Beat Detection Research Articles

Generalizable Beat-by-Beat Arrhythmia Detection by Using Weakly Supervised Deep Learning.

Beat-by-beat arrhythmia detection in ambulatory electrocardiogram (ECG) monitoring is critical for the evaluation and prognosis of cardiac arrhythmias, however, it is a highly professional demanding and time-consuming task. Current methods for automatic beat-by-beat arrhythmia detection suffer from poor generalization ability due to the lack of large-sample and finely-annotated (labels are given to each beat) ECG data for model training. In this work, we propose a weakly supervised deep learning framework for arrhythmia detection (WSDL-AD), which permits training a fine-grained (beat-by-beat) arrhythmia detector with the use of large amounts of coarsely annotated ECG data (labels are given to each recording) to improve the generalization ability. In this framework, heartbeat classification and recording classification are integrated into a deep neural network for end-to-end training with only recording labels. Several techniques, including knowledge-based features, masked aggregation, and supervised pre-training, are proposed to improve the accuracy and stability of the heartbeat classification under weak supervision. The developed WSDL-AD model is trained for the detection of ventricular ectopic beats (VEB) and supraventricular ectopic beats (SVEB) on five large-sample and coarsely-annotated datasets and the model performance is evaluated on three independent benchmarks according to the recommendations from the Association for the Advancement of Medical Instrumentation (AAMI). The experimental results show that our method improves the F 1 score of supraventricular ectopic beats detection by 8%–290% and the F1 of ventricular ectopic beats detection by 4%–11% on the benchmarks compared with the state-of-the-art methods of supervised learning. It demonstrates that the WSDL-AD framework can leverage the abundant coarsely-labeled data to achieve a better generalization ability than previous methods while retaining fine detection granularity. Therefore, this framework has a great potential to be used in clinical and telehealth applications. The source code is available at https://github.com/sdnjly/WSDL-AD.

Detection of focal source and arrhythmogenic substrate from body surface potentials to guide atrial fibrillation ablation.

Focal sources (FS) are believed to be important triggers and a perpetuation mechanism for paroxysmal atrial fibrillation (AF). Detecting FS and determining AF sustainability in atrial tissue can help guide ablation targeting. We hypothesized that sustained rotors during FS-driven episodes indicate an arrhythmogenic substrate for sustained AF, and that non-invasive electrical recordings, like electrocardiograms (ECGs) or body surface potential maps (BSPMs), could be used to detect FS and AF sustainability. Computer simulations were performed on five bi-atrial geometries. FS were induced by pacing at cycle lengths of 120–270 ms from 32 atrial sites and four pulmonary veins. Self-sustained reentrant activities were also initiated around the same 32 atrial sites with inexcitable cores of radii of 0, 0.5 and 1 cm. FS fired for two seconds and then AF inducibility was tested by whether activation was sustained for another second. ECGs and BSPMs were simulated. Equivalent atrial sources were extracted using second-order blind source separation, and their cycle length, periodicity and contribution, were used as features for random forest classifiers. Longer rotor duration during FS-driven episodes indicates higher AF inducibility (area under ROC curve = 0.83). Our method had accuracy of 90.6±1.0% and 90.6±0.6% in detecting FS presence, and 93.1±0.6% and 94.2±1.2% in identifying AF sustainability, and 80.0±6.6% and 61.0±5.2% in determining the atrium of the focal site, from BSPMs and ECGs of five atria. The detection of FS presence and AF sustainability were insensitive to vest placement (±9.6%). On pre-operative BSPMs of 52 paroxysmal AF patients, patients classified with initiator-type FS on a single atrium resulted in improved two-to-three-year AF-free likelihoods (p-value < 0.01, logrank tests). Detection of FS and arrhythmogenic substrate can be performed from ECGs and BSPMs, enabling non-invasive mapping towards mechanism-targeted AF treatment, and malignant ectopic beat detection with likely AF progression.

Automatic detection of arrhythmias from an ECG signal using an auto-encoder and SVM classifier.

