Electrocardiogram (ECG) is a noninvasive, inexpensive, and convenient test for diagnosing cardiovascular diseases and assessing the risk of cardiovascular events. Although there are clear standardized operations and procedures for ECG examination, the interpretation of ECG by even trained physicians can be biased due to differences in diagnostic experience. In recent years, artificial intelligence has become a powerful tool to automatically analyze medical data by building deep neural network models, and has been widely used in the field of medical image diagnosis such as CT, MRI, ultrasound and ECG. This article mainly introduces the application progress of deep neural network models in ECG diagnosis and prediction of cardiovascular diseases, and discusses its limitations and application prospects.
Fetal electrocardiogram signal extraction is of great significance for perinatal fetal monitoring. In order to improve the prediction accuracy of fetal electrocardiogram signal, this paper proposes a fetal electrocardiogram signal extraction method (GA-LSTM) based on genetic algorithm (GA) optimization with long and short term memory (LSTM) network. Firstly, according to the characteristics of the mixed electrocardiogram signal of the maternal abdominal wall, the global search ability of the GA is used to optimize the number of hidden layer neurons, learning rate and training times of the LSTM network, and the optimal combination of parameters is calculated to make the network topology and the mother body match the characteristics of the mixed signals of the abdominal wall. Then, the LSTM network model is constructed using the optimal network parameters obtained by the GA, and the nonlinear transformation of the maternal chest electrocardiogram signals to the abdominal wall is estimated by the GA-LSTM network. Finally, using the non-linear transformation obtained from the maternal chest electrocardiogram signal and the GA-LSTM network model, the maternal electrocardiogram signal contained in the abdominal wall signal is estimated, and the estimated maternal electrocardiogram signal is subtracted from the mixed abdominal wall signal to obtain a pure fetal electrocardiogram signal. This article uses clinical electrocardiogram signals from two databases for experimental analysis. The final results show that compared with the traditional normalized minimum mean square error (NLMS), genetic algorithm-support vector machine method (GA-SVM) and LSTM network methods, the method proposed in this paper can extract a clearer fetal electrocardiogram signal, and its accuracy, sensitivity, accuracy and overall probability have been better improved. Therefore, the method could extract relatively pure fetal electrocardiogram signals, which has certain application value for perinatal fetal health monitoring.
Using LabVIEW programming and highspeed multifunction data acquisition card PCI6251, we designed an electrocardiogram (ECG) signal generator based on Chinese typical ECG database. When the ECG signals are given off by the generator, the generator can also display the ECG information annotations at the same time, including waveform data and diagnostic results. It could be a useful assisting tool of ECG automatic diagnose instruments.
Based on the capacitance coupling principle, we studied a capacitive way of non-contact electrocardiogram (ECG) monitoring, making it possible to obtain ECG on the condition that a patient is habilimented. Conductive fabric with a good electrical conductivity was used as electrodes. The electrodes fixed on a bed sheet is presented in this paper. A capacitance comes into being as long as the body gets close to the surface of electrode, sandwiching the cotton cushion, which acts as dielectric. The surface potential generated by heart is coupled to electrodes through the capacitance. After being processed, the signal is suitable for monitoring. The test results show that 93.5% of R wave could be detected for 9 volunteers and ECG with good signal quality could be acquired for 2 burnt patients. Non-contact ECG is harmless to skin, and it has advantages for those patients to whom stickup electrodes are not suitable. On the other hand, it is convenient to use and good for permanent monitoring.
Aiming at the defects that the traditional pulse transit time (PTT) detection methods are sensitive to changes in photoplethysmography (PPG) signal and require heavy computation, we proposed a new algorithm to detect PTT based on waveform time domain feature and dynamic difference threshold. We calculated the PTT by using dynamic difference threshold method to detect the R-waves of electrocardiogram (ECG), shortening the main peak detection range in PPG signal according to the characteristics of the waveform time domain, and using R wave to detect the main peak of PPG signal. We used the American MIMIC database and laboratory test data to validate the algorithm. The experimental results showed that the proposed method could accurately extract the feature points and detect PTT, and the PTT detection accuracies of the measurements and the database samples were 99.1% and 97.5%, respectively. So the proposed method could be better than the traditional methods.
Lorenz plot (LP) method which gives a global view of long-time electrocardiogram signals, is an efficient simple visualization tool to analyze cardiac arrhythmias, and the morphologies and positions of the extracted attractors may reveal the underlying mechanisms of the onset and termination of arrhythmias. But automatic diagnosis is still impossible because it is lack of the method of extracting attractors by now. We presented here a methodology of attractor extraction and recognition based upon homogeneously statistical properties of the location parameters of scatter points in three dimensional LP (3DLP), which was constructed by three successive RR intervals as X, Y and Z axis in Cartesian coordinate system. Validation experiments were tested in a group of RR-interval time series and tags data with frequent unifocal premature complexes exported from a 24-hour Holter system. The results showed that this method had excellent effective not only on extraction of attractors, but also on automatic recognition of attractors by the location parameters such as the azimuth of the points peak frequency (APF) of eccentric attractors once stereographic projection of 3DLP along the space diagonal. Besides, APF was still a powerful index of differential diagnosis of atrial and ventricular extrasystole. Additional experiments proved that this method was also available on several other arrhythmias. Moreover, there were extremely relevant relationships between 3DLP and two dimensional LPs which indicate any conventional achievement of LPs could be implanted into 3DLP. It would have a broad application prospect to integrate this method into conventional long-time electrocardiogram monitoring and analysis system.
