Noise Maps for Quantitative and Clinical Severity Towards Long-Term ECG Monitoring

Characterization of noise in long-term ECG monitoring with machine learning based on clinical criteria

Noise and artifacts affect strongly the quality of the electrocardiogram (ECG) in long-term ECG monitoring (LTM), making some of its parts impractical for diagnosis. The clinical severity of noise defines a qualitative quality score according to the manner clinicians make the interpretation of the ECG, in contrast to assess noise from a quantitative standpoint. So clinical noise refers to a scale of different levels of qualitative severity of noise which aims at elucidating which ECG fragments are valid to achieve diagnosis from a clinical point of view, unlike the traditional approach, which assesses noise in terms of quantitative severity. This work proposes the use of machine learning (ML) techniques to categorize different qualitative noise severity using a database annotated according to a clinical noise taxonomy as gold standard. A comparative study is carried out using five representative ML methods, namely, K neareast neighbors, decision trees, support vector machine, single-layer perceptron, and random forest. The models are fed by signal quality indexes characterizing the waveform in time and frequency domains, as well as from a statistical viewpoint, to distinguish between clinically valid ECG segments from invalid ones. A solid methodology to prevent overfitting to both the dataset and the patient is developed, taking into account balance of classes, patient separation, and patient rotation in the test set. All the proposed learning systems have demonstrated good classification performance, attaining a recall, precision, and F1 score up to 0.78, 0.80, and 0.77, respectively, in the test set by a single-layer perceptron approach. These systems provide a classification solution for assessing the clinical quality of the ECG taken from LTM recordings. Graphical Abstract Clinical Noise Severity Classification based on Machine Learning techniques towards Long-Term ECG Monitoring.

Manifold analysis of the P-wave changes induced by pulmonary vein isolation during cryoballoon procedure

BACKGROUND/AIM

In atrial fibrillation (AF) ablation procedures, it is desirable to know whether a proper disconnection of the pulmonary veins (PVs) was achieved. We hypothesize that information about their isolation could be provided by analyzing changes in P-wave after ablation. Thus, we present a method to detect PV disconnection using P-wave signal analysis.

METHODS

Conventional P-wave feature extraction was compared to an automatic feature extraction procedure based on creating low-dimensional latent spaces for cardiac signals with the Uniform Manifold Approximation and Projection (UMAP) method. A database of patients (19 controls and 16 AF individuals who underwent a PV ablation procedure) was collected. Standard 12-lead ECG was recorded, and P-waves were segmented and averaged to extract conventional features (duration, amplitude, and area) and their manifold representations provided by UMAP on a 3-dimensional latent space. A virtual patient was used to validate these results further and study the spatial distribution of the extracted characteristics over the whole torso surface.

RESULTS

Both methods showed differences between P-wave before and after ablation. Conventional methods were more prone to noise, P-wave delineation errors, and inter-patient variability. P-wave differences were observed in the standard leads recordings. However, higher differences appeared in the torso region over the precordial leads. Recordings near the left scapula also yielded noticeable differences.

CONCLUSIONS

P-wave analysis based on UMAP parameters detects PV disconnection after ablation in AF patients and is more robust than heuristic parameterization. Moreover, additional leads different from the standard 12-lead ECG should be used to detect PV isolation and possible future reconnections better.

