ML-CDAE: Multi-Lead Convolutional Denoising Autoencoder for Denoising 12-Lead ECG Signals

Blood pressure (BP) and the electrocardiographic (ECG) signal, or electrical signal of the heart, are physical measurements that provide insight into the behavior of the cardiac system. This paper presents a novel methodology, where for the first time the relationship between BP and ECG signal is shown. Initially, to perform this study, a signal sampling of ECG signals was performed on 20 patients: eighteen healthy, between 17 and 26 years old, and two with normal BP between 50 and 78 years old. Powerlab equipment was used to record the ECG signal, with electrodes used to capture the heart signal through the lead. Once the signal samples were obtained, the R and T waves in particular were studied with the aim of reading the systolic and diastolic blood pressure separately. In order to obtain the BP with the ECG signals, we used a wavelet transform to identify the R waves and T waves, and then to perform segmentation on the signal and extract the systole and diastole portions from the original signal. Following this procedure, neural networks were applied in order to have a system with systolic and diastolic pressure values based on the ECG signals. This application led to a total success rate of 97.305% for systole and 95.65% for diastole. In conclusion, this article can be said to demonstrate the existence of a relationship between BP and ECG signals. Blood pressure (BP) and the electrocardiographic (ECG) signal, or electrical signal of the heart, are physical measurements that provide insight into the behavior of the cardiac system. This paper presents a novel methodology, where for the first time the relationship between BP and ECG signal is shown. Initially, to perform this study, a signal sampling of ECG signals was performed on 20 patients: eighteen healthy, between 17 and 26 years old, and two with normal BP between 50 and 78 years old. Powerlab equipment was used to record the ECG signal, with electrodes used to capture the heart signal through the lead. Once the signal samples were obtained, the R and T waves in particular were studied with the aim of reading the systolic and diastolic blood pressure separately. In order to obtain the BP with the ECG signals, we used a wavelet transform to identify the R waves and T waves, and then to perform segmentation on the signal and extract the systole and diastole portions from the original signal. Following this procedure, neural networks were applied in order to have a system with systolic and diastolic pressure values based on the ECG signals. This application led to a total success rate of 97.305% for systole and 95.65% for diastole. In conclusion, this article can be said to demonstrate the existence of a relationship between BP and ECG signals.

The continuous blood pressure measurement research is widely known for helpingthe development of ambulatory blood pressure monitoring where it measures blood pressureevery 15 to 30 minutes throughout the day. The cuff is a problem for the patient withAmbulatory Blood Pressure Monitor. It can make a person feel uncomfortable and must staystill when the cuff starts to inflate. It is limiting and disturbing their daily activity when thedevice is starting to measure the blood pressure. Blood pressure measurement without cuff isbeing proposed in this research, called cuff-less blood pressure measurement. It will be based onPhotoplethysmography (PPG) and Electrocardiography (ECG) signal analysis. ECG (Lead 1,Lead 2, and Lead 3) with PPG signal produced from index finger on the left hand are comparedand analyzed. Then the relation of PPG and ECG signal and the optimum location for daily usecan be obtained. The optimum location will be based on the electrode’s position that producedthe optimum ECG lead Signal to measure blood pressure. Based on the result, PPG and ECGsignal have a linear relation with Blood Pressure Measurement and Lead 1 is more stable inproducing the ECG signal. The equation from Lead 1 appeared as one of the optimum equationsfor measuring Systolic Blood Pressure (SBP) or Diastolic Blood Pressure (DBP).

