Electrocardiography Recording, Feature Extraction and Classification for Emotion Recognition

Deep learning using electrocardiographs

Machine learning (ML) methods for the analysis of electrocardiography (ECG) data are gaining importance, substantially supported by the release of large public datasets. However, these current datasets miss important derived descriptors such as ECG features that have been devised in the past hundred years and still form the basis of most automatic ECG analysis algorithms and are critical for cardiologists’ decision processes. ECG features are available from sophisticated commercial software but are not accessible to the general public. To alleviate this issue, we add ECG features from two leading commercial algorithms and an open-source implementation supplemented by a set of automatic diagnostic statements from a commercial ECG analysis software in preprocessed format. This allows the comparison of ML models trained on clinically versus automatically generated label sets. We provide an extensive technical validation of features and diagnostic statements for ML applications. We believe this release crucially enhances the usability of the PTB-XL dataset as a reference dataset for ML methods in the context of ECG data.

BackgroundAutomatic classification of arrhythmias based on electrocardiography (ECG) data faces several significant challenges, particularly due to the substantial volume of clinical data involved in ECG signal analysis. The volume of clinical data has increased considerably, especially with the emergence of new clinical symptoms and signs in various arrhythmia conditions. These symptoms and signs, which serve as distinguishing features, can number in the tens of thousands. However, the inclusion of irrelevant features can lead to inaccurate classification results.MethodTo identify the most relevant and optimal features for ECG arrhythmia classification, common feature extraction techniques have been applied to ECG signals, specifically shallow and deep feature extraction. Additionally, a feature selection technique based on a metaheuristic optimization algorithm is utilized following the ECG feature extraction process.ResultsOur findings indicate that shallow feature extraction based on the time-domain analysis, combined with feature selection using a metaheuristic optimization algorithm, outperformed other ECG feature extraction and selection techniques. Among eight features of time-domain anaylsis, the selected feature is one to three features from RR-interval assesment, achieving 100% accuracy, sensitivity, specificity, and precision for ECG arrhythmia classification.ConclusionThe proposed end-to-end architecture for ECG arrhythmia classification demonstrates simplicity in parameters and low complexity, making it highly effective for practical applications.

World Health Organization (WHO) reported that heart disease, stroke, lower respiratory infections and chronic obstructive lung disease have remained the top major killers in the world during the past decade. In Indonesia, it covers 37% or about 1.8 millions mortality cases. Hence, human vital sign monitoring such as electrocardiograph (ECG) monitoring system is very essential for mitigating and preventing people hearth's fail to be handled a proper assistance from healthcare physicians. For that purpose, we propose a real time ECG monitoring system, which the ECG data is monitored wirelessly directly without any server for data pooling (centralized database). ECG QRS wave and arrhythmia detection are important features for ECG monitoring system. Up to now, there are several studies ECG wave detection, however, they are mainly focus on internet of things - based telemedicine system for data transfer. In this case, real time monitoring system could be disrupted by a pooling time and packet loss due to its IP-based connection. Hence, we propose a wireless ECG patient monitoring system based on point to point communication scheme at 2.4 GHz band by using 3-lead acquisition system on single channel wireless ECG device. The system is developed based on AD8232 chip for acquisition, NRF24101 wireless link platform for data transfer and Arduino™microcontroller for ECG data processing. In order to evaluate the performance of wireless ECG system, our developed wearable textile antennas (as Tx and Rx antenna) are attached on the human body and experimentally evaluated within indoor environment, simulating off-/on-body communication system. The wearable textile antenna is lightweight, small and insensitive to the resonant frequency due to the human body effect. The measurement is conducted by embedding the antennas at some body positions of the volunteers. The experimental results show that the proposed ECG monitoring system is able to transfer and display real time ECG signals well for off-/on-body communication schemes. In addition, to validate the results, the obtained ECG data are compared and analyzed with the healthcare clinic equipment. According to QRS wave analysis, the ECG signals have reasonable RR, PR, QRS and ST wave interval in comparison with the standard ECG signals. We convince that our developed ECG patient monitoring system is stable by applying our developed textile antennas and it is promising to be adopted in the hospital for future implementation.

