Long-term ST database: a reference for the development and evaluation of automated ischaemia detectors and for the study of the dynamics of myocardial ischaemia

Automatic segmentation of atrial fibrillation and flutter in single-lead electrocardiograms by self-supervised learning and Transformer architecture

OBJECTIVES

Automatic detection of atrial fibrillation and flutter (AF/AFL) is a significant concern in preventing stroke and mitigating hemodynamic instability. Herein, we developed a Transformer-based deep learning model for AF/AFL segmentation in single-lead electrocardiograms (ECGs) by self-supervised learning with masked signal modeling (MSM).

MATERIALS AND METHODS

We retrieved data from 11 open-source databases on PhysioNet; 7 of these databases included labeled ECGs, while the other 4 were without labels. Each database contained ECG recordings with durations of ≥30 s. A total of 24 intradialytic ECGs with paroxysmal AF/AFL during 4 h of hemodialysis sessions at Seoul National University Hospital were used for external validation. The model was pretrained by predicting masked areas of ECG signals and fine-tuned by predicting AF/AFL areas. Cross-database validation was used for evaluation, and the intersection over union (IOU) was used as a main performance metric in external database validation.

RESULTS

In the 7 labeled databases, the areas marked as AF/AFL constituted 41.1% of the total ECG signals, ranging from 0.19% to 51.31%. In the evaluation per ECG segment, the model achieved IOU values of 0.9254 and 0.9477 for AF/AFL segmentation and other segmentation tasks, respectively. When applied to intradialytic ECGs with paroxysmal AF/AFL, the IOUs for the segmentation of AF/AFL and non-AF/AFL were 0.9896 and 0.9650, respectively. Model performance by different training procedure indicated that pretraining with MSM and the application of an appropriate masking ratio both contributed to the model performance. It also showed higher IOUs of AF/AFL labels than in previous studies when training and test databases were matched.

CONCLUSION

The present model with self-supervised learning by MSM performs robustly in segmenting AF/AFL.

Electrocardiogram-based deep learning algorithm for the screening of obstructive coronary artery disease

BACKGROUND

Information on electrocardiogram (ECG) has not been quantified in obstructive coronary artery disease (ObCAD), despite the deep learning (DL) algorithm being proposed as an effective diagnostic tool for acute myocardial infarction (AMI). Therefore, this study adopted a DL algorithm to suggest the screening of ObCAD from ECG.

METHODS

ECG voltage-time traces within a week from coronary angiography (CAG) were extracted for the patients who received CAG for suspected CAD in a single tertiary hospital from 2008 to 2020. After separating the AMI group, those were classified into ObCAD and non-ObCAD groups based on the CAG results. A DL-based model adopting ResNet was built to extract information from ECG data in the patients with ObCAD relative to those with non-ObCAD, and compared the performance with AMI. Moreover, subgroup analysis was conducted using ECG patterns of computer-assisted ECG interpretation.

RESULTS

The DL model demonstrated modest performance in suggesting the probability of ObCAD but excellent performance in detecting AMI. The AUC of the ObCAD model adopting 1D ResNet was 0.693 and 0.923 in detecting AMI. The accuracy, sensitivity, specificity, and F1 score of the DL model for screening ObCAD were 0.638, 0.639, 0.636, and 0.634, respectively, while the figures were up to 0.885, 0.769, 0.921, and 0.758 for detecting AMI, respectively. Subgroup analysis showed that the difference between normal and abnormal/borderline ECG groups was not notable.

CONCLUSIONS

ECG-based DL model showed fair performance for assessing ObCAD and it may serve as an adjunct to the pre-test probability in patients with suspected ObCAD during the initial evaluation. With further refinement and evaluation, ECG coupled with the DL algorithm may provide potential front-line screening support in the resource-intensive diagnostic pathways.

Standardized Database of 12-Lead Electrocardiograms with a Common Standard for the Promotion of Cardiovascular Research: KURIAS-ECG

OBJECTIVES

Electrocardiography (ECG)-based diagnosis by experts cannot maintain uniform quality because individual differences may occur. Previous public databases can be used for clinical studies, but there is no common standard that would allow databases to be combined. For this reason, it is difficult to conduct research that derives results by combining databases. Recent commercial ECG machines offer diagnoses similar to those of a physician. Therefore, the purpose of this study was to construct a standardized ECG database using computerized diagnoses.

METHODS

The constructed database was standardized using Systematized Nomenclature of Medicine Clinical Terms (SNOMED CT) and Observational Medical Outcomes Partnership-common data model (OMOP-CDM), and data were then categorized into 10 groups based on the Minnesota classification. In addition, to extract high-quality waveforms, poor-quality ECGs were removed, and database bias was minimized by extracting at least 2,000 cases for each group. To check database quality, the difference in baseline displacement according to whether poor ECGs were removed was analyzed, and the usefulness of the database was verified with seven classification models using waveforms.

RESULTS

The standardized KURIAS-ECG database consists of high-quality ECGs from 13,862 patients, with about 20,000 data points, making it possible to obtain more than 2,000 for each Minnesota classification. An artificial intelligence classification model using the data extracted through SNOMED-CT showed an average accuracy of 88.03%.

CONCLUSIONS

The KURIAS-ECG database contains standardized ECG data extracted from various machines. The proposed protocol should promote cardiovascular disease research using big data and artificial intelligence.