Background: Electrocardiography (ECG) signals are commonly used to detect cardiac disorders, with 12-lead ECGs being the standard method for acquiring these signals. The primary objective of this research is to propose a new feature engineering model that achieves both high classification accuracy and explainable results using ECG signals. To this end, a symbolic language, named Cardioish, has been introduced. Methods: In this research, two publicly available datasets were used: (i) a mental disorder classification dataset and (ii) a myocardial infarction (MI) dataset. These datasets contain ECG beats and include 4 and 11 classes, respectively. To obtain explainable results from these ECG signal datasets, a new explainable feature engineering (XFE) model has been proposed. The Cardioish-based XFE model consists of four main phases: (i) lead transformation and transition table feature extraction, (ii) iterative neighborhood component analysis (INCA) for feature selection, (iii) classification, and (iv) explainable results generation using the recommended Cardioish. In the feature extraction phase, the lead transformer converts ECG signals into lead indexes. To extract features from the transformed signals, a transition table-based feature extractor is applied, resulting in 144 features (12 × 12) from each ECG signal. In the feature selection phase, INCA is used to select the most informative features from the 144 generated, which are then classified using the k-nearest neighbors (kNN) classifier. The final phase is the explainable artificial intelligence (XAI) phase. In this phase, Cardioish symbols are created, forming a Cardioish sentence. By analyzing the extracted sentence, XAI results are obtained. Additionally, these results can be integrated into connectome theory for applications in cardiology. Results: The presented Cardioish-based XFE model achieved over 99% classification accuracy on both datasets. Moreover, the XAI results related to these disorders have been presented in this research. Conclusions: The recommended Cardioish-based XFE model achieved high classification performance for both datasets and provided explainable results. In this regard, our proposal paves a new way for ECG classification and interpretation.

Electrocardiography (ECGs) signals are the primary means by which physicians diagnose cardiovascular-related illnesses such as abnormal heart rhythms, heart attack, and rheumatic heart. Automatically analyzing electrocardiogram (ECG) signals is a complex machine learning problem. This is because ECG waveforms can exhibit significant variability in their morphological (shape) and temporal (time-based) characteristics across different individuals. Doctors can reliably analyze electrocardiogram (ECG) signals using visual inspection of the signal waveform. However, doctors often find it challenging to analyze lengthy ECG records within a short time frame. Furthermore, the human eye has limitations in detecting subtle morphological variations within ECG signals. Although ECG signals can reveal a diverse range of heart conditions, the task of observing and categorizing long-term ECG beats can be challenging even for experts. Furthermore, because of the large volume of data, there is a significant risk of missing important information. As a result, effective computational techniques are essential to tackle this challenge. This paper introduces a deep learning approach for improving the classification of electrocardiogram (ECG) signals. The novelty in our approach is applying range normalization, which scales input data to a range of 0 to 1 before feeding it into neural network layers. The method classifies ECG signals into five categories, evaluated using the Massachusetts Institute of Technology and Boston Hospital and PTB datasets and adhering to AAMI standards. A comparison of normalization techniques with a convolutional neural network (CNN) classifier shows that the proposed method achieves average F1-scores of 99%, 85%, 95%, 81%, and 99% for the N, S, V, F, and Q classes, respectively. The overall accuracy of 98.73% demonstrates that the proposed technique outperforms existing methods. Received: 6 August 2024 | Revised: 10 February 2025 | Accepted: 19 February 2025 Conflicts of Interest The authors declare that they have no conflicts of interest to this work. Data Availability Statement Data sharing is not applicable to this article as no new data were created or analyzed in this study. Author Contribution Statement Jonah Kenei: Conceptualization, Formal analysis, Writing – original draft, Writing – review & editing, Supervision, Project administration. Juliet Moso: Methodology, Software, Validation, Investigation, Resources, Data curation, Visualization.

The continuous analysis of electrocardiographic (ECG) signals is complicated by morphological variability in the ECG due to movement of the heart. By aligning vectorcardiographic loops, movement-induced ECG variations can be partly corrected for. Existing methods for loop alignment can account for loop rotation, scaling, and time delays, but they lack the possibility to include a priori information on any of these transformations, and they are unreliable in case of low-quality signals, such as fetal ECG signals. The inclusion of a priori information might aid in the robustness of loop alignment and is, hence, proposed in this paper. We provide a generic Bayesian framework to derive our loop alignment method. In this framework, existing methods can be readily derived as well, as a simplification of our method. The loop alignment is evaluated by comparing its performance in loop alignment to two existing methods, for both adult and fetal ECG recordings. For the adult ECG recordings, a quantitative performance assessment shows that the developed method outperforms the existing method in terms of robustness. For the fetal ECG recordings, it is demonstrated that the developed method can be used to correct ECG signals for movement-induced morphology changes (enabling diagnostics) and that the method is capable of classifying recorded ECG signals to periods of fetal movement or rest ( 0.01). This information on fetal movement can also serve as a valuable diagnostic tool.