Background: Desmin null mice (Des-/-) develop a multisystem disorder involving cardiac, skeletal, and smooth muscle. Myocardial abnormalities include progressive degeneration, cardiomyocyte death, extensive calcification and fibrosis which lead to a dilated cardiomyopathy pattern. The aim of this study is to investigate the presence of arrhythmias in a Des-/- mouse model as compared to wild type (WT) mice. Methods: Mice lacking desmin were generated by gene targeting via homologous recombination as previously describe. Electrocardiographic (ECG) recordings were obtained in 8-months old Des-/- and WT mice (n=17) using a Data Sciences International Telemetry System (St. Paul, MN, USA). ECG data were collected for 5 min every 30 min for 24 consecutive hour, giving 240 min of recording. Arrhythmias were recorded and heart rate, QRS wave duration and PR and QTc (Bazett's formula) intervals and were measured using Ponemah software. Mice also underwent a swimming stress test protocol of 20min total duration under ECG recording. Results: Premature ventricular beats were more often in Des-/- mice (n=16, 85.5±37.9/240min, range 0-612) when compared to the WT (n=8, 3.9±1.3/240min, range 0-15). Heart rate (523±20 vs 566±19 bpm), QRS (16.9±1.3 vs 17.0±0.9 ms) and PR duration (37.1±1.1 vs 34.1±1.5 ms, p=0.07) did not differ between Des-/- and WT mice. Only QTc interval was prolonged in Des-/- mice compared to the WT (186.1±5.9 vs 173.3±3.6 ms, p=0.04). Supraventricular ectopic beats were only present in Des-/- mice (range 3-111). During endurance, more premature ventricular beats were observed in Des-/- mice (range 14-510) compared to the WT (range 2-14) but first and second degree AV block were only present in Des-/- mice. Conclusion: Ventricular and supraventricular arrhythmias are common in Des-/- mice compared to the WT making this genetic model a possible experimental model of arrhythmogenic cardiomyopathy.

Background and Objective: The reliability of low-cost mobile systems for recording Electrocardiographic (ECG) data is mostly unknown, posing questions regarding the quality of the recorded data and the validity of the extracted physiological parameters. The present study compared the BITalino toolkit with an established medical-grade ECG system (BrainAmp-ExG). Methods: Participants underwent simultaneous ECG recordings with the two instruments while watching pleasant and unpleasant pictures of the “International Affective Picture System” (IAPS). Common ECG parameters were extracted and compared between the two systems. The Intraclass Correlation Coefficients (ICCs) and the Bland–Altman Limits of Agreement (LoA) method served as criteria for measurement agreement. Results: All but one parameter showed an excellent agreement (>80%) between both devices in the ICC analysis. No criteria for Bland–Altman LoA and bias were found in the literature regarding ECG parameters. Conclusion: The results of the ICC and Bland–Altman methods demonstrate that the BITalino system can be considered as an equivalent recording device for stationary ECG recordings in psychophysiological experiments.

This brief presents an ultralow power electrocardiography (ECG) feature extraction engine. ECG signal represents the cardiac cycle and contains key features, such as QRS complex, ${P}$ -wave, and ${T}$ -wave, which provide important diagnostic information about cardiovascular diseases. The ECG feature extraction is based on combined techniques of curve length transform (CLT) and discrete wavelet transform. A novel pipelined architecture for implementing CLT is proposed. The system was fabricated using GF-65-nm technology and consumed 642 nW only when operating at a frequency of 7.5 kHz from a supply voltage of 0.6 V. Ultralow power consumption of the SoC made it suitable for self-powered wearable devices.