ECG segmentation algorithm based on bidirectional hidden semi-Markov model

Accurate segmentation of electrocardiogram (ECG) waves is crucial for cardiovascular diseases (CVDs). In this study, a bidirectional hidden semi-Markov model (BI-HSMM) based on the probability distributions of ECG waveform duration was proposed for ECG wave segmentation. Four feature-vectors of ECG signals were extracted as the observation sequence of the hidden Markov model (HMM), and the statistical probability distribution of each waveform duration was counted. Logistic regression (LR) was used to train model parameters. The starting and ending positions of the QRS wave were first detected, and thereafter, bidirectional prediction was employed for the other waves. Forwardly, ST segment, T wave, and TP segment were predicted. Backwardly, P wave and PQ segments were detected. The Viterbi algorithm was improved by integrating the recursive formula of the forward prediction and backward backtracking algorithms. In the QT database, the proposed method demonstrated excellent performance (Acc = 97.98%, F1 score of P wave = 98.37%, F1 score of QRS wave = 97.60%, F1 score of T wave = 97.79%). For the wearable dynamic electrocardiography (DCG) signals collected by the Shandong Provincial Hospital (SPH), the detection accuracy was 99.71% and the F1 of each waveform was above 99%. The experimental results and real DCG signal validation confirmed that the proposed new BI-HSMM method exhibits significant ability to segment the resting and DCG signals; this is conducive to the detection and monitoring of CVDs.

Wearable Electrocardiogram Quality Assessment Using Wavelet Scattering and LSTM

As the fast development of wearable devices and Internet of things technologies, real-time monitoring of ECG signals is quite critical for cardiovascular diseases. However, dynamic ECG signals recorded in free-living conditions suffered from extremely serious noise pollution. Presently, most algorithms for ECG signal evaluation were designed to divide signals into acceptable and unacceptable. Such classifications were not enough for real-time cardiovascular disease monitoring. In the study, a wearable ECG quality database with 50,085 recordings was built, including A/B/C (or high quality/medium quality/low quality) three quality grades (A: high quality signals can be used for CVD detection; B: slight contaminated signals can be used for heart rate extracting; C: heavily polluted signals need to be abandoned). A new SQA classification method based on a three-layer wavelet scattering network and transfer learning LSTM was proposed in this study, which can extract more systematic and comprehensive characteristics by analyzing the signals thoroughly and deeply. Experimental results (mACC = 98.56%, mF₁ = 98.55%, Se_A = 97.90%, Se_B = 98.16%, Se_C = 99.60%, +P_A = 98.52%, +P_B = 97.60%, +P_C = 99.54%, F_1A = 98.20%, F_1B = 97.90%, F_1C = 99.60%) and real data validations proved that this proposed method showed the high accuracy, robustness, and computationally efficiency. It has the ability to evaluate the long-term dynamic ECG signal quality. It is advantageous to promoting cardiovascular disease monitoring by removing contaminating signals and selecting high-quality signal segments for further analysis.

Short-Term Beat-to-Beat QT Variability Appears Influenced More Strongly by Recording Quality Than by Beat-to-Beat RR Variability

Increases in beat-to-beat variability of electrocardiographic QT interval duration have repeatedly been associated with increased risk of cardiovascular events and complications. The measurements of QT variability are frequently normalized for the underlying RR interval variability. Such normalization supports the concept of the so-called immediate RR effect which relates each QT interval to the preceding RR interval. The validity of this concept was investigated in the present study together with the analysis of the influence of electrocardiographic morphological stability on QT variability measurements. The analyses involved QT and RR measurements in 6,114,562 individual beats of 642,708 separate 10-s ECG samples recorded in 523 healthy volunteers (259 females). Only beats with high morphology correlation (r > 0.99) with representative waveforms of the 10-s ECG samples were analyzed, assuring that only good quality recordings were included. In addition to these high correlations, SDs of the ECG signal difference between representative waveforms and individual beats expressed morphological instability and ECG noise. In the intra-subject analyses of both individual beats and of 10-s averages, QT interval variability was substantially more strongly related to the ECG noise than to the underlying RR variability. In approximately one-third of the analyzed ECG beats, the prolongation or shortening of the preceding RR interval was followed by the opposite change of the QT interval. In linear regression analyses, underlying RR variability within each 10-s ECG sample explained only 5.7 and 11.1% of QT interval variability in females and males, respectively. On the contrary, the underlying ECG noise contents of the 10-s samples explained 56.5 and 60.1% of the QT interval variability in females and males, respectively. The study concludes that the concept of stable and uniform immediate RR interval effect on the duration of subsequent QT interval duration is highly questionable. Even if only stable beat-to-beat measurements of QT interval are used, the QT interval variability is still substantially influenced by morphological variability and noise pollution of the source ECG recordings. Even when good quality recordings are used, noise contents of the electrocardiograms should be objectively examined in future studies of QT interval variability.