Millions of people around the world are affected by arrhythmias, which are abnormal activities of the functioning of the heart. Most arrhythmias are harmful to the heart and can suddenly become life-threatening. The electrocardiogram (ECG) is an important non-invasive tool in cardiology for the diagnosis of arrhythmias. This work proposes a computer-aided diagnosis (CAD) system to automatically classify different types of arrhythmias from ECG signals. First, the auto-encoder convolutional network (ACN) model is used, which is based on a one-dimensional convolutional neural network (1D-CNN)that automatically learns the best features from the raw ECG signals. After that, the support vector machine (SVM) classifier is applied to the features learned by the ACN model to improve the detection of arrhythmic beats. This classifier detects four different types of arrhythmias, namely the left bundle branch block (LBBB), right bundle branch block (RBBB), paced beat (PB), and premature ventricular contractions (PVC), along with the normal sinus rhythms (NSR). Among these arrhythmias, PVC is particularly a dangerous type of heartbeat in ECG signals. The performance of the model is measured in terms of accuracy, sensitivity, and precision using a tenfold cross-validation strategy on the MIT-BIH arrhythmia database. The obtained overall accuracy of the SVM classifier was 98.84%. The result of this model is portrayed as a better performance than in other literary works. Thus, this approach may also help in further clinical studies of cardiac cases.

Atrial fibrillation detection using ambulatory smartwatch photoplethysmography and validation with simultaneous holter recording

BackgroundAtrial fibrillation (AF)-associated embolic stroke is preventable, and AF detection may help to prevent stroke in subjects with paroxysmal AF. We aimed to evaluate the AF detection performance of smartwatch photoplethysmography (PPG) and the feasibility of ambulatory monitoring for AF detection in the daily life. Design and methodsConsecutive subjects who underwent ambulatory Holter electrocardiogram (ECG) monitoring for AF detection or AF burden evaluation were enrolled. The participants underwent 24 hours of simultaneous Holter ECG monitoring and continuous PPG recording using a Garmin smartwatch. The PPG signals were processed for noise rejection, beat detection, beat labeling, and rhythm labeling for each 5-minute segment. The accuracy of the PPG AF detection was calculated using the corresponding simultaneous Holter ECG as the AF diagnostic standard. ResultsAmong the 200 available participants, 112 participants (56%) developed AF (the AF group). The sensitivity, specificity, and positive predicted value of AF detection in participants were 97.3%, 88.6%, and 91.6%, respectively. The area under the receiver operating characteristic curve was 0.90. When the performance was analyzed in these 5-minute segments, the sensitivity, specificity, and positive predicted values of AF detection were 97.1%, 86.8%, and 89.7%, respectively. ConclusionsThis study demonstrated the feasibility of ambulatory monitoring for AF detection using a commercial smartwatch in daily life. A smartwatch may be an alternative screening tool to standard ambulatory Holter monitoring.

A Multimodal Data Fusion Technique for Heartbeat Detection in Wearable IoT Sensors

The accurate detection of heartbeats is of paramount importance in the current healthcare scenario as they act as an indicator for various underlying cardiac conditions and provides an indication of cardiorespiratory fitness. The article presents a novel multimodal data fusion technique using the discrete wavelet transform (DWT) and an application for fusing electrocardiogram (ECG), and photoplethysmogram (PPG) signals to improve beat detection accuracy in ambulatory monitoring using Internet of Things (IoT) sensors. The characteristics of interest from the input signals are first isolated in the wavelet domain and then combined to form a fused feature signal using a weighted average. The weights used are derived from a signal quality index calculation algorithm, suitable for periodic/quasiperiodic signals of different wave morphologies. The peak detection process to identify the heartbeat locations is carried out on the final fused signal. The research evaluates the algorithm performance when different types of noises at varying amplitudes corrupt the ECG and PPG signal inputs, affecting the signal-to-noise ratios (SNRs). The algorithm consistently exhibited a sensitivity of 99.69%, positive predictive value (PPV) of 99.64%, mean beat-to-beat interval relative error of 0.01, and an error spread (corresponding to 90th percentile of relative errors) of 0.02 in the −30 to 50 dB SNR range for all noise scenarios considered. The proposed algorithm exhibits improved detection sensitivities and PPVs under ambulatory conditions compared to state-of-the-art beat detection algorithms and can be used to accurately detect heartbeats where single-channel monitoring tends to fail in IoT devices.

A VLSI Chip for the Abnormal Heart Beat Detection Using Convolutional Neural Network.