Hierarchical deep learning with Generative Adversarial Network for automatic cardiac diagnosis from ECG signals

Cardiac disease is the leading cause of death in the US. Accurate heart disease detection is critical to timely medical treatment to save patients' lives. Routine use of the electrocardiogram (ECG) is the most common method for physicians to assess the cardiac electrical activities and detect possible abnormal conditions. Fully utilizing the ECG data for reliable heart disease detection depends on developing effective analytical models. In this paper, we propose a two-level hierarchical deep learning framework with Generative Adversarial Network (GAN) for ECG signal analysis. The first-level model is composed of a Memory-Augmented Deep AutoEncoder with GAN (MadeGAN), which aims to differentiate abnormal signals from normal ECGs for anomaly detection. The second-level learning aims at robust multi-class classification for different arrhythmia identification, which is achieved by integrating the transfer learning technique to transfer knowledge from the first-level learning with the multi-branching architecture to handle the data-lacking and imbalanced data issues. We evaluate the performance of the proposed framework using real-world ECG data from the MIT-BIH arrhythmia database. Experimental results show that our proposed model outperforms existing methods that are commonly used in current practice.

Enhancing Electrocardiogram Classification with Multiple Datasets and Distant Transfer Learning

Electrocardiogram classification is crucial for various applications such as the medical diagnosis of cardiovascular diseases, the level of heart damage, and stress. One of the typical challenges of electrocardiogram classification problems is the small size of the datasets, which may lead to limitation in the performance of the classification models, particularly for models based on deep-learning algorithms. Transfer learning has demonstrated effectiveness in transferring knowledge from a source model with a similar domain and can enhance the performance of the target model. Nevertheless, the consideration of datasets with similar domains restricts the selection of source domains. In this paper, electrocardiogram classification was enhanced by distant transfer learning where a generative-adversarial-network-based auxiliary domain with a domain-feature-classifier negative-transfer-avoidance (GANAD-DFCNTA) algorithm was proposed to bridge the knowledge transfer from distant sources to target domains. To evaluate the performance of the proposed algorithm, eight benchmark datasets were chosen, with four from electrocardiogram datasets and four from the following distant domains: ImageNet, COCO, WordNet, and Sentiment140. The results showed an average accuracy improvement of 3.67 to 4.89%. The proposed algorithm was also compared with existing works using traditional transfer learning, revealing an average accuracy improvement of 0.303-5.19%. Ablation studies confirmed the effectiveness of the components of GANAD-DFCNTA.

Using Multi-Task Learning-Based Framework to Detect ST-Segment and J-Point Deviation From Holter

Artificial intelligence is increasingly being used on the clinical electrocardiogram workflows. Few electrocardiograms based on artificial intelligence algorithms have focused on detecting myocardial ischemia using long-term electrocardiogram data. A main reason for this is that interference signals generated from daily activities while wearing the Holter monitor lowered the ability of artificial intelligence to detect myocardial ischemia. In this study, an automatic system combining denoising and segmentation modules was developed to detect the deviation of the ST-segment and J point. We proposed a ECG Bidirectional Transformer network that applied in both denoising and segmentation tasks. The denoising model achieved RMSEde, SNRimp, and PRD values of 0.074, 10.006, and 16.327, respectively. The segmentation model achieved precision, sensitivity (recall), and F1-score of 96.00, 93.06, and 94.51%, respectively. The system’s ability to distinguish the depression and elevation of the ST-segment and J point was also verified by cardiologists as well. From our ECG dataset, 103 patients with ST-segment depression and 10 patients with ST-segment elevation were detected with positive predictive values of 80.6 and 60% respectively. Using Holter ECG and transformer-based deep neural networks, we can detect subtle ST-segment changes in noisy ECG signals. This system has the potential to improve the efficacy of daily medicine and to provide a broader population-level screening for asymptomatic myocardial ischemia.

Time Synchronization of Multimodal Physiological Signals through Alignment of Common Signal Types and Its Technical Considerations in Digital Health

Background: Despite advancements in digital health, it remains challenging to obtain precise time synchronization of multimodal physiological signals collected through different devices. Existing algorithms mainly rely on specific physiological features that restrict the use cases to certain signal types. The present study aims to complement previous algorithms and solve a niche time alignment problem when a common signal type is available across different devices. Methods: We proposed a simple time alignment approach based on the direct cross-correlation of temporal amplitudes, making it agnostic and thus generalizable to different signal types. The approach was tested on a public electrocardiographic (ECG) dataset to simulate the synchronization of signals collected from an ECG watch and an ECG patch. The algorithm was evaluated considering key practical factors, including sample durations, signal quality index (SQI), resilience to noise, and varying sampling rates. Results: The proposed approach requires a short sample duration (30 s) to operate, and demonstrates stable performance across varying sampling rates and resilience to common noise. The lowest synchronization delay achieved by the algorithm is 0.13 s with the integration of SQI thresholding. Conclusions: Our findings help improve the time alignment of multimodal signals in digital health and advance healthcare toward precise remote monitoring and disease prevention.

A Super Fast Algorithm for Estimating Sample Entropy

Sample entropy, an approximation of the Kolmogorov entropy, was proposed to characterize complexity of a time series, which is essentially defined as −log(B/A), where B denotes the number of matched template pairs with length m and A denotes the number of matched template pairs with m+1, for a predetermined positive integer m. It has been widely used to analyze physiological signals. As computing sample entropy is time consuming, the box-assisted, bucket-assisted, x-sort, assisted sliding box, and kd-tree-based algorithms were proposed to accelerate its computation. These algorithms require O(N2) or O(N2−1m+1) computational complexity, where N is the length of the time series analyzed. When N is big, the computational costs of these algorithms are large. We propose a super fast algorithm to estimate sample entropy based on Monte Carlo, with computational costs independent of N (the length of the time series) and the estimation converging to the exact sample entropy as the number of repeating experiments becomes large. The convergence rate of the algorithm is also established. Numerical experiments are performed for electrocardiogram time series, electroencephalogram time series, cardiac inter-beat time series, mechanical vibration signals (MVS), meteorological data (MD), and 1/f noise. Numerical results show that the proposed algorithm can gain 100–1000 times speedup compared to the kd-tree and assisted sliding box algorithms while providing satisfactory approximate accuracy.