Application of Dense Neural Networks for Detection of Atrial Fibrillation and Ranking of Augmented ECG Feature Set

Considering the significant burden to patients and healthcare systems globally related to atrial fibrillation (AF) complications, the early AF diagnosis is of crucial importance. In the view of prominent perspectives for fast and accurate point-of-care arrhythmia detection, our study optimizes an artificial neural network (NN) classifier and ranks the importance of enhanced 137 diagnostic ECG features computed from time and frequency ECG signal representations of short single-lead strips available in 2017 Physionet/CinC Challenge database. Based on hyperparameters' grid search of densely connected NN layers, we derive the optimal topology with three layers and 128, 32, 4 neurons per layer (DenseNet-3@128-32-4), which presents maximal F1-scores for classification of Normal rhythms (0.883, 5076 strips), AF (0.825, 758 strips), Other rhythms (0.705, 2415 strips), Noise (0.618, 279 strips) and total F1 relevant to the CinC Challenge of 0.804, derived by five-fold cross-validation. DenseNet-3@128-32-4 performs equally well with 137 to 32 features and presents tolerable reduction by about 0.03 to 0.06 points for limited input sets, including 8 and 16 features, respectively. The feature reduction is linked to effective application of a comprehensive method for computation of the feature map importance based on the weights of the activated neurons through the total path from input to specific output in DenseNet. The detailed analysis of 20 top-ranked ECG features with greatest importance to the detection of each rhythm and overall of all rhythms reveals DenseNet decision-making process, noticeably corresponding to the cardiologists' diagnostic point of view.

Signal Quality Assessment of a Novel ECG Electrode for Motion Artifact Reduction

BACKGROUND

The presence of noise is problematic in the analysis and interpretation of the ECG, especially in ambulatory monitoring. Restricting the analysis to high-quality signal segments only comes with the risk of excluding significant arrhythmia episodes. Therefore, the development of novel electrode technology, robust to noise, continues to be warranted.

METHODS

The signal quality of a novel wet ECG electrode (Piotrode) is assessed and compared to a commercially available, commonly used electrode (Ambu). The assessment involves indices of QRS detection and atrial fibrillation detection performance, as well as signal quality indices (ensemble standard deviation and time-frequency repeatability), computed from ECGs recorded simultaneously from 20 healthy subjects performing everyday activities.

RESULTS

The QRS detection performance using the Piotrode was considerably better than when using the Ambu, especially for running but also for lighter activities. The two signal quality indices demonstrated similar trends: the gap in quality became increasingly larger as the subjects became increasingly more active.

CONCLUSIONS

The novel wet ECG electrode produces signals with less motion artifacts, thereby offering the potential to reduce the review burden, and accordingly the cost, associated with ambulatory monitoring.

Spatial-Temporal Signals and Clinical Indices in Electrocardiographic Imaging (I): Preprocessing and Bipolar Potentials