The heart is one of the human body’s vital organs. An electrocardiogram (ECG) provides continuous tracings of the electrophysiological activity originated from heart, thus being widely used for a variety of diagnostic purposes. This study aims to design and realize an artificial intelligence (AI)-based abnormal heart beat detection with applications for early detection and timely treatment for heart diseases. A convolutional neural network (CNN) was employed to achieve a fast and accurate identification. In order to meet the requirements of the modularity and scalability of the circuit, modular and efficient processing element (PE) units and activation function modules were designed. The proposed CNN was implemented using a TSMC 0.18 m CMOS technology and had an operating frequency of 60 MHz with chip area of 1.42 mm and maximum power dissipation of mW. Furthermore, six types of ECG signals drawn from the MIT-BIH arrhythmia database were used for performance evaluation. Results produced by the proposed hardware showed that the discrimination rate was % with high efficiency in calculation, suggesting that it may be suitable for wearable devices in healthcare.

Enhancing dynamic ECG heartbeat classification with lightweight transformer model

Arrhythmia is a common class of Cardiovascular disease which is the cause for over 31% of all death over the world, according to WHOs' report. Automatic detection and classification of arrhythmia, as an effective tool of early warning, has recently been received more and more attention, especially in the applications of wearable devices for data capturing. However, different from traditional application scenarios, wearable electrocardiogram (ECG) devices have some drawbacks, such as being subject to multiple abnormal interferences, thus making accurate ventricular contraction (PVC) and supraventricular premature beat (SPB) detection to be more challenging. The traditional models for heartbeat classification suffer from the problem of large-scale parameters and the performance in dynamic ECG heartbeat classification is not satisfactory. In this paper, we propose a novel light model Lightweight Fussing Transformer to address these problems. We developed a more lightweight structure named LightConv Attention (LCA) to replace the self-attention of Fussing Transformer. LCA has reached remarkable performance level equal to or higher than self-attention with fewer parameters. In particular, we designed a stronger embedding structure (Convolutional Neural Network with attention mechanism) to enhance the weight of features of internal morphology of the heartbeat. Furthermore, we have implemented the proposed methods on real datasets and experimental results have demonstrated outstanding accuracy of detecting PVC and SPB.

Multilevel Classification and Detection of Cardiac Arrhythmias With High-Resolution Superlet Transform and Deep Convolution Neural Network

Atrial fibrillation and ventricular fibrillation are the two most common cardiac arrhythmia. These cardiac arrhythmias cause heart strokes and other heart complications leading increased risk of heart failure. Early and accurate cardiac arrhythmia detection is vital to preventing various heart-related diseases. Electrocardiogram (ECG) is a popular and reliable method to detect cardiac arrhythmia and heart-related diseases. However, sometimes it becomes difficult and time-consuming to interpret the ECG signal, even for a cardiac expert. This study proposes a deep learning-based method to effectively classify and detect cardiac arrhythmias (atrial fibrillation and ventricular fibrillation) using a time-frequency spectrogram of ECG records. The proposed framework utilizes superlet transform (SLT) to transform the one-dimensional (1-D) ECG signal into a two-dimensional (2-D) time-frequency spectrogram. The last layer of the pre-trained convolutional neural networks, namely, AlexNet, GoogLeNet, and DenseNet, are modified and then used to classify the ECG records into healthy heart, atrial fibrillation, and ventricular fibrillation. The proposed method is tested using ECG signals from the MIT-BIH arrhythmia database, MIT-BIH malignant database, Fantasia database, AF termination challenge database, and CU ventricular tachyarrhythmia database. The proposed method with DenseNet-201 architecture provides the best performance with an overall test accuracy of 96.2%. The proposed method reduces the system complexity as it does not require noise removal, hand-crafted features, and beat detection. Further, the proposed system automatically classifies the ECG signal into healthy heart, atrial fibrillation, and ventricular fibrillation.

A Fully Analog Autonomous QRS Complex Detection and Low-Complexity Asystole, Extreme Bradycardia, and Tachycardia Classification System