Comparability of Heart Rate Turbulence Methodology: 15 Intervals Suffice to Calculate Turbulence Slope – A Methodological Analysis Using PhysioNet Data of 1074 Patients

Heart rate turbulence (HRT) is a characteristic heart rate pattern triggered by a ventricular premature contraction (VPC). It can be used to assess autonomic function and health risk for various conditions, e.g., coronary artery disease or cardiomyopathy. While comparability is essential for scientific analysis, especially for research focusing on clinical application, the methodology of HRT still varies widely in the literature. Particularly, the ECG measurement and parameter calculation of HRT differs, including the calculation of turbulence slope (TS). In this article, we focus on common variations in the number of intervals after the VPC that are used to calculate TS (#TSRR) posing two questions: 1) Does a change in #TSRR introduce noticeable changes in HRT parameter values and classification? and 2) Do larger values of turbulence timing (TT) enabled by a larger #TSRR still represent distinct HRT? We compiled a free-access data set of 1,080 annotated long-term ECGs provided by Physionet. HRT parameter values and risk classes were determined both with #TSRR 15 and 20. A standard local tachogram was created by averaging the tachograms of only the files with the best heart rate variability values. The shape of this standard VPC sequence was compared to all VPC sequences grouped by their TT value using dynamic time warping (DTW) in order to identify HRT shapes. When calculated with different #TSRR, our results show only a little difference between the number of files with enough valid VPC sequences to calculate HRT (<1%) and files with different risk classes (5 and 6% for HRT0-2 and HRTA-C, respectively). In the DTW analysis, the difference between averaged sequences with a specific TT and the standard sequence increased with increasing TT. Our analysis suggests that HRT occurs in the early intervals after the VPC and TS calculated from late intervals reflects common heart rate variability rather than a distinct response to the VPC. Even though the differences in classification are marginal, this can lead to problems in clinical application and scientific research. Therefore, we recommend uniformly using #TSRR 15 in HRT analysis.

Deep Learning for Detecting and Locating Myocardial Infarction by Electrocardiogram: A Literature Review

Myocardial infarction is a common cardiovascular disorder caused by prolonged ischemia, and early diagnosis of myocardial infarction (MI) is critical for lifesaving. ECG is a simple and non-invasive approach in MI detection, localization, diagnosis, and prognosis. Population-based screening with ECG can detect MI early and help prevent it but this method is too labor-intensive and time-consuming to carry out in practice unless artificial intelligence (AI) would be able to reduce the workload. Recent advances in using deep learning (DL) for ECG screening might rekindle this hope. This review aims to take stock of 59 major DL studies applied to the ECG for MI detection and localization published in recent 5 years, covering convolutional neural network (CNN), long short-term memory (LSTM), convolutional recurrent neural network (CRNN), gated recurrent unit (GRU), residual neural network (ResNet), and autoencoder (AE). In this period, CNN obtained the best popularity in both MI detection and localization, and the highest performance has been obtained from CNN and ResNet model. The reported maximum accuracies of the six different methods are all beyond 97%. Considering the usage of different datasets and ECG leads, the network that trained on 12 leads ECG data of PTB database has obtained higher accuracy than that on smaller number leads data of other datasets. In addition, some limitations and challenges of the DL techniques are also discussed in this review.

Dynamic detection and reversal of myocardial ischemia using an artificially intelligent bioelectronic medicine

Myocardial ischemia is spontaneous, frequently asymptomatic, and contributes to fatal cardiovascular consequences. Importantly, myocardial sensory networks cannot reliably detect and correct myocardial ischemia on their own. Here, we demonstrate an artificially intelligent and responsive bioelectronic medicine, where an artificial neural network (ANN) supplements myocardial sensory networks, enabling reliable detection and correction of myocardial ischemia. ANNs were first trained to decode spontaneous cardiovascular stress and myocardial ischemia with an overall accuracy of ~92%. ANN-controlled vagus nerve stimulation (VNS) significantly mitigated major physiological features of myocardial ischemia, including ST depression and arrhythmias. In contrast, open-loop VNS or ANN-controlled VNS following a caudal vagotomy essentially failed to reverse cardiovascular pathophysiology. Last, variants of ANNs were used to meet clinically relevant needs, including interpretable visualizations and unsupervised detection of emerging cardiovascular stress. Overall, these preclinical results suggest that ANNs can potentially supplement deficient myocardial sensory networks via an artificially intelligent bioelectronic medicine system.

Artificial Intelligence for Detection of Cardiovascular-Related Diseases from Wearable Devices: A Systematic Review and Meta-Analysis

PURPOSE

Several artificial intelligence (AI) models for the detection and prediction of cardiovascular-related diseases, including arrhythmias, diabetes, and sleep apnea, have been reported. This systematic review and meta-analysis aimed to identify AI models developed for or applicable to wearable and mobile devices for diverse cardiovascular-related diseases.

MATERIALS AND METHODS

The searched databases included Medline, Embase, and Cochrane Library. For AI models for atrial fibrillation (AF) detection, a meta-analysis of diagnostic accuracy was performed to summarize sensitivity and specificity.

RESULTS

A total of 102 studies were included in the qualitative review. There were AI models for the detection of arrythmia (n=62), followed by sleep apnea (n=11), peripheral vascular diseases (n=6), diabetes mellitus (n=5), hyper/hypotension (n=5), valvular heart disease (n=4), heart failure (n=3), myocardial infarction and cardiac arrest (n=2), and others (n=4). For quantitative analysis of 26 studies reporting AI models for AF detection, meta-analyzed sensitivity was 94.80% and specificity was 96.96%. Deep neural networks showed superior performance [meta-analyzed area under receiver operating characteristics curve (AUROC) of 0.981] compared to conventional machine learning algorithms (meta-analyzed AUROC of 0.961). However, AI models tested with proprietary dataset (meta-analyzed AUROC of 0.972) or data acquired from wearable devices (meta-analyzed AUROC of 0.977) showed inferior performance than those with public dataset (meta-analyzed AUROC of 0.986) or data from in-hospital devices (meta-analyzed AUROC of 0.983).

