parameters of weaning readiness do not show statistical differences. However, the capacity to changes of the BRS, C, estimated via BPRSA, exhibits significant differences between the two groups. Negative values of C, and with higher absolute values, were obtained for the S-group. Temporal indices of HRV also show differences, but not significant. These results suggest that BRS should be further explored for predicting weaning outcomes.
Automated ECG classification is a standard feature in many commercial 12-Lead ECG machines. As part of the Physionet/CinC Challenge 2020, our team, “Mad-hardmax”, developed an XGBoost based classification method for the analysis of 12-Lead ECGs acquired from four different countries. Our aim is to develop an interpretable classifier that outputs diagnoses which can be traced to specific ECG features, while also testing the potential of information theoretic features for ECG diagnosis. These measures capture high-level interdependencies across ECG leads which are effective for discriminating conditions with multiple complex morphologies. On unseen test data, our algorithm achieved a challenge score of 0.155 relative to a winning score of 0.533, putting our submission in 24th position from 41 successful entries.
The morphological ECG features for arrhythmia diagnosis are usually identified and combined on different scales. For example, morphological ECG features can be identified on the scale of length or amplitude of QRS waves. Professionals can then make a diagnosis based on the combination of these identified features. Attention-based deep neural networks have been proved to boost meaningful features on different scales and suppress weak features. To boost and combine ECG features on different scales for arrhythmia classification, we proposed MADNN: a multi-scale attention deep neural network for arrhythmia classification. Our proposed network was designed combining kernel-wise and branch-wise attention modules based on a backbone of 1-dimensional convolutional neural networks. MADNN with properly tuned hyper-parameters was tested for arrhythmia classification in the PhysioNet/Computing in Cardiology Challenge 2020. In this challenge, MADNN officially achieved a validation score of 0.446, and a full test set score of 0.236. Our team named Minibus ranked the 18th out of 41 teams.
VoluMetrix has developed a new version of its NIVA, a wrist device that measures pressure variations in the veins together with the photoplethysmogram (PPG) at the same point. Previous studies have shown that the venous pressure signal (NIVA) reflected an increased HF power with respect to the electrocardiogram. This suggests that it may be useful for parasympathetic characterization guided by signal-derived respiration. Performance of NIVA signal is compared to that of PPG in a controlled breathing experiment (8 subjects) with different respiratory rates (6, 12 and 18 bpm), where a downward trend in parasympathetic estimates is expected with increasing respiratory rates. The NIVA signal is able to accurately estimate the respiratory rate (less than 0.03 Hz estimation error) in all the subjects, outperforming the PPG in the same task. In addition, respiratory-guided parasympathetic estimates significantly decreases with increased respiratory rate.
Automatic and fast atrial fibrillation (AF) diagnosis is still a major concern for the healthcare professional. Several algorithms based on univariate and multivariate analysis have been developed to detect AF. Although the published results do show satisfactory detection accuracy, computational complexity of such methods is still questionable. This study proposes an alternative way to diagnosis AF arrhythmia which is based on the combination of seven univariate data analysis-based detectors followed by a majority voting in order to build a digital fingerprint of AF. Four publicly-accessible sets of clinical data were used for AF assessment. The time series were segmented in 10 s RR interval window. The features of the four databases were merged in order to give rise huge variability and therefore to better characterize AF arrhythmia. Afterwards, a receiver operating characteristic curve analysis has been conducted to fix optimal thresholds for AF detection. Finally, the seven obtained detectors have been concatenated and then a majority rule was applied to yield a final decision on AF diagnosis. The results showed that this strategy performed better than some existing algorithms do, with 98.50% for sensitivity and 95.1 % specificity.
Various technologies, such as electrocardiography, optical mapping, and patch clamping, have been developed to monitor cardiac electrophysiological behavior in live tissue.One limitation is that none of the available measurement methods is capable of monitoring simultaneously all quantities, such as intracellular ionic concentrations and ion-channel gating states, that may be important contributors to arrhythmia formation.Data assimilation strategies such as Kalman filtering can be used to fill in missing measurements, but to our knowledge, there have been few comparisons of different state estimation algorithms applied to the same cardiac action potential model.To help develop a framework for comparing performances of estimators, we applied two estimation algorithms, an unscented Kalman filter (UKF) and a gainscheduled Kalman filter (GSKF), to a two-variable Karma model of a cardiac cell.We generated simulated data from the model and compared the abilities of the algorithms to infer the slow variable of the model from measurements of the fast variable and vice versa.The UKF performed well when the process noise variance was low relative to measurement noise, while the opposite was often true for the GSKF, and estimation errors tended to be smaller when the fast variable was chosen as the measurement.
An essential part for the accurate classification of electrocardiogram (ECG) signals is the extraction of informative yet general features, which are able to discriminate diseases.Cardiovascular abnormalities manifest themselves in features on different time scales: small scale morphological features, such as missing P-waves, as well as rhythmical features apparent on heart rate scales.For this reason we incorporate a variant of the complex wavelet transform, called a scatter transform, in a deep residual neural network (ResNet).The former has the advantage of being derived from theory, making it well behaved under certain transformations of the input.The latter has proven useful in ECG classification, allowing feature extraction and classification to be learned in an end-to-end manner.Through the incorporation of trainable layers in between scatter transforms, the model gains the ability to combine information from different channels, yielding more informative features for the classification task and adapting them to the specific domain.For evaluation, we submitted our model in the official phase in the PhysioNet/Computing in Cardiology Challenge 2020.Our (Team Triage) approach achieved a challenge validation score of 0.640, and full test score of 0.485, placing us 4th out of 41 in the official ranking.