In healthcare, electrocardiography (ECG) sensors generate a large amount of heart electrical signal that must be efficiently compressed to enable fast data transfer and reduce storage costs. Existing methods for ECG data compression do not fully exploit the characteristics of ECG signals, leading to suboptimal compression. This study proposes a data compression technique for ECG data by exploiting the known characteristics of ECG signals. Our approach combines Savitzky-Golay filtering, detrending, discrete cosine transform, scalar quantization, run-length encoding, and Huffman coding for the effective compression. To optimize the compression performance, we generated quantization intervals tailored to the ECG data characteristics. The proposed method experimentally produces a high compression ratio of 127.61 for a design parameter K = 8, a minimum percentage root mean square difference of 1.03% for K = 128, and a maximum quality score (QS) of 39.78, where K is the number of quantization intervals. Moreover, we compared the proposed method to state-of-the-art methods on a widely used ECG benchmark dataset. We found that the proposed method outperforms the others in terms of the QS, which measures the overall compression-decompression ability. By enabling more storage and faster data transfer, the proposed method can facilitate the widespread use and analysis of large volumes of ECG data, thereby contributing to advances in healthcare.

Introduction: Atrial fibrillation (AF) is one of the most prevalent cardiac arrhythmias, yet current screening and monitoring methods require contact-based electrocardiographic (ECG) equipment, leading to higher costs and potential diagnostic delays. Facial remote photoplethysmography (rPPG) uses video to capture subtle light absorption changes in the skin and is a promising avenue for accessible AF monitoring. While algorithms exist for detecting AF from R-R intervals extracted from ECG, the inherent noisiness of rPPG signals makes direct application difficult. Deep learning models have been proposed; however, they require high computational costs and offer limited explainability. Aims: This study aimed to develop and validate a low-cost, explainable approach to detect AF from beat-to-beat (B-B) intervals extracted from facial rPPG. Our goal was to avoid complex model architectures and to instead train a simple model using only R-R intervals extracted from ECG data. Methods: We first trained a model using 1172 single-lead recordings from the 2017 CinC Challenge dataset, of which 567 were labeled as AF. From these recordings, R-R intervals were extracted and then used for computing seven handcrafted features. We then trained a support vector machine (SVM) model solely on this dataset. For external validation, we conducted a prospective clinical study in which facial video (320×240 pixels, 150 Hz) and 12-lead ECG were recorded from 15 AF and 52 normal sinus rhythm (NSR) patients. ECGs were annotated by board-certified physicians. Videos were segmented into 30-second clips. The green channel was isolated, bandpass filtered and then temporally smoothed. Signals from the forehead and cheek regions were averaged to produce a representative rPPG signal. A heuristic algorithm was used to assess signal quality. Beat onsets were detected using an adaptive thresholding algorithm and then used to compute the B-B intervals for classification. Results: The model has achieved a sensitivity of 100.00%, a specificity of 97.73% and an area under the curve (AUC) of 0.9773. The indeterminate rate due to low signal quality was 11.94%. Conclusion: We haved successfully demonstrated that an interpretable, low-cost machine learning model trained solely on ECG data can accurately detect AF from facial rPPG. This shows significant promise for developing accessible, low-cost, non-contact rPPG-based systems for AF screening and monitoring, with possible future applications to other arrhythmias.

Electrocardiographic features of patients with COVID-19: One year of unexpected manifestations