CONCLUSION

This review found that AI models for diverse cardiovascular-related diseases are being developed, and that they are gradually developing into a form that is suitable for wearable and mobile devices.

Tensor-Based ECG Anomaly Detection toward Cardiac Monitoring in the Internet of Health Things

Advanced heart monitors, especially those enabled by the Internet of Health Things (IoHT), provide a great opportunity for continuous collection of the electrocardiogram (ECG), which contains rich information about underlying cardiac conditions. Realizing the full potential of IoHT-enabled cardiac monitoring hinges, to a great extent, on the detection of disease-induced anomalies from collected ECGs. However, challenges exist in the current literature for IoHT-based cardiac monitoring: (1) Most existing methods are based on supervised learning, which requires both normal and abnormal samples for training. This is impractical as it is generally unknown when and what kind of anomalies will occur during cardiac monitoring. (2) Furthermore, it is difficult to leverage advanced machine learning approaches for information processing of 1D ECG signals, as most of them are designed for 2D images and higher-dimensional data. To address these challenges, a new sensor-based unsupervised framework is developed for IoHT-based cardiac monitoring. First, a high-dimensional tensor is generated from the multi-channel ECG signals through the Gramian Angular Difference Field (GADF). Then, multi-linear principal component analysis (MPCA) is employed to unfold the ECG tensor and delineate the disease-altered patterns. Obtained principal components are used as features for anomaly detection using machine learning models (e.g., deep support vector data description (deep SVDD)) as well as statistical control charts (e.g., Hotelling T2 chart). The developed framework is evaluated and validated using real-world ECG datasets. Comparing to the state-of-the-art approaches, the developed framework with deep SVDD achieves superior performances in detecting abnormal ECG patterns induced by various types of cardiac disease, e.g., an F-score of 0.9771 is achieved for detecting atrial fibrillation, 0.9986 for detecting right bundle branch block, and 0.9550 for detecting ST-depression. Additionally, the developed framework with the T² control chart facilitates personalized cycle-to-cycle monitoring with timely detected abnormal ECG patterns. The developed framework has a great potential to be implemented in IoHT-enabled cardiac monitoring and smart management of cardiac health.

Edge-Computing Architectures for Internet of Things Applications: A Survey

The rapid growth of the Internet of Things (IoT) applications and their interference with our daily life tasks have led to a large number of IoT devices and enormous sizes of IoT-generated data. The resources of IoT devices are limited; therefore, the processing and storing IoT data in these devices are inefficient. Traditional cloud-computing resources are used to partially handle some of the IoT resource-limitation issues; however, using the resources in cloud centers leads to other issues, such as latency in time-critical IoT applications. Therefore, edge-cloud-computing technology has recently evolved. This technology allows for data processing and storage at the edge of the network. This paper studies, in-depth, edge-computing architectures for IoT (ECAs-IoT), and then classifies them according to different factors such as data placement, orchestration services, security, and big data. Besides, the paper studies each architecture in depth and compares them according to various features. Additionally, ECAs-IoT is mapped according to two existing IoT layered models, which helps in identifying the capabilities, features, and gaps of every architecture. Moreover, the paper presents the most important limitations of existing ECAs-IoT and recommends solutions to them. Furthermore, this survey details the IoT applications in the edge-computing domain. Lastly, the paper recommends four different scenarios for using ECAs-IoT by IoT applications.

Classification of ischemic and non-ischemic cardiac events in Holter recordings based on the continuous wavelet transform

Holter recordings are widely used to detect cardiac events that occur transiently, such as ischemic events. Much effort has been made to detect early ischemia, thus preventing myocardial infarction. However, after detection, classification of ischemia has still not been fully solved. The main difficulty relies on the false positives produced because of non-ischemic events, such as changes in the heart rate, the intraventricular conduction or the cardiac electrical axis. In this work, the classification of ischemic and non-ischemic events from the long-term ST database has been improved, using novel spectral parameters based on the continuous wavelet transform (CWT) together with temporal parameters (such as ST level and slope, T wave width and peak, R wave peak, QRS complex width). This was achieved by using a nearest neighbour classifier of six neighbours. Results indicated a sensitivity and specificity of 84.1% and 92.9% between ischemic and non-ischemic events, respectively, resulting a 10% increase of the sensitivity found in the literature. Extracted features based on the CWT applied on the ECG in the frequency band 0.5-4 Hz provided a substantial improvement in classifying ischemic and non-ischemic events, when comparing with the same classifier using only temporal parameters. Graphical Abstract In this work it is improved the classification of ischemic and non-ischemic events. The main difficulty of ischemic detectors relies on the false positives produced because of non-ischemic events. After a preprocessing stage, temporal and spectral parameters are extracted from events of the Long Term ST Database. The novel parameters proposed in this work are extracted from the Continuous Wavelet Transform. A nearest Neighbor Classifier is used, obtaining a sensitivity and specificity of 84.1% and 92.9%, respectively.

High arrhythmic risk in antero-septal acute myocardial ischemia is explained by increased transmural reentry occurrence

Acute myocardial ischemia is a precursor of sudden arrhythmic death. Variability in its manifestation hampers understanding of arrhythmia mechanisms and challenges risk stratification. Our aim is to unravel the mechanisms underlying how size, transmural extent and location of ischemia determine arrhythmia vulnerability and ECG alterations. High performance computing simulations using a human torso/biventricular biophysically-detailed model were conducted to quantify the impact of varying ischemic region properties, including location (LAD/LCX occlusion), transmural/subendocardial ischemia, size, and normal/slow myocardial propagation. ECG biomarkers and vulnerability window for reentry were computed in over 400 simulations for 18 cases evaluated. Two distinct mechanisms explained larger vulnerability to reentry in transmural versus subendocardial ischemia. Macro-reentry around the ischemic region was the primary mechanism increasing arrhythmic risk in transmural versus subendocardial ischemia, for both LAD and LCX occlusion. Transmural micro-reentry at the ischemic border zone explained arrhythmic vulnerability in subendocardial ischemia, especially in LAD occlusion, as reentries were favoured by the ischemic region intersecting the septo-apical region. ST elevation reflected ischemic extent in transmural ischemia for LCX and LAD occlusion but not in subendocardial ischemia (associated with mild ST depression). The technology and results presented can inform safety and efficacy evaluation of anti-arrhythmic therapy in acute myocardial ischemia.