During the last years, Electrocardiographic Imaging (ECGI) has emerged as a powerful and promising clinical tool to support cardiologists. Starting from a plurality of potential measurements on the torso, ECGI yields a noninvasive estimation of their causing potentials on the epicardium. This unprecedented amount of measured cardiac signals needs to be conditioned and adapted to current knowledge and methods in cardiac electrophysiology in order to maximize its support to the clinical practice. In this setting, many cardiac indices are defined in terms of the so-called bipolar electrograms, which correspond with differential potentials between two spatially close potential measurements. Our aim was to contribute to the usefulness of ECGI recordings in the current knowledge and methods of cardiac electrophysiology. For this purpose, we first analyzed the basic stages of conventional cardiac signal processing and scrutinized the implications of the spatial-temporal nature of signals in ECGI scenarios. Specifically, the stages of baseline wander removal, low-pass filtering, and beat segmentation and synchronization were considered. We also aimed to establish a mathematical operator to provide suitable bipolar electrograms from the ECGI-estimated epicardium potentials. Results were obtained on data from an infarction patient and from a healthy subject. First, the low-frequency and high-frequency noises are shown to be non-independently distributed in the ECGI-estimated recordings due to their spatial dimension. Second, bipolar electrograms are better estimated when using the criterion of the maximum-amplitude difference between spatial neighbors, but also a temporal delay in discrete time of about 40 samples has to be included to obtain the usual morphology in clinical bipolar electrograms from catheters. We conclude that spatial-temporal digital signal processing and bipolar electrograms can pave the way towards the usefulness of ECGI recordings in the cardiological clinical practice. The companion paper is devoted to analyzing clinical indices obtained from ECGI epicardial electrograms measuring waveform variability and repolarization tissue properties.

Multiple Physiological Signals Fusion Techniques for Improving Heartbeat Detection: A Review

This paper presents a review of the techniques found in the literature that aim to achieve a robust heartbeat detection from fusing multi-modal physiological signals (e.g., electrocardiogram (ECG), blood pressure (BP), artificial blood pressure (ABP), stroke volume (SV), photoplethysmogram (PPG), electroencephalogram (EEG), electromyogram (EMG), and electrooculogram (EOG), among others). Techniques typically employ ECG, BP, and ABP, of which usage has been shown to obtain the best performance under challenging conditions. SV, PPG, EMG, EEG, and EOG signals can help increase performance when included within the fusion. Filtering, signal normalization, and resampling are common preprocessing steps. Delay correction between the heartbeats obtained over some of the physiological signals must also be considered, and signal-quality assessment to retain the best signal/s must be considered as well. Fusion is usually accomplished by exploiting regularities in the RR intervals; by selecting the most promising signal for the detection at every moment; by a voting process; or by performing simultaneous detection and fusion using Bayesian techniques, hidden Markov models, or neural networks. Based on the results of the review, guidelines to facilitate future comparison of the performance of the different proposals are given and promising future lines of research are pointed out.

Enabling Heart Self-Monitoring for All and for AAL-Portable Device within a Complete Telemedicine System

During the last decades there has been a rapidly growing elderly population and the number of patients with chronic heart-related diseases has exploded. Many of them (such as those with congestive heart failure or some types of arrhythmias) require close medical supervision, thus imposing a big burden on healthcare costs in most western economies. Specifically, continuous or frequent Arterial Blood Pressure (ABP) and electrocardiogram (ECG) monitoring are important tools in the follow-up of many of these patients. In this work, we present a novel remote non-ambulatory and clinically validated heart self-monitoring system, which allows ABP and ECG monitoring to effectively identify clinically relevant arrhythmias. The system integrates digital transmission of the ECG and tensiometer measurements, within a patient-comfortable support, easy to recharge and with a multi-function software, all of them aiming to adapt for elderly people. The main novelty is that both physiological variables (ABP and ECG) are simultaneously measured in an ambulatory environment, which to our best knowledge is not readily available in the clinical market. Different processing techniques were implemented to analyze the heart rhythm, including pause detection, rhythm alterations and atrial fibrillation, hence allowing early detection of these diseases. Our results achieved clinical quality both for in-lab hardware testing and for ambulatory scenario validations. The proposed active assisted living (AAL) Sensor-based system is an end-to-end multidisciplinary system, fully connected to a platform and tested by the clinical team from beginning to end.