This article presents a power-efficient fully analog classifier architecture for the detection of critical cardiac abnormalities, i.e., asystole, extreme bradycardia, and tachycardia. To reduce power consumption and hardware complexity, an analog QRS complex detection circuit and arithmetic counter-based classification modules are introduced. The QRS detection circuit is autonomous and consumes an average current of 34 nA only, vis-à-vis state-of-the-art QRS detection designs that are digital signal processor (DSP) assisted and consume tens of microwatts of power. Furthermore, a heart-rate estimator provides the number of QRS complexes per minute. Each of the proposed modules is successfully validated through real electrocardiogram (ECG) test signals taken from the PhysioNet database. The proposed beat detector circuit exhibits a sensitivity of 97.85% and a positive prediction of 98.3%. Experimental results based on bedside monitor data show that the proposed classification module provides an overall sensitivity of 96.25% and a positive prediction of 96.97%. The complete classification architecture, implemented fully on analog platform, is simulated in the UMC 0.18- <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$\mu \text{m}$ </tex-math></inline-formula> CMOS process and consumes 119-nW power with an active silicon area of <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$450\times 800~\mu \text{m}$ </tex-math></inline-formula> . Moreover, the low-power implementation makes it suitable for long-term battery-operated remote ECG monitoring systems.

A hybrid approach of a deep learning technique for real-time ECG beat detection

A hybrid approach of a deep learning technique for real-time ECG beat detection

Arrhythmia and Disease Classification Based on Deep Learning Techniques

Electrocardiography (ECG) is a method for monitoring the human heart’s electrical activity. ECG signal is often used by clinical experts in the collected time arrangement for the evaluation of any rhythmic circumstances of a topic. The research was carried to make the assignment computerized by displaying the problem with encoder-decoder methods, by using misfortune appropriation to predict standard or anomalous information. The two Convolutional Neural Networks (CNNs) and the Long Short-Term Memory (LSTM) fully connected layer (FCL) have shown improved levels over deep learning networks (DLNs) across a wide range of applications such as speech recognition, prediction etc., As CNNs are suitable to reduce recurrence types, LSTMs are reasonable for temporary displays and DNNs are appropriate for preparing highlights for a more divisible area. CNN, LSTM, and DNNs are appropriate to view. The complementarity of CNNs, LSTMs, and DNNs was explored in this paper by consolidating them through a single architecture firm. Our findings show that the methodology suggested can expressively explain ECG series and of detection of anomalies through scores that beat other techniques supervised as well as unsupervised technique. The LSTM-Network and FL also showed that the imbalanced data sets of the ECG beat detection issue have been consistently solved and that they have not been prone to the accuracy of ECG-Signals. The novel approach should be used to assist cardiologists in their accurate and unbiased analysis of ECG signals in telemedicine scenarios.

Improving the Efficacy of Deep-Learning Models for Heart Beat Detection on Heterogeneous Datasets.

Deep learning (DL) has greatly contributed to bioelectric signal processing, in particular to extract physiological markers. However, the efficacy and applicability of the results proposed in the literature is often constrained to the population represented by the data used to train the models. In this study, we investigate the issues related to applying a DL model on heterogeneous datasets. In particular, by focusing on heart beat detection from electrocardiogram signals (ECG), we show that the performance of a model trained on data from healthy subjects decreases when applied to patients with cardiac conditions and to signals collected with different devices. We then evaluate the use of transfer learning (TL) to adapt the model to the different datasets. In particular, we show that the classification performance is improved, even with datasets with a small sample size. These results suggest that a greater effort should be made towards the generalizability of DL models applied on bioelectric signals, in particular, by retrieving more representative datasets.

Ear and Finger PPG Wearables for Night and Day Beat-to-Beat Interval Detection.

In this work, we study the accuracy of ear and finger photoplethysmography (PPG) based inter-beat interval (IBI) detection and estimation compared to the R-to-R interval (RRI) values derived from the electrocardiography (ECG). Seven male subjects with a mean age of 34.29±5.28 years are asked to wear simultaneously the Senbiosys earbud SBE2200 and the Senbiosys ring SBF2200 together with the Shimmer3 ECG development kit. The study includes 43 recordings with a total duration of 72.21 hours divided into 37.10 and 35.11 hours of sleep and wake recordings, respectively. The obtained results show that the earbud PPG enables a higher beat detection rate and a more accurate IBI estimation than the ring. They also show that the performance of the beat detection and estimation is significantly better for the sleep recordings compared to the wake recordings with an increase of ∼ 1.5% in the detection rate and a decrease of ∼ 1 ms and ∼ 4 ms in the mean absolute error (MAE) and the root mean square error (RMSE), respectively. Moreover, we propose a novel fusion scheme that smartly combines the IBI values from both devices and achieves a superior performance with a beat detection rate of 99.22% and an IBI estimation with MAE and RMSE values of 7.42 ms and 13.45 ms, respectively.