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.

Low Resource Complexity R-peak Detection Based on Triangle Template Matching and Moving Average Filter

A novel R-peak detection algorithm suitable for wearable electrocardiogram (ECG) devices is proposed with four objectives: robustness to noise, low latency processing, low resource complexity, and automatic tuning of parameters. The approach is a two-pronged algorithm comprising (1) triangle template matching to accentuate the slope information of the R-peaks and (2) a single moving average filter to define a dynamic threshold for peak detection. The proposed algorithm was validated on eight ECG public databases. The obtained results not only presented good accuracy, but also low resource complexity, all of which show great potential for detection R-peaks in ECG signals collected from wearable devices.

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.

Deep Learning in Cardiology

The medical field is creating large amount of data that physicians are unable to decipher and use efficiently. Moreover, rule-based expert systems are inefficient in solving complicated medical tasks or for creating insights using big data. Deep learning has emerged as a more accurate and effective technology in a wide range of medical problems such as diagnosis, prediction, and intervention. Deep learning is a representation learning method that consists of layers that transform data nonlinearly, thus, revealing hierarchical relationships and structures. In this review, we survey deep learning application papers that use structured data, and signal and imaging modalities from cardiology. We discuss the advantages and limitations of applying deep learning in cardiology that also apply in medicine in general, while proposing certain directions as the most viable for clinical use.

Analyzing single-lead short ECG recordings using dense convolutional neural networks and feature-based post-processing to detect atrial fibrillation

OBJECTIVE

The prevalence of atrial fibrillation (AF) in the general population is 0.5%-1%. As AF is the most common sustained cardiac arrhythmia that is associated with an increased morbidity and mortality, its timely diagnosis is clinically desirable. The main aim of this study as our contribution to the PhysioNet/CinC Challenge 2017 was to develop an automatic algorithm for classification of normal sinus rhythm (NSR), AF, other rhythm (O), and noise using a short single-channel ECG. Furthermore, the impact of changing labels/annotations on performance of the proposed algorithm was studied in this article.

APPROACH

The challenge training dataset (8528 ECG recordings) and a complementary dataset (6312 ECG recordings) from other sources were used for algorithm development. Version 3 (v3), which is an updated version of the annotations at the official phase of the challenge (v2), was used in this study. In the proposed algorithm, densely connected convolutional networks were combined with feature-based post-processing after initial signal quality analysis for the classification of ECG recordings.

MAIN RESULTS

The F1 scores for classification of NSR, AF, and O were 0.91, 0.83, and 0.72, respectively, which led to a F1 of 0.82. There was a small or no performance difference between the top entries in the official phase of the challenge and our proposed method. An increase of 2.5% in F1 score was observed when the same annotations for training and test was used (using v3 annotations) compared to using different annotations (v2 annotations for training and v3 annotations for the test).

SIGNIFICANCE

Our promising results suggest that the availability of more data with improved labeling along with improvement in signal quality analysis make our algorithm suitable for practical clinical applications.

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.

Wearable brain imaging with multimodal physiological monitoring

The brain is a central component of cognitive and physical human performance. Measures, including functional brain activation, cerebral perfusion, cerebral oxygenation, evoked electrical responses, and resting hemodynamic and electrical activity are all related to, or can predict, health status or performance decrements. However, measuring brain physiology typically requires large, stationary machines that are not suitable for mobile or self-monitoring. Moreover, when individuals are ambulatory, systemic physiological fluctuations—e.g., in heart rate, blood pressure, skin perfusion, and more—can interfere with noninvasive brain measurements. In efforts to address the physiological monitoring and performance assessment needs for astronauts during spaceflight, we have developed easy-to-use, wearable prototypes, such as NINscan, for near-infrared scanning, which can collect synchronized multimodal physiology data, including hemodynamic deep-tissue imaging (including brain and muscles), electroencephalography, electrocardiography, electromyography, electrooculography, accelerometry, gyroscopy, pressure, respiration, and temperature measurements. Given their self-contained and portable nature, these devices can be deployed in a much broader range of settings—including austere environments—thereby, enabling a wider range of novel medical and research physiology applications. We review these, including high-altitude assessments, self-deployable multimodal e.g., (polysomnographic) recordings in remote or low-resource environments, fluid shifts in variable-gravity, or spaceflight analog environments, intracranial brain motion during high-impact sports, and long-duration monitoring for clinical symptom-capture in various clinical conditions. In addition to further enhancing sensitivity and miniaturization, advanced computational algorithms could help support real-time feedback and alerts regarding performance and health.

Unsupervised heart-rate estimation in wearables with Liquid states and a probabilistic readout

Heart-rate estimation is a fundamental feature of modern wearable devices. In this paper we propose a machine learning technique to estimate heart-rate from electrocardiogram (ECG) data collected using wearable devices. The novelty of our approach lies in (1) encoding spatio-temporal properties of ECG signals directly into spike train and using this to excite recurrently connected spiking neurons in a Liquid State Machine computation model; (2) a novel learning algorithm; and (3) an intelligently designed unsupervised readout based on Fuzzy c-Means clustering of spike responses from a subset of neurons (Liquid states), selected using particle swarm optimization. Our approach differs from existing works by learning directly from ECG signals (allowing personalization), without requiring costly data annotations. Additionally, our approach can be easily implemented on state-of-the-art spiking-based neuromorphic systems, offering high accuracy, yet significantly low energy footprint, leading to an extended battery-life of wearable devices. We validated our approach with CARLsim, a GPU accelerated spiking neural network simulator modeling Izhikevich spiking neurons with Spike Timing Dependent Plasticity (STDP) and homeostatic scaling. A range of subjects is considered from in-house clinical trials and public ECG databases. Results show high accuracy and low energy footprint in heart-rate estimation across subjects with and without cardiac irregularities, signifying the strong potential of this approach to be integrated in future wearable devices.