: We compared the electrocardiographic (ECG) and heart rate data from a multi lead electrocardiographic recorder to the pulse rate data obtained from an oxygen saturation (O2 Sat) monitor in 77 pediatric patients undergoing exercise testing for the evaluation of exercise induced bronchospasm (EIB). PURPOSE: To assess the accuracy of pulse rate data during exercise in pediatric patients tested for EIB and the ability to document significant diagnostic findings using a multi lead electrocardiographic recorder. METHODS: Seventy-seven pediatric patients were evaluated for EIB utilizing a treadmill protocol designed to exhaust the patient within 6 to 8 minutes and elicit an 85% or greater maximal heart rate response for the last 4 minutes of the protocol. The patients were continuously monitored with a GE Medical Case 8000 ECG recorder and a Nellcor N-595 O2 Sat monitor utilizing a temple oxisensor. The heart rate (HR) and pulse rate (PR) data were indexed against the measured maximal heart rate as a measure of maximal exercise intensity. RESULTS: The average age was 13.6±2.4 years old with 36 males and 71 Caucasian, 5 African Americans and 1 Asian. Weight was 54±16Kilograms; Body Mass Index (BMI) was 21±4. The referral pattern for the patients was 57% from Pediatric Pulmonologists, 26% from community based Pediatricians, 14% from Pediatric Cardiologists and 3% from Pediatric Thoracic Surgeons. The O2 Sat monitor pulse rate function failed in 4/77 patients (5.2%).TableSignificant electrocardiographic findings included pathologic ST depression in 1 patient, continued prolonged QTc during exercise in 1 patient, exercise induced ventricular arrhythmia in 20 patients and exercise induced atrial arrhythmia in 9 patients. CONCLUSIONS: These findings support the use of a multi lead electrocardiographic recording system for exercise testing in patients evaluated for EIB. This conclusion is based on the documentation of frequent exercise induced electrocardiographic abnormalities and the significantly low pulse rate readings from the O2 sat monitor.

Electrocardiography (ECG) has been a subject of research interest in human identification because it is a promising biometric trait that is believed to have discriminatory characteristics. However, features of ECGs that are recorded at different times are often likely to vary significantly. To address the variability of ECG features over multiple records, we propose a new methodology for human identification using ECGs recorded on different days. To demonstrate the applicability of our method, we use the publicly available ECG ID dataset. The main goal of this work is to extract the most significant and discriminative wavelet components of the ECG signal, followed by utilizing the ECG spectral change for human identification using multi-level filtering technique. Our proposed multi-channel identification system is based on using the Maximal Overlap Discrete Wavelet Transform (MODWT) and its inverse (the IMODWT) to create multiple filtered ECG signals. The discriminative feature that we utilize for human identification is based on modeling the dynamic change of the frequency components in these multiple filtered signals. To reach the best possible identification performance, we use the Weighted Majority Voting Method (WMVM) for ECG classification. We evaluated the robustness of our proposed method over several random experiments and obtained 92.29% average identification accuracy, 0.9495 precision, 0.9229 recall, 0.0771 FRR and 0.0013 FAR. These results indicate that filtering some of the ECG wavelet components along with performing data fusion technique can be utilized for human identification.

Funding Acknowledgements None Background Acute myocarditis is a clinical and pathological condition defined as an inflammation of the myocardium. Its diagnosis is often challenging and requires multiple information derived from different diagnostic modalities. Purpose The aim of the study is to evaluate the correlation between electrocardiographic and imaging data in patients with acute myocarditis. Methods We made a retrospective analysis of 102 patients admitted to our Centre between January 2012 and April 2019 for suspected acute myocarditis. Diagnosis was confirmed with cardiac magnetic resonance (CMR) by identification of myocardial edema in T2-weighted images and/or typical subepicardial or midwall pattern of late gadolinium enhancement (LGE). Significant coronary artery disease was ruled out with coronary angiography. Electrocardiogram (ECG) was analysed on admission - in order to evaluate the presence of ST segment abnormalities, atrio-ventricular or bundle-branch block and heart rhythm disorders - and at the time of discharge. Every patient underwent echocardiography and CMR: from both these exams we reported the presence of regional wall motion abnormalities and left ventricular ejection fraction (LVEF). Results Mean age of our population was 39 ± 18 years; 92 people (90%) were males. At admission, 85 patients (83%) presented ECG abnormalities; the most frequent was ST-segment elevation (65 cases). Conduction or rhythm disorders were observed in 26 cases (25%). At the time of discharge, 41 out of 85 patients had complete regression of ECG changes. Mean value of LVEF measured with echocardiography was 56.4 ± 7.6%. In patients with normal ECG on admission it was 59.9 ± 3.1%, whereas in patients with abnormal ECG 55.7 ± 7.9% (p = 0.045). Considering CMR, mean LVEF in the population was 58.5 ± 8.6%, varying between 64.0 ± 8.9% in the group with normal ECG and 57.4 ± 10.1% in the group with ECG abnormalities (p = 0.02). Moreover, subjects with altered ECG on admission had a higher prevalence of wall motion abnormalities both in echocardiography (47/85 – 55% vs 3/17 – 18%, p < 0.01) and in CMR (45/85 – 53% vs 3/17 – 18%, p < 0.01). Patients with ECG normalization at discharge had a higher prevalence of ST-segment elevation (88 vs 66%, p = 0.02), while the group with persistent ECG alterations had a higher incidence of AV or bundle-branch block (23 vs 7%, p = 0.048). No statistical difference was noted between these two groups regarding echocardiographic or CMR values. Conclusion In our experience evaluation of ECG at admission in patients with suspected acute myocarditis identifies a subgroup of individuals with lower values of LVEF and a higher prevalence of wall motion abnormalities both in echocardiography and in CMR, while data derived by imaging techniques had no significant predictive value on ECG evolution at the time of discharge.