IKardo: An Intelligent ECG Device for Automatic Critical Beat Identification for Smart Healthcare

In medical practices, the detection of diseases highly depends on different medical tests. Electrocardiogram (ECG) technique is commonly used for heart disease diagnosis. Doctors can measure pulse and other heart boundaries with the aid of it. Fast and precise detection of forms of arrhythmia is critical while identifying heart disease. In this work, we proposed an intelligent ECG device (called iKardo) with the built-in automatic capability to classify into critical and non-critical data from an imbalanced ECG dataset for the smart IoT or Internet of Things based smart healthcare device. Particular emphasis is given to the reduction of data misclassification by converting imbalanced data into a balanced dataset using necessary techniques. This proposed iKardo helps in the accurate detection of critical ECG beats with an accuracy of 99.58% and result in a smart healthcare monitoring device that would make the disease detection fast and precise.

Applications of Machine Learning in Ambulatory ECG

The ambulatory ECG (AECG) is an important diagnostic tool for many heart electrophysiology-related cases. AECG covers a wide spectrum of devices and applications. At the core of these devices and applications are the algorithms responsible for signal conditioning, ECG beat detection and classification, and event detections. Over the years, there has been huge progress for algorithm development and implementation thanks to great efforts by researchers, engineers, and physicians, alongside the rapid development of electronics and signal processing, especially machine learning (ML). The current efforts and progress in machine learning fields are unprecedented, and many of these ML algorithms have also been successfully applied to AECG applications. This review covers some key AECG applications of ML algorithms. However, instead of doing a general review of ML algorithms, we are focusing on the central tasks of AECG and discussing what ML can bring to solve the key challenges AECG is facing. The center tasks of AECG signal processing listed in the review include signal preprocessing, beat detection and classification, event detection, and event prediction. Each AECG device/system might have different portions and forms of those signal components depending on its application and the target, but these are the topics most relevant and of greatest concern to the people working in this area.

Heart rate variability analyzed by computer science

Computer science is being used in medical applications with very good results. In this work, artificial neural networks are used as intelligent techniques that allow analyzing the heart rate variable to anticipate changes in the autonomic nervous system, which seeks to maintain the homeostasis of the organism and the execution of adaptive responses to environmental changes. This clinical examination has been little explored, and we want to contribute with this work to correct ectopic beats and artifacts detected in an electrocardiographic signal for the precise calculation of heart rate variability. The reverse propagation algorithm is used. The ectopic beat detection and correction program is validated based on the calculation of the sensitivity, specificity, and accuracy of ectopias and artifacts detection. A sample of 40 electrocardiogram signals is taken using for training and testing. With the proposed method, an accuracy of 82.34%, a sensitivity of 89.22% and a specificity of 95.78%.

Effects of temporal fine structure preservation on spatial hearing in bilateral cochlear implant users.

Typically, the coding strategies of cochlear implant audio processors discard acoustic temporal fine structure information (TFS), which may be related to the poor perception of interaural time differences (ITDs) and the resulting reduced spatial hearing capabilities compared to normal-hearing individuals. This study aimed to investigate to what extent bilateral cochlear implant (BiCI) recipients can exploit ITD cues provided by a TFS preserving coding strategy (FS4) in a series of sound field spatial hearing tests. As a baseline, we assessed the sensitivity to ITDs and binaural beats of 12 BiCI subjects with a coding strategy disregarding fine structure (HDCIS) and the FS4 strategy. For 250 Hz pure-tone stimuli but not for broadband noise, the BiCI users had significantly improved ITD discrimination using the FS4 strategy. In the binaural beat detection task and the broadband sound localization, spatial discrimination, and tracking tasks, no significant differences between the two tested coding strategies were observed. These results suggest that ITD sensitivity did not generalize to broadband stimuli or sound field spatial hearing tests, suggesting that it would not be useful for real-world listening.