From Pacemaker to Wearable: Techniques for ECG Detection Systems

With the alarming rise in the deaths due to cardiovascular diseases (CVD), present medical research scenario places notable importance on techniques and methods to detect CVDs. As adduced by world health organization, technological proceeds in the field of cardiac function assessment have become the nucleus and heart of all leading research studies in CVDs in which electrocardiogram (ECG) analysis is the most functional and convenient tool used to test the range of heart-related irregularities. Most of the approaches present in the literature of ECG signal analysis consider noise removal, rhythm-based analysis, and heartbeat detection to improve the performance of a cardiac pacemaker. Advancements achieved in the field of ECG segments detection and beat classification have a limited evaluation and still require clinical approvals. In this paper, approaches on techniques to implement on-chip ECG detector for a cardiac pacemaker system are discussed. Moreover, different challenges regarding the ECG signal morphology analysis deriving from medical literature is extensively reviewed. It is found that robustness to noise, wavelet parameter choice, numerical efficiency, and detection performance are essential performance indicators required by a state-of-the-art ECG detector. Furthermore, many algorithms described in the existing literature are not verified using ECG data from the standard databases. Some ECG detection algorithms show very high detection performance with the total number of detected QRS complexes. However, the high detection performance of the algorithm is verified using only a few datasets. Finally, gaps in current advancements and testing are identified, and the primary challenge remains to be implementing bullseye test for morphology analysis evaluation.

A Review of Automated Methods for Detection of Myocardial Ischemia and Infarction Using Electrocardiogram and Electronic Health Records

There is a growing body of research focusing on automatic detection of ischemia and myocardial infarction (MI) using computer algorithms. In clinical settings, ischemia and MI are diagnosed using electrocardiogram (ECG) recordings as well as medical context including patient symptoms, medical history, and risk factors-information that is often stored in the electronic health records. The ECG signal is inspected to identify changes in the morphology such as ST-segment deviation and T-wave changes. Some of the proposed methods compute similar features automatically while others use nonconventional features such as wavelet coefficients. This review provides an overview of the methods that have been proposed in this area, focusing on their historical evolution, the publicly available datasets that they have used to evaluate their performance, and the details of their algorithms for ECG and EHR analysis. The validation strategies that have been used to evaluate the performance of the proposed methods are also presented. Finally, the paper provides recommendations for future research to address the shortcomings of the currently existing methods and practical considerations to make the proposed technical solutions applicable in clinical practice.

CSE database: extended annotations and new recommendations for ECG software testing

Nowadays, cardiovascular diseases represent the most common cause of death in western countries. Among various examination techniques, electrocardiography (ECG) is still a highly valuable tool used for the diagnosis of many cardiovascular disorders. In order to diagnose a person based on ECG, cardiologists can use automatic diagnostic algorithms. Research in this area is still necessary. In order to compare various algorithms correctly, it is necessary to test them on standard annotated databases, such as the Common Standards for Quantitative Electrocardiography (CSE) database. According to Scopus, the CSE database is the second most cited standard database. There were two main objectives in this work. First, new diagnoses were added to the CSE database, which extended its original annotations. Second, new recommendations for diagnostic software quality estimation were established. The ECG recordings were diagnosed by five new cardiologists independently, and in total, 59 different diagnoses were found. Such a large number of diagnoses is unique, even in terms of standard databases. Based on the cardiologists' diagnoses, a four-round consensus (4R consensus) was established. Such a 4R consensus means a correct final diagnosis, which should ideally be the output of any tested classification software. The accuracy of the cardiologists' diagnoses compared with the 4R consensus was the basis for the establishment of accuracy recommendations. The accuracy was determined in terms of sensitivity = 79.20-86.81%, positive predictive value = 79.10-87.11%, and the Jaccard coefficient = 72.21-81.14%, respectively. Within these ranges, the accuracy of the software is comparable with the accuracy of cardiologists. The accuracy quantification of the correct classification is unique. Diagnostic software developers can objectively evaluate the success of their algorithm and promote its further development. The annotations and recommendations proposed in this work will allow for faster development and testing of classification software. As a result, this might facilitate cardiologists' work and lead to faster diagnoses and earlier treatment.

Signal Quality Analysis of Ambulatory Electrocardiograms to Gate False Myocardial Ischemia Alarms

OBJECTIVE

The objective of this study is to propose and validate an alarm gating system for a myocardial ischemia monitoring system that uses ambulatory electrocardiogram. The PeriOperative ISchemic Evaluation study recommended the selective administration of β blockers to patients at risk of cardiac events following noncardiac surgery. Patients at risk are identified by monitoring ST segment deviations in the electrocardiogram (ECG); however, patients are encouraged to ambulate to improve recovery, which deteriorates the signal quality of the ECG leading to false alarms.

METHODS

The proposed alarm gating system computes a signal quality index (SQI) to quantify the ECG signal quality and rejects alarms with a low SQI. The system was validated by artificially contaminating ECG records with motion artifact records obtained from the long-term ST database and MIT-BIH noise stress test database, respectively.

RESULTS

Without alarm gating, the myocardial ischemia monitoring system attained a Precision of 0.31 and a Recall of 0.78. The alarm gating improved the Precision to 0.58 with a reduction of Recall to 0.77.

CONCLUSION

The proposed system successfully gated false alarms with future work exploring the misidentification of fiducial points by myocardial ischemia monitoring systems.