Electrocardiographic Fragmented Activity (I): Physiological Meaning of Multivariate Signal Decompositions

Recent research has proven the existence of statistical relation among fragmented QRS and several highly prevalence diseases, such as cardiac sarcoidosis, acute coronary syndrome, arrythmogenic cardiomyopathies, Brugada syndrome, and hypertrophic cardiomyopathy. One out of five hundred people suffer from hypertrophic cardiomyopathies. The relation among the fragmentation and arrhythmias drives the objective of this work, which is to propose a valid method for QRS fragmentation detection. With that aim, we followed a two-stage approach. First, we identified the features that better characterize the fragmentation by analyzing the physiological interpretation of multivariate approaches, such as principal component analysis (PCA) and independent component analysis (ICA). Second, we created an invariant transformation method for the multilead electrocardiogram (ECG), by scrutinizing the statistical distributions of the PCA eigenvectors and of the ICA transformation arrays, in order to anchor the desired elements in the suitable leads in the feature space. A complete database was compounded incorporating real fragmented ECGs, surrogate registers by synthetically adding fragmented activity to real non-fragmented ECG registers, and standard clean ECGs. Results showed that the creation of beat templates together with the application of PCA over eight independent leads achieves 0.995 fragmentation enhancement ratio and 0.07 dispersion coefficient. In the case of ICA over twelve leads, the results were 0.995 fragmentation enhancement ratio and 0.70 dispersion coefficient. We conclude that the algorithm presented in this work constructs a new paradigm, by creating a systematic and powerful tool for clinical anamnesis and evaluation based on multilead ECG. This approach consistently consolidates the inconspicuous elements present in multiple leads onto designated variables in the output space, hence offering additional and valid visual and non-visual information to standard clinical review, and opening the door to a more accurate automatic detection and statistically valid systematic approach for a wide number of applications. In this direction and within the companion paper, further developments are presented applying this technique to fragmentation detection.

Electrocardiographic Fragmented Activity (II): A Machine Learning Approach to Detection

Hypertrophic cardiomyopathy, according to its prevalence, is a comparatively common disease related to the risk of suffering sudden cardiac death, heart failure and stroke. This illness is characterized by the excessive deposition of collagen among healthy myocardium cells. This situation, which is medically known as fibrosis, constitutes effective conduction obstacles in the myocardium electrical path, and when severe enough, it can be outlined as additional peaks or notches in the QRS, clinically entitled as fragmentation. Nowadays, the fragmentation detection is performed by visual inspection, but the fragmented QRS can be confused with the noise present in the electrocardiogram (ECG). On the other hand, fibrosis detection is performed by magnetic resonance imaging with late gadolinium enhancement, the main drawback of this technique being its cost in terms of time and money. In this work, we propose two automatic algorithms, one for fragmented QRS detection and another for fibrosis detection. For this purpose, we used four different databases, including the subrogated database described in the companion paper and incorporating three additional ones, one compounded by more accurate subrogated ECG signals and two compounded by real and affected subjects as labeled by expert clinicians. The first real-world database contains QRS fragmented records and the second one contains records with fibrosis and both were recorded in Hospital Clínico Universitario Virgen de la Arrixaca (Spain). To deeply analyze the scope of these datasets, we benchmarked several classifiers such as Neural Networks, Support Vector Machines (SVM), Decision Trees and Gaussian Naïve Bayes (NB). For the fragmentation dataset, the best results were 0.94 sensitivity, 0.88 specificity, 0.89 positive predictive value, 0.93 negative predictive value and 0.91 accuracy when using SVM with Gaussian kernel. For the fibrosis databases, more limited accuracy was reached, with 0.47 sensitivity, 0.91 specificity, 0.82 predictive positive value, 0.66 negative predictive value and 0.70 accuracy when using Gaussian NB. Nevertheless, this is the first time that fibrosis detection is attempted automatically from ECG postprocessing, paving the way towards improved algorithms and methods for it. Therefore, we can conclude that the proposed techniques could offer a valuable tool to clinicians for both fragmentation and fibrosis diagnoses support.