As a novel technology, wearable physiological parameter monitoring technology represents the future of monitoring technology. However, there are still many problems in the application of this kind of technology. In this paper, a pilot study was conducted to evaluate the quality of electrocardiogram (ECG) signals of the wearable physiological monitoring system (SensEcho-5B). Firstly, an evaluation algorithm of ECG signal quality was developed based on template matching method, which was used for automatic and quantitative evaluation of ECG signals. The algorithm performance was tested on a randomly selected 100 h dataset of ECG signals from 100 subjects (15 healthy subjects and 85 patients with cardiovascular diseases). On this basis, 24-hour ECG data of 30 subjects (7 healthy subjects and 23 patients with cardiovascular diseases) were collected synchronously by SensEcho-5B and ECG Holter. The evaluation algorithm was used to evaluate the quality of ECG signals recorded synchronously by the two systems. Algorithm validation results: sensitivity was 100%, specificity was 99.51%, and accuracy was 99.99%. Results of controlled test of 30 subjects: the median (Q1, Q3) of ECG signal detected by SensEcho-5B with poor signal quality time was 8.93 (0.84, 32.53) minutes, and the median (Q1, Q3) of ECG signal detected by Holter with poor signal quality time was 14.75 (4.39, 35.98) minutes (Rank sum test, P=0.133). The results show that the ECG signal quality algorithm proposed in this paper can effectively evaluate the ECG signal quality of the wearable physiological monitoring system. Compared with signal measured by Holter, the ECG signal measured by SensEcho-5B has the same ECG signal quality. Follow-up studies will further collect physiological data of large samples in real clinical environment, analyze and evaluate the quality of ECG signals, so as to continuously optimize the performance of the monitoring system.
Recently, deep neural networks (DNNs) have been widely used in the field of electrocardiogram (ECG) signal classification, but the previous models have limited ability to extract features from raw ECG data. In this paper, a deep residual network model based on pyramidal convolutional layers (PC-DRN) was proposed to implement ECG signal classification. The pyramidal convolutional (PC) layer could simultaneously extract multi-scale features from the original ECG data. And then, a deep residual network was designed to train the classification model for arrhythmia detection. The public dataset provided by the physionet computing in cardiology challenge 2017(CinC2017) was used to validate the classification experiment of 4 types of ECG data. In this paper, the harmonic mean F1 of classification accuracy and recall was selected as the evaluation indexes. The experimental results showed that the average sequence level F1 (SeqF1) of PC-DRN was improved from 0.857 to 0.920, and the average set level F1 (SetF1) was improved from 0.876 to 0.925. Therefore, the PC-DRN model proposed in this paper provided a promising way for the feature extraction and classification of ECG signals, and provided an effective tool for arrhythmia classification.
Arrhythmia is a kind of common cardiac electrical activity abnormalities. Heartbeats classification based on electrocardiogram (ECG) is of great significance for clinical diagnosis of arrhythmia. This paper proposes a feature extraction method based on manifold learning, neighborhood preserving embedding (NPE) algorithm, to achieve the automatic classification of arrhythmia heartbeats. With classification system, we obtained low dimensional manifold structure features of high dimensional ECG signals by NPE algorithm, then we inputted the feature vectors into support vector machine (SVM) classifier for heartbeats diagnosis. Based on MIT-BIH arrhythmia database, we clustered 14 classes of arrhythmia heartbeats in the experiment, which yielded a high overall classification accuracy of 98.51%. Experimental result showed that the proposed method was an effective classification method for arrhythmia heartbeats.
Heart rate variability (HRV) is the difference between the successive changes in the heartbeat cycle, and it is produced in the autonomic nervous system modulation of the sinus node of the heart. The HRV is a valuable indicator in predicting the sudden cardiac death and arrhythmic events. Traditional analysis of HRV is based on a multi-electrocardiogram (ECG), but the ECG signal acquisition is complex, so we have designed an HRV analysis system based on photoplethysmography (PPG). PPG signal is collected by a microcontroller from human’s finger, and it is sent to the terminal via USB-Serial module. The terminal software not only collects the data and plot waveforms, but also stores the data for future HRV analysis. The system is small in size, low in power consumption, and easy for operation. It is suitable for daily care no matter whether it is used at home or in a hospital.