This paper introduced an efficient compression technique that uses the compressive sensing (CS) method to obtain and recover sparse electrocardiography (ECG) signals. The recovery of the signal can be achieved by using sampling rates lower than the Nyquist frequency. A novel analysis was proposed in this paper. To apply CS on ECG signal, the first step is to generate a sparse signal, which can be obtained using Modified Discrete Cosine Transform (MDCT) on the given ECG signal. This transformation is a promising key for other transformations used in this search domain and can be considered as the main contribution of this paper. A small number of wavelet components can describe the ECG signal as related work to obtain a sparse ECG signal. A sensing technique for ECG signal compression, which is a novel area of research, is proposed. ECG signals are introduced randomly between any successive beats of the heart. MIT-BIH database can be represented as the experimental database in this domain of research. The MIT-BIH database consists of various ECG signals involving a patient and standard ECG signals. MATLAB can be considered as the simulation tool used in this work. The proposed method's uniqueness was inspired by the compression ratio (CR) and achieved by MDCT. The performance measurement of the recovered signal was done by calculating the percentage root mean difference (PRD), mean square error (MSE), and peak signal to noise ratio (PSNR) besides the calculation of CR. Finally, the simulation results indicated that this work is one of the most important works in ECG signal compression.

The electrocardiographic (ECG) signal is a transthoracic manifestation of the electrical activity of the heart and is widely used in clinical applications. This chapter describes an unbiased linear adaptive filter (ULAF) to attenuate high-frequency random noise present in ECG signals. The ULAF does not contain a bias in its summation unit and the filter coefficients are normalized. During the adaptation process, the normalized coefficients are updated with the steepest-descent algorithm to achieve efficient filtering of noisy ECG signals. A total of 16 ECG signals were tested in the adaptive filtering experiments with the ULAF, the least-mean-square (LMS), and the recursive-least-squares (RLS) adaptive filters. The filtering performance was quantified in terms of the root-mean-squared error (RMSE), normalized correlation coefficient (NCC), and filtered noise entropy (FNE). A template derived from each ECG signal was used as the reference to compute the measures of filtering performance. The results indicated that the ULAF was able to provide noise-free ECG signals with an average RMSE of 0.0287, which was lower than the second-best RMSE obtained with the LMS filter. With respect to waveform fidelity, the ULAF provided the highest average NCC (0.9964) among the three filters studied. In addition, the ULAF effectively removed more noise, measured by FNE, in comparison with the LMS and RLS filters in most of the ECG signals tested. The issues of adaptive filter setting for noise reduction in ECG signals are discussed at the end of this chapter.

A convolutional autoencoder framework for ECG signal analysis.

Electrocardiograph (ECG) is a diagnostic tool that can record the electrical activity of the human heart. By analyzing the resulting waveforms of the recorded electrical activity of the heart, it is possible to record and diagnose disease. Given the importance of the ECG recording device, it is necessary to check the function of the ECG recording device, namely by performing a device calibration procedure using the Phantom ECG which aims to simulate the ECG signal. The purpose of this research is to check the ECG device during repairs, besides that the Electrocardiograph (EKG) tool functions for research purposes on ECG signals or for educational purposes. Electrocardiograph (EKG) simulator or often called Phantom ECG is in principle a signal generator in the form of an ECG like signal or a recorded ECG signal. This device can be realized based on microcontroller and analog circuit. The advantage of this simulator research is that the ECG signal displayed is the original ECG recording and has an adequate ECG signal database. ECG This simulator also has the advantage of providing convenience for research on digital signal processing applications for ECG signal processing. In its application this simulator can be used as a tool to study various forms of ECG signals. Based on the measurement results, the error value at BPM 30 and 60 is 0.00% at the sensitivity of 0.5mV, 1.0mV, and 2.0mV, then the measurement results for the error value at BPM 120 are 0.33% and at the BPM 180 value, the error value is 0.22%. From these results, it can be concluded that the highest error value is at BPM 120 with sensitivities of 0.5mV, 1.0mV, and 2.0mV.