B-PO02-185 DEEP LEARNING FOR ECG WAVEFORM SEGMENTATION

Deep learning has become the basis for a new generation of automated ECG interpretation algorithms that are capable of achieving human levels of accuracy for rhythm and beat detection. However, these algorithms are generally not capable of identifying P, QRS, and T waves, which are required for measuring PR interval, QRS duration, ST elevation, and QT interval. Novel deep learning techniques capable of waveform segmentation are required to create fully automated ECG interpretation systems. To describe and validate a deep learning algorithm for performing ECG waveform segmentation using real-world data. ECG was recorded using the Preventice BodyGaurdian® Heart or MINI remote monitor (sample rates of 256 and 250, respectively). A deep learning network was trained and optimized using 9166 one-to-six-minute ECG records from 8063 patients. Beats were annotated and adjudicated by at least 3 certified ECG technicians using custom software designed for waveform segmentation. The deep learning network consisted of 81 1D convolutional layers connected in a preactivation residual block design terminating with a bidirectional LSTM layer and linear activation function. The network produces P onset, P offset, QRS onset, QRS offset and T offset locations for each beat. Error was measured using real-world ECG from 505 patients not used during training. The validation dataset contained a diverse array of normal and ectopic rhythms. The average errors made by the deep learning algorithm were smaller than 4.5 msec for all waves. Average and standard deviation of the errors were: P onset 0.69 +/- 19.54 msec, P offset -0.01 +/- 23.32 msec, QRS onset 0.43 +/- 9.30 msec, QRS offset -4.29 +/- 12.32 msec, T offset -4.35 +/- 35.83 msec. On average, wave segmentation errors were less than two samples (8 msec) in the recorded ECG. This level of accuracy is comparable to intra and inter-observer errors reported for the publicly available Physionet QT database, which was annotated by two board certified electrophysiologists. As part of the Preventice BeatLogic® deep learning platform, this wave segmentation algorithm enables high-performance automated ECG interpretation for rhythms, beats and waves.

Use of the Polar V800 and Actiheart 5 heart rate monitors for the assessment of heart rate variability (HRV) in horses

Heart rate variability (HRV) is a commonly used metric in animal science to quantify an aspect of the physiological response to stressors. HRV analysis relies upon accurate detection of the electrical activity of the heart to define inter-beat interval (IBI). In horses, this can be problematic due to specific features of the equine ECG waveform and a high incidence of 2nd-degree atrioventricular (AV) block in clinically normal horses at rest.The current study validates the collection and interpretation of IBI data from two heart rate monitors (HRMs) principally designed for human personal monitoring (Polar V800 and Actiheart 5). 30 min of synchronous IBI data was collected using the two HRMs and a 4-lead telemetric ECG that is widely used in equine veterinary clinical practice (Televet 100). The study was conducted using a mixed sample (n = 17) of general riding horses, with (n = 4) and without (n = 13) cardiac dysrhythmias. HRV data was derived from each IBI data set and compared following application of a series of correction factors.There was a very strong inter-class correlation of HRV parameters from both the Actiheart 5 and Polar V800 with the Televet 100 data. Both HRMs providing an acceptable level of agreement with both time and frequency domain HRV parameters. The data from the Actiheart 5 were found to be interchangeable with the Televet 100 in respect of beat detection whereas the Polar V800 data required a 0.4 s artefact correction to give best agreement. Both HRMs have utility in equine research for demonstrating the direction and magnitude of change in HRV parameters, although users must apply appropriate artefact correction to maximise data validity.

Premature beats detection based on a novel convolutional neural network

Objective. Automatic detection of premature beats on long electrocardiogram (ECG) recordings is of great significance for clinical diagnosis. In this paper, we propose a novel deep learning model, the ECGDet, to detect premature beats, including premature ventricular contractions (PVCs) and supraventricular premature beats (SPBs) on single-lead long-term ECGs. Approach. The ECGDet is proposed based on a convolutional neural network and squeeze-and-excitation network. It outputs the probabilities that the ECG samples belong to a premature contraction. Non-max suppression was used to select the most appropriate locations for the premature beats. The ECGDet was trained and tested on the MIT-BIH arrhythmia database (MITDB) using a five-fold cross-validation approach. A novel loss calculation method was introduced in the model training process. Then it was tuned and further tested on the China Physiological Signal Challenge (2020) database (CPSCDB). Main results. The results showed that the average F1 value of PVC detection was 92.6%, while that of SPB detection was 72.2% on MITDB. The ECGDet bagged the 2nd place for PVC detection and ranked 7th place of SPB detection in the China Physiological Signal Challenge (2020). Significance. The proposed ECGDet can automatically detect premature heartbeats without manually extracting the features. This technique can be used for long-term ECG signal analysis and has potential for clinical applications.