Feature fusion for imbalanced ECG data analysis

One-shot screening: Utilization of a two-dimensional convolutional neural network for automatic detection of left ventricular hypertrophy using electrocardiograms

BACKGROUND AND OBJECTIVE

Left ventricular hypertrophy (LVH) can impair ejection function and elevate the risk of heart failure. Therefore, early detection through screening is crucial. This study aimed to propose a novel method to enhance LVH detection using 12-lead electrocardiogram (ECG) waveforms with a two-dimensional (2D) convolutional neural network (CNN).

METHODS

Utilizing 42,127 pairs of ECG-transthoracic echocardiogram data, we pre-processed raw data into single-shot images derived from each ECG lead and conducted lead selection to optimize LVH diagnosis. Our proposed one-shot screening method, implemented during pre-processing, enables the superimposition of waveform source data of any length onto a single-frame image, thereby addressing the limitations of the one-dimensional (1D) approach. We developed a deep learning model with a 2D-CNN structure and machine learning models for LVH detection. To assess our method, we also compared our results with conventional ECG criteria and those of a prior study that used a 1D-CNN approach, utilizing the same dataset from the University of Tokyo Hospital for LVH diagnosis.

RESULTS

For LVH detection, the average area under the receiver operating characteristic curve (AUROC) was 0.916 for the 2D-CNN model, which was significantly higher than that obtained using logistic regression and random forest methods, as well as the two conventional ECG criteria (AUROC of 0.766, 0.790, 0.599, and 0.622, respectively). Incorporating additional metadata, such as ECG measurement data, further improved the average AUROC to 0.921. The model's performance remained stable across two different annotation criteria and demonstrated significant superiority over the performance of the 1D-CNN model used in a previous study (AUROC of 0.807).

CONCLUSIONS

This study introduces a robust and computationally efficient method that outperforms 1D-CNN models utilized in previous studies for LVH detection. Our method can transform waveforms of any length into fixed-size images and leverage the selected lead of the ECG, ensuring adaptability in environments with limited computational resources. The proposed method holds promise for integration into clinical practice as a tool for early diagnosis, potentially enhancing patient outcomes by facilitating earlier treatment and management.

Pre-Processing techniques and artificial intelligence algorithms for electrocardiogram (ECG) signals analysis: A comprehensive review

Electrocardiogram (ECG) are the physiological signals and a standard test to measure the heart's electrical activity that depicts the movement of cardiac muscles. A review study has been conducted on ECG signals analysis with the help of artificial intelligence (AI) methods over the last ten years i.e., 2012-22. Primarily, the method of ECG analysis by software systems was divided into classical signal processing (e.g. spectrograms or filters), machine learning (ML) and deep learning (DL), including recursive models, transformers and hybrid. Secondly, the data sources and benchmark datasets were depicted. Authors grouped resources by ECG acquisition methods into hospital-based portable machines and wearable devices. Authors also included new trends like advanced pre-processing, data augmentation, simulations and agent-based modeling. The study found improvement in ECG examination perfection made each year through ML, DL, hybrid models, and transformers. Convolutional neural networks and hybrid models were more targeted and proved efficient. The transformer model extended the accuracy from 90% to 98%. The Physio-Net library helps acquire ECG signals, including the popular benchmark databases such as MIT-BIH, PTB, and challenging datasets. Similarly, wearable devices have been established as a appropriate option for monitoring patient health without the time and place limitations and are also helpful for AI model calibration with so far accuracy of 82%-83% on Samsung smartwatch. In the pre-processing signals, spectrogram generation through Fourier and wavelet transformations are erected leading approaches promoting on average accuracy of 90%-95%. Likewise, data enhancement using geometrical techniques is well-considered; however, extraction and concatenation-based methods need attention. As the what-if analysis in healthcare or cardiac issues can be performed using a complex simulation, the study reviews agent-based modeling and simulation approaches for cardiovascular risk event assessment.