Electrocardiography (ECG) is one of the simplest and fastest methods for examining the cardiovascular system. Despite its widespread use in assessing cardiac health, collecting ECG data can be a challenge, often requiring the attachment of multiple sensors to the body or the use of specialized equipment. To overcome these limitations, many research has been conducted to find alternative approaches. One promising avenue is the use of photoplethysmography (PPG) data, which has a high correlation with ECG data and can be easily collected in real-time through wearable devices. Consequently, there has been growing interest in using PPG data to generate ECG data for real-time cardiac analysis. In this study, we propose a diffusion-based PPG-to-ECG translation with dynamic region of interest (ROI) selection, using the diffusion model success in various domains, including image, video, audio, and time-series data. Our proposed methodology addresses a key limitation of traditional diffusion models, which struggle to accurately generate the QRS complex which is an essential component of ECG data. We enhance the model performance by selectively adding noise to the ROI, focusing on the most critical segments. Instead of relying on hyperparameters for ROI selection, we develop a dynamic ROI selection algorithm. Furthermore, instead of using a standard U-Net architecture commonly used in image processing, we propose using a diffusion model suitable for time-series data, DiffWave, enabling more precise and detailed extraction of signal characteristics. To validate the performance of our model, we use a benchmark comprising five bio-signal datasets. Experimental results show that, compared with the baseline RDDM, our method reduced the average RMSE from 0.533 to 0.212 and FD from 129.88 to 14.28 across five benchmark datasets.

Atrial fibrillation (AF) is the type of arrhythmia that raises possibility of severe health problems such as heart failure and stroke and it is known that a major risk factor of AF includes overweight and obesity. Based on this association between such health-related indicators, we propose a smart scale that is capable of measuring weight and electrocardiography (ECG) simultaneously. The scale was developed using Arduino Uno, a Wheatstone bridge load cell, and ECG sensors. The ECG signals were processed to compute heart rate (in other words, RR interval). The smart scale was evaluated with four healthy volunteers in terms of reliability showing high agreement with a commercial device for ECG monitoring. In addition, it implements Atrial Fibrillation (AF) detection using machine-learning classifiers including a k-Nearest Neighbor (kNN) method, a Decision Tree (DT), and a Neural Network (NN) on relatively short recordings of ECG obtained while using the scale. The root mean square of the successive differences between heart beats (RMSSD) and the Shannon entropy of the RR interval (ECG features) were extracted from ECG signals for AF detection. Performance of AF detection was tested with patients who were treated at a Cardiology Center after balancing data by applying over- and under-sampling techniques such as Synthetic Minority Over-sampling Technique (SMOTE) and the Tomek Link (T-Link) algorithm. After addressing the data imbalance, the AF detection performance of each classifier (kNN, DT, and NNs) was 98.9%, 97.8%, and 98.9% respectively. This work has successfully demonstrated weight and cardio activity monitoring features while using a scale that may help keep the records of sensitive health related indexes on a daily basis.