SIGNIFICANCE

The reduction of false alarms due to the proposed system will decrease the incidence of the alarm fatigue condition typically found in clinicians. Alarm fatigue condition was rated as the top patient safety hazard from 2012 to 2015 by the Emergency Care Research Institute.

T Wave Amplitude Correction of QT Interval Variability for Improved Repolarization Lability Measurement

Objectives: The inverse relationship between QT interval variability (QTV) and T wave amplitude potentially confounds QT variability assessment. We quantified the influence of the T wave amplitude on QTV in a comprehensive dataset and devised a correction formula.

Methods: Three ECG datasets of healthy subjects were analyzed to model the relationship between T wave amplitude and QTV. To derive a generally valid correction formula, linear regression analysis was used. The proposed correction formula was applied to patients enrolled in the Evaluation of Defibrillator in Non-Ischemic Cardiomyopathy Treatment Evaluation trial (DEFINITE) to assess the prognostic significance of QTV for all-cause mortality in patients with non-ischemic dilated cardiomyopathy.

Results: A strong inverse relationship between T wave amplitude and QTV was demonstrated, both in healthy subjects (R² = 0.68, p < 0.001) and DEFINITE patients (R² = 0.20, p < 0.001). Applying the T wave amplitude correction to QTV achieved 2.5-times better group discrimination between patients enrolled in the DEFINITE study and healthy subjects. Kaplan-Meier estimator analysis showed that T wave amplitude corrected QTVi is inversely related to survival (p < 0.01) and a significant predictor of all-cause mortality.

Conclusion: We have proposed a simple correction formula for improved QTV assessment. Using this correction, predictive value of QTV for all-cause mortality in patients with non-ischemic cardiomyopathy has been demonstrated.

Electrocardiogram ST-Segment Morphology Delineation Method Using Orthogonal Transformations

Differentiation between ischaemic and non-ischaemic transient ST segment events of long term ambulatory electrocardiograms is a persisting weakness in present ischaemia detection systems. Traditional ST segment level measuring is not a sufficiently precise technique due to the single point of measurement and severe noise which is often present. We developed a robust noise resistant orthogonal-transformation based delineation method, which allows tracing the shape of transient ST segment morphology changes from the entire ST segment in terms of diagnostic and morphologic feature-vector time series, and also allows further analysis. For these purposes, we developed a new Legendre Polynomials based Transformation (LPT) of ST segment. Its basis functions have similar shapes to typical transient changes of ST segment morphology categories during myocardial ischaemia (level, slope and scooping), thus providing direct insight into the types of time domain morphology changes through the LPT feature-vector space. We also generated new Karhunen and Lo ève Transformation (KLT) ST segment basis functions using a robust covariance matrix constructed from the ST segment pattern vectors derived from the Long Term ST Database (LTST DB). As for the delineation of significant transient ischaemic and non-ischaemic ST segment episodes, we present a study on the representation of transient ST segment morphology categories, and an evaluation study on the classification power of the KLT- and LPT-based feature vectors to classify between ischaemic and non-ischaemic ST segment episodes of the LTST DB. Classification accuracy using the KLT and LPT feature vectors was 90% and 82%, respectively, when using the k-Nearest Neighbors (k = 3) classifier and 10-fold cross-validation. New sets of feature-vector time series for both transformations were derived for the records of the LTST DB which is freely available on the PhysioNet website and were contributed to the LTST DB. The KLT and LPT present new possibilities for human-expert diagnostics, and for automated ischaemia detection.

Individual identification via electrocardiogram analysis

BACKGROUND

During last decade the use of ECG recordings in biometric recognition studies has increased. ECG characteristics made it suitable for subject identification: it is unique, present in all living individuals, and hard to forge. However, in spite of the great number of approaches found in literature, no agreement exists on the most appropriate methodology. This study aimed at providing a survey of the techniques used so far in ECG-based human identification. Specifically, a pattern recognition perspective is here proposed providing a unifying framework to appreciate previous studies and, hopefully, guide future research.

METHODS

We searched for papers on the subject from the earliest available date using relevant electronic databases (Medline, IEEEXplore, Scopus, and Web of Knowledge). The following terms were used in different combinations: electrocardiogram, ECG, human identification, biometric, authentication and individual variability. The electronic sources were last searched on 1st March 2015. In our selection we included published research on peer-reviewed journals, books chapters and conferences proceedings. The search was performed for English language documents.

RESULTS

100 pertinent papers were found. Number of subjects involved in the journal studies ranges from 10 to 502, age from 16 to 86, male and female subjects are generally present. Number of analysed leads varies as well as the recording conditions. Identification performance differs widely as well as verification rate. Many studies refer to publicly available databases (Physionet ECG databases repository) while others rely on proprietary recordings making difficult them to compare. As a measure of overall accuracy we computed a weighted average of the identification rate and equal error rate in authentication scenarios. Identification rate resulted equal to 94.95 % while the equal error rate equal to 0.92 %.

CONCLUSIONS

Biometric recognition is a mature field of research. Nevertheless, the use of physiological signals features, such as the ECG traits, needs further improvements. ECG features have the potential to be used in daily activities such as access control and patient handling as well as in wearable electronics applications. However, some barriers still limit its growth. Further analysis should be addressed on the use of single lead recordings and the study of features which are not dependent on the recording sites (e.g. fingers, hand palms). Moreover, it is expected that new techniques will be developed using fiducials and non-fiducial based features in order to catch the best of both approaches. ECG recognition in pathological subjects is also worth of additional investigations.