Efficient Classification of ECG Images Using a Lightweight CNN with Attention Module and IoT

Cardiac disorders are a leading cause of global casualties, emphasizing the need for the initial diagnosis and prevention of cardiovascular diseases (CVDs). Electrocardiogram (ECG) procedures are highly recommended as they provide crucial cardiology information. Telemedicine offers an opportunity to provide low-cost tools and widespread availability for CVD management. In this research, we proposed an IoT-based monitoring and detection system for cardiac patients, employing a two-stage approach. In the initial stage, we used a routing protocol that combines routing by energy and link quality (REL) with dynamic source routing (DSR) to efficiently collect data on an IoT healthcare platform. The second stage involves the classification of ECG images using hybrid-based deep features. Our classification system utilizes the "ECG Images dataset of Cardiac Patients", comprising 12-lead ECG images with four distinct categories: abnormal heartbeat, myocardial infarction (MI), previous history of MI, and normal ECG. For feature extraction, we employed a lightweight CNN, which automatically extracts relevant ECG features. These features were further optimized through an attention module, which is the method's main focus. The model achieved a remarkable accuracy of 98.39%. Our findings suggest that this system can effectively aid in the identification of cardiac disorders. The proposed approach combines IoT, deep learning, and efficient routing protocols, showcasing its potential for improving CVD diagnosis and management.

Self-Attention LSTM-FCN model for arrhythmia classification and uncertainty assessment

This paper presents ArrhyMon, a self-attention-based LSTM-FCN model for arrhythmia classification from ECG signal inputs. ArrhyMon targets to detect and classify six different types of arrhythmia apart from normal ECG patterns. To the best of our knowledge, ArrhyMon is the first end-to-end classification model that successfully targets the classification of six detailed arrhythmia types and compared to previous work does not require additional preprocessing and/or feature extraction operations separate from the classification model. ArrhyMon's deep learning model is designed to capture and exploit both global and local features embedded in ECG sequences by integrating fully convolutional network (FCN) layers and a self-attention-based long and short-term memory (LSTM) architecture. Moreover, to enhance its practicality, ArrhyMon incorporates a deep ensemble-based uncertainty model that generates a confidence-level measure for each classification result. We evaluate ArrhyMon's effectiveness using three publicly available arrhythmia datasets (i.e., MIT-BIH, Physionet Cardiology Challenge 2017 and 2020/2021) to show that ArrhyMon achieves state-of-the-art classification performance (average accuracy 99.63%), and that confidence measures show close correlation with subjective diagnosis made from practitioners.

ECG Heartbeat Classification Using Machine Learning and Metaheuristic Optimization for Smart Healthcare Systems

Early diagnosis and classification of arrhythmia from an electrocardiogram (ECG) plays a significant role in smart healthcare systems for the health monitoring of individuals with cardiovascular diseases. Unfortunately, the nonlinearity and low amplitude of ECG recordings make the classification process difficult. Thus, the performance of most traditional machine learning (ML) classifiers is questionable, as the interrelationship between the learning parameters is not well modeled, especially for data features with high dimensions. To address the limitations of ML classifiers, this paper introduces an automatic arrhythmia classification approach based on the integration of a recent metaheuristic optimization (MHO) algorithm and ML classifiers. The role of the MHO is to optimize the search parameters of the classifiers. The approach consists of three steps: the preprocessing of the ECG signal, the extraction of the features, and the classification. The learning parameters of four supervised ML classifiers were utilized for the classification task; support vector machine (SVM), k-nearest neighbors (kNNs), gradient boosting decision tree (GBDT), and random forest (RF) were optimized using the MHO algorithm. To validate the advantage of the proposed approach, several experiments were conducted on three common databases, including the Massachusetts Institute of Technology (MIT-BIH), the European Society of Cardiology ST-T (EDB), and the St. Petersburg Institute of Cardiological Techniques 12-lead Arrhythmia (INCART). The obtained results showed that the performance of all the tested classifiers were significantly improved after integrating the MHO algorithm, with the average ECG arrhythmia classification accuracy reaching 99.92% and a sensitivity of 99.81%, outperforming the state-of the-art methods.

A Two-Step Approach to Overcoming Data Imbalance in the Development of an Electrocardiography Data Quality Assessment Algorithm: A Real-World Data Challenge

Continuously acquired biosignals from patient monitors contain significant amounts of unusable data. During the development of a decision support system based on continuously acquired biosignals, we developed machine and deep learning algorithms to automatically classify the quality of ECG data. A total of 31,127 twenty-s ECG segments of 250 Hz were used as the training/validation dataset. Data quality was categorized into three classes: acceptable, unacceptable, and uncertain. In the training/validation dataset, 29,606 segments (95%) were in the acceptable class. Two one-step, three-class approaches and two two-step binary sequential approaches were developed using random forest (RF) and two-dimensional convolutional neural network (2D CNN) classifiers. Four approaches were tested on 9779 test samples from another hospital. On the test dataset, the two-step 2D CNN approach showed the best overall accuracy (0.85), and the one-step, three-class 2D CNN approach showed the worst overall accuracy (0.54). The most important parameter, precision in the acceptable class, was greater than 0.9 for all approaches, but recall in the acceptable class was better for the two-step approaches: one-step (0.77) vs. two-step RF (0.89) and one-step (0.51) vs. two-step 2D CNN (0.94) (p < 0.001 for both comparisons). For the ECG quality classification, where substantial data imbalance exists, the 2-step approaches showed more robust performance than the one-step approach. This algorithm can be used as a preprocessing step in artificial intelligence research using continuously acquired biosignals.