Deep Learning and Big Data in Healthcare: A Double Review for Critical Beginners

In the last few years, there has been a growing expectation created about the analysis of large amounts of data often available in organizations, which has been both scrutinized by the academic world and successfully exploited by industry. Nowadays, two of the most common terms heard in scientific circles are Big Data and Deep Learning. In this double review, we aim to shed some light on the current state of these different, yet somehow related branches of Data Science, in order to understand the current state and future evolution within the healthcare area. We start by giving a simple description of the technical elements of Big Data technologies, as well as an overview of the elements of Deep Learning techniques, according to their usual description in scientific literature. Then, we pay attention to the application fields that can be said to have delivered relevant real-world success stories, with emphasis on examples from large technology companies and financial institutions, among others. The academic effort that has been put into bringing these technologies to the healthcare sector are then summarized and analyzed from a twofold view as follows: first, the landscape of application examples is globally scrutinized according to the varying nature of medical data, including the data forms in electronic health recordings, medical time signals, and medical images; second, a specific application field is given special attention, in particular the electrocardiographic signal analysis, where a number of works have been published in the last two years. A set of toy application examples are provided with the publicly-available MIMIC dataset, aiming to help the beginners start with some principled, basic, and structured material and available code. Critical discussion is provided for current and forthcoming challenges on the use of both sets of techniques in our future healthcare.

A new approach to the intracardiac inverse problem using Laplacian distance kernel

Background

The inverse problem in electrophysiology consists of the accurate estimation of the intracardiac electrical sources from a reduced set of electrodes at short distances and from outside the heart. This estimation can provide an image with relevant knowledge on arrhythmia mechanisms for the clinical practice. Methods based on truncated singular value decomposition (TSVD) and regularized least squares require a matrix inversion, which limits their resolution due to the unavoidable low-pass filter effect of the Tikhonov regularization techniques.

Methods

We propose to use, for the first time, a Mercer’s kernel given by the Laplacian of the distance in the quasielectrostatic field equations, hence providing a Support Vector Regression (SVR) formulation by following the principles of the Dual Signal Model (DSM) principles for creating kernel algorithms.

Results

Simulations in one- and two-dimensional models show the performance of our Laplacian distance kernel technique versus several conventional methods. Firstly, the one-dimensional model is adjusted for yielding recorded electrograms, similar to the ones that are usually observed in electrophysiological studies, and suitable strategy is designed for the free-parameter search. Secondly, simulations both in one- and two-dimensional models show larger noise sensitivity in the estimated transfer matrix than in the observation measurements, and DSM−SVR is shown to be more robust to noisy transfer matrix than TSVD.

Conclusion

These results suggest that our proposed DSM−SVR with Laplacian distance kernel can be an efficient alternative to improve the resolution in current and emerging intracardiac imaging systems.

On the Beat Detection Performance in Long-Term ECG Monitoring Scenarios

Despite the wide literature on R-wave detection algorithms for ECG Holter recordings, the long-term monitoring applications are bringing new requirements, and it is not clear that the existing methods can be straightforwardly used in those scenarios. Our aim in this work was twofold: First, we scrutinized the scope and limitations of existing methods for Holter monitoring when moving to long-term monitoring; Second, we proposed and benchmarked a beat detection method with adequate accuracy and usefulness in long-term scenarios. A longitudinal study was made with the most widely used waveform analysis algorithms, which allowed us to tune the free parameters of the required blocks, and a transversal study analyzed how these parameters change when moving to different databases. With all the above, the extension to long-term monitoring in a database of 7-day Holter monitoring was proposed and analyzed, by using an optimized simultaneous-multilead processing. We considered both own and public databases. In this new scenario, the noise-avoid mechanisms are more important due to the amount of noise that exists in these recordings, moreover, the computational efficiency is a key parameter in order to export the algorithm to the clinical practice. The method based on a Polling function outperformed the others in terms of accuracy and computational efficiency, yielding 99.48% sensitivity, 99.54% specificity, 99.69% positive predictive value, 99.46% accuracy, and 0.85% error for MIT-BIH arrhythmia database. We conclude that the method can be used in long-term Holter monitoring systems.