Independent component analysis and decision trees for ECG holter recording de-noising

We have developed a method focusing on ECG signal de-noising using Independent component analysis (ICA). This approach combines JADE source separation and binary decision tree for identification and subsequent ECG noise removal. In order to to test the efficiency of this method comparison to standard filtering a wavelet- based de-noising method was used. Freely data available at Physionet medical data storage were evaluated. Evaluation criteria was root mean square error (RMSE) between original ECG and filtered data contaminated with artificial noise. Proposed algorithm achieved comparable result in terms of standard noises (power line interference, base line wander, EMG), but noticeably significantly better results were achieved when uncommon noise (electrode cable movement artefact) were compared.

CCDD: AN ENHANCED STANDARD ECG DATABASE WITH ITS MANAGEMENT AND ANNOTATION TOOLS

Standard Electrocardiogram (ECG) database is created for validating and comparing different algorithms on feature detection and disease classification. At present, there are four frequently used standard databases: MIT-BIH arrhythmia database, QT database, CSE multi-lead database and AHA database. With the development in equipment and diagnosis approach, severe deficiencies are discovered and a new modern ECG database is needed for further research. So Chinese Cardiovascular Disease Database (CCDD or CCD database), which contains 12-Lead ECG data, detailed annotation features and beat diagnosis result is proposed. It is advanced for not only improving the raw ECG data's technical parameters, but also introducing valuable morphology features which are utilized by experienced cardiologists effectively. CCDD is employed by our group as well as aiming for supporting other research groups that work in automated ECG analysis.

Automatic classification of long-term ambulatory ECG records according to type of ischemic heart disease

BACKGROUND

Elevated transient ischemic ST segment episodes in the ambulatory electrocardiographic (AECG) records appear generally in patients with transmural ischemia (e. g. Prinzmetal's angina) while depressed ischemic episodes appear in patients with subendocardial ischemia (e. g. unstable or stable angina). Huge amount of AECG data necessitates automatic methods for analysis. We present an algorithm which determines type of transient ischemic episodes in the leads of records (elevations/depressions) and classifies AECG records according to type of ischemic heart disease (Prinzmetal's angina; coronary artery diseases excluding patients with Prinzmetal's angina; other heart diseases).

METHODS

The algorithm was developed using 24-hour AECG records of the Long Term ST Database (LTST DB). The algorithm robustly generates ST segment level function in each AECG lead of the records, and tracks time varying non-ischemic ST segment changes such as slow drifts and axis shifts to construct the ST segment reference function. The ST segment reference function is then subtracted from the ST segment level function to obtain the ST segment deviation function. Using the third statistical moment of the histogram of the ST segment deviation function, the algorithm determines deflections of leads according to type of ischemic episodes present (elevations, depressions), and then classifies records according to type of ischemic heart disease.

RESULTS

Using 74 records of the LTST DB (containing elevated or depressed ischemic episodes, mixed ischemic episodes, or no episodes), the algorithm correctly determined deflections of the majority of the leads of the records and correctly classified majority of the records with Prinzmetal's angina into the Prinzmetal's angina category (7 out of 8); majority of the records with other coronary artery diseases into the coronary artery diseases excluding patients with Prinzmetal's angina category (47 out of 55); and correctly classified one out of 11 records with other heart diseases into the other heart diseases category.

CONCLUSIONS

The developed algorithm is suitable for processing long AECG data, efficient, and correctly classified the majority of records of the LTST DB according to type of transient ischemic heart disease.

Myocardial ischemia analysis based on electrocardiogram QRS complex

Electrocardiogram (ECG) is an economic, convenient, and non-invasive detecting tool in myocardial ischemia (MI), and its clinical appearance is mainly exhibited by the changes in ST-T complex. Recently, QRS complex characters were proposed to analyze MI by more and more researchers. In this paper, various QRS complex characters were extracted in ECG signals, and their relationship was analyzed systematically. As a result, these characters were divided into two groups, and there existed good relationship among them for each group, while the poor relationship between the groups. Then these QRS complex characters were applied for statistical analysis on MI, and five characters had significant differences after ECG recording verification, which were: QRS upward and downward slopes, transient heart rate, angle R and angle Q. On the other hand, these QRS complex characters were analyzed in frequency domain. Experimental results showed that the frequency features of RR interval series (Heart Rate Variability, HRV), and QRS barycenter sequence had significant differences between MI states and normal states. Moreover, QRS barycenter sequence performed better.

Clinically accurate fetal ECG parameters acquired from maternal abdominal sensors

OBJECTIVE

We sought to evaluate the accuracy of a novel system for measuring fetal heart rate (FHR) and ST-segment changes using noninvasive electrodes on the maternal abdomen.

STUDY DESIGN

Fetal electrocardiograms were recorded using abdominal sensors from 32 term laboring women who had a fetal scalp electrode (FSE) placed for a clinical indication.

RESULTS

Good-quality data for FHR estimation were available in 91.2% of the FSE segments and 89.9% of the abdominal electrode segments. The root mean square error between the FHR data calculated by both methods over all processed segments was 0.36 beats per minute. ST deviation from the isoelectric point ranged from 0-14.2% of R-wave amplitude. The root mean square error between the ST change calculated by both methods averaged over all processed segments was 3.2%.

CONCLUSION

FHR and ST change acquired from the maternal abdomen is highly accurate and, on average, is clinically indistinguishable from FHR and ST change calculated using FSE data.

Research on electrocardiogram baseline wandering correction based on wavelet transform, QRS barycenter fitting, and regional method

Baseline wandering in electrocardiogram (ECG) is one of the biggest interferences in visualization and computerized detection of waveforms (especially ST-segment) based on threshold decision. A new method based on wavelet transform, QRS barycenter fitting and regional method was proposed in this paper. Firstly, wavelet transform as a coarse correction was used to remove the baseline wandering, whose frequency bands were non-overlapping with that of ST-segment. Secondly, QRS barycenter fitting was applied as a detailed correction. The third, the regional method was used to transfer baseline to zero. Finally, the method in this paper was proved to perform better than filtering and function fitting methods in baseline wandering correction after the long-term ST database (LTST) verification. In addition, the proposed method is simple and easy to carry out, and in current use.