Emotion recognition while applying cosmetic cream using deep learning from EEG data; cross-subject analysis

We report a deep learning-based emotion recognition method using EEG data collected while applying cosmetic creams. Four creams with different textures were randomly applied, and they were divided into two classes, “like (positive)” and “dislike (negative)”, according to the preference score given by the subject. We extracted frequency features using well-known frequency bands, i.e., alpha, beta and low and high gamma bands, and then we created a matrix including frequency and spatial information of the EEG data. We developed seven CNN-based models: (1) inception-like CNN with four-band merged input, (2) stacked CNN with four-band merged input, (3) stacked CNN with four-band parallel input, and stacked CNN with single-band input of (4) alpha, (5) beta, (6) low gamma, and (7) high gamma. The models were evaluated by the Leave-One-Subject-Out Cross-Validation method. In like/dislike two-class classification, the average accuracies of all subjects were 73.2%, 75.4%, 73.9%, 68.8%, 68.0%, 70.7%, and 69.7%, respectively. We found that the classification performance is higher when using multi-band features than when using single-band feature. This is the first study to apply a CNN-based deep learning method based on EEG data to evaluate preference for cosmetic creams.

Arrhythmia Classification Algorithm Based on a Two-Dimensional Image and Modified EfficientNet

The classification and identification of arrhythmias using electrocardiogram (ECG) signals are of great practical significance in the early prevention and diagnosis of cardiovascular diseases. In this study, we propose an arrhythmia classification algorithm based on two-dimensional (2D) images and modified EfficientNet. First, we developed a method for converting original one-dimensional (1D) ECG signals into 2D image signals. In contrast with the existing classification method that uses only the time-domain features of a 1D ECG signal, the classification of 2D images can consider the spatiotemporal characteristics of the signal. Then, to better assign feature weights, we introduced an attention feature fusion module (AFF) into the EfficientNet network to replace the addition operation in the mobile inverted bottleneck convolution (MBConv) structure of the network. We selected EfficientNet for modification because, compared with most convolutional neural networks (CNNs), EfficientNet does not require manual adjustment of parameters, which improves the accuracy and speed of the network. Finally, we combined the 2D images and the improved EfficientNet network and tested its performance as an arrhythmia classification method. Our experimental results show that the network training of the proposed method requires less equipment and training time, and this method can effectively distinguish eight types of heartbeats in the MIT-BIH arrhythmia database, with a classification accuracy of 99.54%. Thus, the model has a good classification effect.

Cost-Sensitive Learning for Anomaly Detection in Imbalanced ECG Data Using Convolutional Neural Networks

Arrhythmia detection algorithms based on deep learning are attracting considerable interest due to their vital role in the diagnosis of cardiac abnormalities. Despite this interest, deep feature representation for ECG is still challenging and intriguing due to the inter-patient variability of the ECG’s morphological characteristics. The aim of this study was to learn a balanced deep feature representation that incorporates both the short-term and long-term morphological characteristics of ECG beats. For efficient feature extraction, we designed a temporal transition module that uses convolutional layers with different kernel sizes to capture a wide range of morphological patterns. Imbalanced data are a key issue in developing an efficient and generalized model for arrhythmia detection as they cause over-fitting to minority class samples (abnormal beats) of primary interest. To mitigate the imbalanced data issue, we proposed a novel, cost-sensitive loss function that ensures a balanced deep representation of class samples by assigning effective weights to each class. The cost-sensitive loss function dynamically alters class weights for every batch based on class distribution and model performance. The proposed method acquired an overall accuracy of 99.81% for intra-patient classification and 96.36% for the inter-patient classification of heartbeats. The experimental results reveal that the proposed approach learned a balanced representation of ECG beats by mitigating the issue of imbalanced data and achieved an improved classification performance as compared to other studies.