To evaluate and assess the cardiovascular system during fetal development in the utero of pregnant mouse, it is essential to understand the effect of mandatory anesthesia treatment on sympathetic and parasympathetic nervous system activities. The preliminary study presented in this paper explores the changes in fetal and maternal Heart Rate Variability (HRV) parameters as well as fetal-maternal Heart Rate (HR) coupling measures during anesthesia. ECG signals of 6 pregnant mice and 10 fetuses were recordedfor 15 min. The obtained ECG signals were segmented into three periods, each for a duration of 5 min. Maternal and fetal HRV parameters in addition to fetal-maternal coupling patterns were computed for each of the three segments of the ECG signals. During the first 10 min, results show that mean and root mean square of successive differences (RMSSD) of maternal HR did not change, but significantly decreased after the first 10 min. A similar result was observed for the mean, RMSSD and standard deviation of NN intervals in fetal HR. On the other hand, no significant changes were observed for the coupling patterns between fetal-maternal heartbeats. These observations suggest that fetal nervous system activities were suppressed by anesthesia treatment applied to pregnant mice for more than 10 min.
In 19 healthy subjects we assessed the effects of chirped respiratory frequency (RF) from 0.05 to 0.8 Hz, and of standing (STC) on the 130-s time courses of the central frequency and power of the high frequency components of RR ( CFE HF RR , PE HF RR ) and of respiration ( CFE HF RES , PE HF RES ), estimated by a time-frequency distribution.We took as indexes of respiratory sinus arrhythmia (RSA) frequency coupling (RSA FC ) the CFE HF RES -CFE HF RR relation, their difference (Δ CFE HF) and coherence (RSA CO ), and the alpha index as RSA sensitivity (RSA S ).The effects of RF on RSA measures were distinctive in three RF ranges, with precise limits at 0.09±0.005,0.18±0.03,0.51±0.10 and 0.81±0.03Hz.In the low, mid and high RF ranges, respectively: CFE HF RR was first unchanged, proportional to RF (r=0.97±0.03),then constant again; RSA CO was 0.73±0.06,0.97±0.03and 0.78±0.08;RSA S was 135±34 ms/l, proportional to RF (r=-0.79±0.08),and 62±30 ms/l; Δ CFE HF was greater than 0.02 Hz in the three RF stages.STC decreased mean RSA S (p<0.02) in all RF stages.RSA FC and RSA S measures vary as function of RF, showing three stages with precise RF limits and distinctive functionality, respectively: low for RSA FC but high for RSA S , optimal and linear for both, and reduced for both measures.Baroreflex activation significantly depresses RSA S .
This study presents a novel non-invasive method to detect His potentials from high-resolution body surface signals.35 patients were included in this study.All patients received an invasive electrophysiological study to determine ground truth His-Ventricular (HV) intervals.Prior to these procedures, body surface potentials were recorded using 128 electrodes sampled at 2048 Hz for 8 minutes.Signal averaging was performed on the body surface signals using only the beats occurring during the exhalation phase of respiration.4 "wide" bipolar (2 electrodes spaced 50 mm) in the vertical, horizontal, right diagonal and left diagonal directions, and 24 Laplacians signals were created and high-pass filtered at 30Hz.Noninvasive HV interval measurements were not performed on 2 (6%) patients because body surface signals were too noisy.His potentials were invisible on the bipolar signals for 5 (14%) patients and on the Laplacian signals for 7 patients (20%).Comparison between invasive (58.5±15ms) and non-invasive HV interval (53.6±14ms) measured with bipolar signals, revealed a squared correlation coefficient (SCC) of 0.66.HV intervals measured with Laplacian signals were less correlated (53.8±18ms) with a SCC of 0.50.This study shows promising results that HV interval can be measured noninvasively using bipolar and Laplacians signals.
Monitoring of cardiac and respiratory activity results crucial in clinical settings. However, specific, expensive and usually wired devices are required, which results in cumbersome configurations. Therefore, there is a large research interest in the development of noninvasive monitoring systems. Some authors have proposed the use of fiber-optic speckle interferometry, enabling for contact-less patient monitoring. In this study, we exploited this technology, and explored the performance of different fiber-optic configurations, in combination with robust signal processing algorithms, for heart and respiratory rates estimation. Several configurations were tested, all of them resulting in very low estimation errors. A qualitative analysis revealed that our system can be suited to track fast changes in respiration, as well as the presence of apneas. Such a system might be useful for noninvasive patient monitoring, specially in overnight recordings (e.g., OSAS or AF) or specific scenarios, such as magnetic resonances.
The fluctuations of the duration of the electrical activity of the heart, measured as the time distance between Q-wave onset and T-wave end (QT), is under autonomic control. We studied the complexity of the QT variability regulation via the computation of sample entropy of QT variability during sympathetic activation induced by graded head-up tilt. Sample entropy was computed over the original QT series and after factorizing it into partial processes describing QT variability related to heart period, measured as the time interval between consecutive R-wave peaks (RR), linked to respiration (R) and unrelated to RR and R. We found that QT variability complexity is high and does not vary with the intensity of the stimulus. This result was the consequence of a non-significant tendency of the complexity of the QT variability related to RR to decrease and a significant raise of the complexity of the QT variability unrelated to RR and R with the magnitude of the orthostatic challenge. We suggest that the sample entropy of the QT variability unrelated to RR and R could quantify the increased heterogeneity of the neural inputs genuinely modulating QT during a sympathetic arousal.