An Effective and Lightweight Deep Electrocardiography Arrhythmia Recognition Model Using Novel Special and Native Structural Regularization Techniques on Cardiac Signal

Recently, cardiac arrhythmia recognition from electrocardiography (ECG) with deep learning approaches is becoming popular in clinical diagnosis systems due to its good prognosis findings, where expert data preprocessing and feature engineering are not usually required. But a lightweight and effective deep model is highly demanded to face the challenges of deploying the model in real-life applications and diagnosis accurately. In this work, two effective and lightweight deep learning models named Deep-SR and Deep-NSR are proposed to recognize ECG beats, which are based on two-dimensional convolution neural networks (2D CNNs) while using different structural regularizations. First, 97720 ECG beats extracted from all records of a benchmark MIT-BIH arrhythmia dataset have been transformed into 2D RGB (red, green, and blue) images that act as the inputs to the proposed 2D CNN models. Then, the optimization of the proposed models is performed through the proper initialization of model layers, on-the-fly augmentation, regularization techniques, Adam optimizer, and weighted random sampler. Finally, the performance of the proposed models is evaluated by a stratified 5-fold cross-validation strategy along with callback features. The obtained overall accuracy of recognizing normal beat and three arrhythmias (V-ventricular ectopic, S-supraventricular ectopic, and F-fusion) based on the Association for the Advancement of Medical Instrumentation (AAMI) is 99.93%, and 99.96% for the proposed Deep-SR model and Deep-NSR model, which demonstrate that the effectiveness of the proposed models has surpassed the state-of-the-art models and also expresses the higher model generalization. The received results with model size suggest that the proposed CNN models especially Deep-NSR could be more useful in wearable devices such as medical vests, bracelets for long-term monitoring of cardiac conditions, and in telemedicine to accurate diagnose the arrhythmia from ECG automatically. As a result, medical costs of patients and work pressure on physicians in medicals and clinics would be reduced effectively.

Classification of Imbalanced Electrocardiosignal Data using Convolutional Neural Network

BACKGROUND AND OBJECTIVE

In the application of wearable heart-monitors, it is of great significance to analyze electrocardiogram (ECG) signals for anomaly detection. ECG arrhythmia classification remains an open problem in that it cannot easily recognize data from minority classes due to the imbalanced dataset and particular characteristic of the time series signal. In this study, a novel method is presented as a possible solution to imbalanced classification problems.

METHODS

An improved data augmentation method based on variational auto-encoder (VAE) and auxiliary classifier generative adversarial network (ACGAN) is implemented to address the difficulties resulting from the imbalanced dataset. Based on the augmented dataset, convolutional neural network (CNN) classifiers are employed to automatically recognize arrhythmias using two-dimensional ECG images.

RESULTS

In experimental studies conducted with the MIT-BIH arrhythmia database, the proposed method achieves 98.45% accuracy and 97.03% sensitivity. The sensitivities of two minority classes achieve 95.83% and 97.37%, respectively.

CONCLUSION

In imbalanced classification, the sensitivity of minority class is a key evaluation indicator. One of the significant contributions of this study is that the proposed method can obtain higher sensitivity of minority class. The experimental results demonstrate that the proposed method for ECG arrhythmia calssification under imbalanced data has better performance compared with traditional cropping augmentation methods and traditional classifiers.

Diagnostic Interpretation of Non-Uniformly Sampled Electrocardiogram

We present a set of three fundamental methods for electrocardiogram (ECG) diagnostic interpretation adapted to process non-uniformly sampled signal. The growing volume of ECGs recorded daily all over the world (roughly estimated to be 600 TB) and the expectance of long persistence of these data (on the order of 40 years) motivated us to challenge the feasibility of medical-grade diagnostics directly based on arbitrary non-uniform (i.e., storage-efficient) ECG representation. We used a refined time-independent QRS detection method based on a moving shape matching technique. We applied a graph data representation to quantify the similarity of asynchronously sampled heartbeats. Finally, we applied a correlation-based non-uniform to time-scale transform to get a multiresolution ECG representation on a regular dyadic grid and to find precise P, QRS and T wave delimitation points. The whole processing chain was implemented and tested with MIT-BIH Database (probably the most referenced cardiac database) and CSE Multilead Database (used for conformance testing of medical instruments) signals arbitrarily sampled accordingly to a perceptual model (set for variable sampling frequency of 100-500 Hz, compression ratio 3.1). The QRS detection shows an accuracy of 99.93% with false detection ratio of only 0.18%. The classification shows an accuracy of 99.27% for 14 most frequent MIT-BIH beat types and 99.37% according to AAMI beat labels. The wave delineation shows cumulative (i.e., sampling model and non-uniform processing) errors of: 9.7 ms for P wave duration, 3.4 ms for QRS, 6.7 ms for P-Q segment and 17.7 ms for Q-T segment, all the values being acceptable for medical-grade interpretive software.

Heartbeat Classification Based on Multifeature Combination and Stacking-DWKNN Algorithm

Arrhythmia is one of the most common abnormal symptoms that can threaten human life. In order to distinguish arrhythmia more accurately, the classification strategy of the multifeature combination and Stacking-DWKNN algorithm is proposed in this paper. The method consists of four modules. In the preprocessing module, the signal is denoised and segmented. Then, multiple different features are extracted based on single heartbeat morphology, P length, QRS length, T length, PR interval, ST segment, QT interval, RR interval, R amplitude, and T amplitude. Subsequently, the features are combined and normalized, and the effect of different feature combinations on heartbeat classification is analyzed to select the optimal feature combination. Finally, the four types of normal and abnormal heartbeats were identified using the Stacking-DWKNN algorithm. This method is performed on the MIT-BIH arrhythmia database. The result shows a sensitivity of 89.42% and a positive predictive value of 94.90% of S-type beats and a sensitivity of 97.21% and a positive predictive value of 97.07% of V-type beats. The obtained average accuracy is 99.01%. Compared to other models with the same features, this method can improve accuracy and has a higher positive predictive value and sensitivity, which is important for clinical decision-making.

Using the VQ-VAE to improve the recognition of abnormalities in short-duration 12-lead electrocardiogram records

BACKGROUND AND OBJECTIVE

Morphological diagnosis is a basic clinical task of the short-duration 12-lead electrocardiogram (ECG). Due to the scarcity of positive samples and other factors, there is currently no algorithm that is comparable to human experts in ECG morphological recognition. Our objective is to develop an ECG specialist-level deep learning method that can accurately identify ten ECG morphological abnormalities in real scene data.

METHODS

We established a short-duration 12-lead ECG image dataset that consists of approximately 200,000 samples. To address the problems with small positive samples, a data augmentation method was proposed. We solved it by interpolating in the latent space of the vector quantized variational autoencoder (VQ-VAE) and generating new samples via sampling. The trained final classifier, general doctors, and ECG specialists evaluated the diagnostic performance on a test set that consisted of 1000 samples.

RESULTS

Relative to that of unaugmented data, the F1 score was improved by 0-6%. Compared with ECG specialists, the deep neural network achieved higher F1 scores and sensitivity in most categories.

CONCLUSIONS

Our method can improve the classification performance of ECG data with insufficient positive samples and reach the level of ECG specialists. This approach can provide specialized reference opinions for ordinary clinicians and reduce the errors of ECG specialists.

A Metaheuristic Optimization Approach for Parameter Estimation in Arrhythmia Classification from Unbalanced Data

The electrocardiogram records the heart's electrical activity and generates a significant amount of data. The analysis of these data helps us to detect diseases and disorders via heart bio-signal abnormality classification. In unbalanced-data contexts, where the classes are not equally represented, the optimization and configuration of the classification models are highly complex, reflecting on the use of computational resources. Moreover, the performance of electrocardiogram classification depends on the approach and parameter estimation to generate the model with high accuracy, sensitivity, and precision. Previous works have proposed hybrid approaches and only a few implemented parameter optimization. Instead, they generally applied an empirical tuning of parameters at a data level or an algorithm level. Hence, a scheme, including metrics of sensitivity in a higher precision and accuracy scale, deserves special attention. In this article, a metaheuristic optimization approach for parameter estimations in arrhythmia classification from unbalanced data is presented. We selected an unbalanced subset of those databases to classify eight types of arrhythmia. It is important to highlight that we combined undersampling based on the clustering method (data level) and feature selection method (algorithmic level) to tackle the unbalanced class problem. To explore parameter estimation and improve the classification for our model, we compared two metaheuristic approaches based on differential evolution and particle swarm optimization. The final results showed an accuracy of 99.95%, a F1 score of 99.88%, a sensitivity of 99.87%, a precision of 99.89%, and a specificity of 99.99%, which are high, even in the presence of unbalanced data.

Multi-branch fusion network for Myocardial infarction screening from 12-lead ECG images

BACKGROUND AND OBJECTIVE

Myocardial infarction (MI) is a myocardial anoxic incapacitation caused by severe cardiovascular obstruction that can cause irreversible injury or even death. In medical field, the electrocardiogram (ECG) is a common and effective way to diagnose myocardial infarction, which often requires a wealth of medical knowledge. It is necessary to develop an approach that can detect the MI automatically.

METHODS

In this paper, we propose a multi-branch fusion framework for automatic MI screening from 12-lead ECG images, which consists of multi-branch network, feature fusion and classification network. First, we use text detection and position alignment to automatically separate twelve leads from ECG images. Then, those 12 leads are input into the multi-branch network constructed by a shallow neural network to get 12 feature maps. After concatenating those feature maps by depth fusion, classification is explored to judge the given ECG is MI or not.

RESULTS

Based on extensive experiments on an ECG image dataset, performances of different combinations of structures are analyzed. The proposed network is compared with other networks and also compared with physicians in the practical use. All the experiments verify that the proposed method is effective for MI screening based on ECG images, which achieves accuracy, sensitivity, specificity and F1-score of 94.73%, 96.41%, 95.94% and 93.79% respectively.

CONCLUSIONS

Rather than using the typical one-dimensional electrical ECG signal, this paper gives an effective model to screen MI by analyzing 12-lead ECG images. Extracting and analyzing these 12 leads from their corresponding ECG images is a good attempt in the application of MI screening.

A High Precision Real-time Premature Ventricular Contraction Assessment Method based on the Complex Feature Set

This paper presents a high precision and low computational complexity premature ventricular contraction (PVC) assessment method for the ECG human-machine interface device. The original signals are preprocessed by integrated filters. Then, R points and surrounding feature points are determined by corresponding detection algorithms. On this basis, a complex feature set and feature matrices are obtained according to the position feature points. Finally, an exponential Minkowski distance method is proposed for PVC recognition. Both public dataset and clinical experiments were utilized to verify the effectiveness and superiority of the proposed method. The results show that our R peak detection algorithm can substantially reduce the error rate, and obtained 98.97% accuracy for QRS complexes. Meanwhile, the accuracy of PVC recognition was 98.69% for the MIT-BIH database and 98.49% for clinical tests. Moreover, benefiting from the lightweight of our model, it can be easily applied to portable healthcare devices for human-computer interaction.

ECG Signal Denoising and Features Extraction Using Unbiased FIR Smoothing

Methods of the electrocardiography (ECG) signal features extraction are required to detect heart abnormalities and different kinds of diseases. However, different artefacts and measurement noise often hinder providing accurate features extraction. One of the standard techniques developed for ECG signals employs linear prediction. Referring to the fact that prediction is not required for ECG signal processing, smoothing can be more efficient. In this paper, we employ the p-shift unbiased finite impulse response (UFIR) filter, which becomes smooth by p < 0. We develop this filter to have an adaptive averaging horizon: optimal for slow ECG behaviours and minimal for fast excursions. It is shown that the adaptive UFIR algorithm developed in such a way provides better denoising and suboptimal features extraction in terms of the output signal-noise ratio (SNR). The algorithm is developed to detect durations and amplitudes of the P-wave, QRS-complex, and T-wave in the standard ECG signal map. Better performance of the algorithm designed is demonstrated in a comparison with the standard linear predictor, UFIR filter, and UFIR predictive filter based on real ECG data associated with normal heartbeats.

AUTOMATIC DETECTION OF CARDIOVASCULAR DISEASE USING DEEP KERNEL EXTREME LEARNING MACHINE

The electrocardiogram (ECG) is a principal signal employed to automatically diagnose cardiovascular disease in shallow and deep learning models. However, ECG feature extraction is required and this may reduce diagnosis accuracy in traditional shallow learning models, while backward propagation (BP) algorithm used by the traditional deep learning models has the disadvantages of local minimization and slow convergence rate. To solve these problems, a new deep learning algorithm called deep kernel extreme learning machine (DKELM) is proposed by combining the extreme learning machine auto-encoder (ELM-AE) and kernel ELM (KELM). In the new DKELM architecture with [Formula: see text] hidden layers, ELM-AEs are employed by the front [Formula: see text] hidden layers for feature extraction in the unsupervised learning process, which can effectively extract abstract features from the original ECG signal. To overcome the “dimension disaster” problem, the kernel function is introduced into ELM to act as classifier by the [Formula: see text]th hidden layer in the supervised learning process. The experiments demonstrate that DKELM outperforms the BP neural network, support vector machine (SVM), extreme learning machine (ELM), deep auto-encoder (DAE), deep belief network (DBN) in classification accuracy. Though the accuracy of convolutional neural network (CNN) is almost the same as DKELM, the computing time of CNN is much longer than DKELM.