In this study, a new hybrid prediction model was proposed by combining ECG (Electrocardiography) and PPG (Photoplethysmographic) signals with a repetitive neural network (RNN) structure to estimate blood pressure continuously. The proposed method consists of two steps. In the first step, a total of 22 time-domain attributes were obtained from PPG and ECG signals to estimate systolic blood pressure (SBP) and diastolic blood pressure (DBP) values. In the second phase, these time-domain attributes are set as input to the RNN model and then the blood pressure is estimated. Within the RNN structure, there are two-way long short-term memory BLSTM (Bidirectational Long-Short Term Memory), LSTM and ReLU (Rectified-Linear unit) layers. The bidirectional LSTM layer has been used to remove the negative affects the blood pressure value of past and future effects of nonlinear physiological changes. The LSTM layers has ensured that learning is deep and that mistakes made are reduced. The ReLU layer has been allowed the neural network to quickly reach its conclusion. The same ECG and PPG signals obtained from the database have been cleared from noise and artifacts. And then ECG and PPG signals have been segmented according to peak values of these signals. The results have shown that RMSE (Root Mean Square Error) between the estimated SBP and the measured SBP with RNN model was 3.63 and the RMSE between the estimated DBP and the measured DBP values was 1.48 with RNN model. It has been seen that the used model has a more learning ability. Thanks to the proposed method, a calibration free blood pressure measurement system using PPG and ECG signals, was developed.

A microprocessor measurement system is developed for simultaneously capturing electrocardiography (ECG) and impedance plethysmogram (IPG) signals. All the measurement electrodes are coupled with human skin in a non-contact capacitive way. A capacitive coupled pickup circuit is used for obtaining the original ECG and IPG signals. Three differences of electrode configurations are experimentally tested for picking up the ECG signals, i.e., from the plantar area of the feet, from the instep area of the feet, and from a single hand and a single foot. Two different configurations (i.e., two-electrode IPG and four-electrode IPG) are tested to get IPG signals. A singular spectrum analysis (SSA)-based algorithm is embedded in a microprocessor system for signal processing. It is found that the original signals, especially the IPG signals, are quite noisy and can only be used after SSA processing. When both the ECG and IPG signals are measured simultaneously, the ECG signal will be severely ruined by the interference from the excitation current signal of IPG measurement. Relatively satisfied signals can only be simultaneously obtained when one ECG electrode is placed in a hand. The time difference between the ECG R-peak and IPG C-peak (the peak point in IPG) is experimentally tested for its potential application in blood pressure measurement.

Electrocardiographic (ECG) signal is essential to diagnose and analyse cardiac disease. However, ECG signals are susceptible to be contaminated with various noises, which affect the application value of ECG signals. In this paper, we propose an ECG signal de-noising method using wavelet energy and a sub-band smoothing filter. Unlike the traditional wavelet threshold de-noising method, which carries out threshold processing for all wavelet coefficients, the wavelet coefficients that require threshold de-noising are selected according to the wavelet energy and other wavelet coefficients remain unchanged in the proposed method. Moreover, The sub-band smoothing filter is adopted to further de-noise the ECG signal and improve the ECG signal quality. The ECG signals of the standard MIT-BIH database are adopted to verify the proposed method using MATLAB software. The performance of the proposed approach is assessed using Signal-To-Noise ratio (SNR), Mean Square Error (MSE) and percent root mean square difference (PRD). The experimental results illustrate that the proposed method can effectively remove noise from the noisy ECG signals in comparison to the existing methods.

Identification of full-night sleep parameters using morphological features of ECG signals: A practical alternative to EEG and EOG signals

Given that current cuffless blood pressure (BP) measurement technologies feature acceptable overall accuracy, this paper proposed a sufficiently accurate cuffless BP estimation method based on photoplethysmography (PPG) and electrocardiography (ECG) signals. This study used single-channel PPG and ECG signals to estimate heart rate (HR), diastolic BP (DBP), and systolic BP (SBP). A modified long-term recurrent convolutional network comprising a multi-scale convolution network and a long short-term memory (LSTM) network was used to develop a deep learning model for accurately estimating BP and HR. The PPG and ECG signal data of 1551 patients were obtained from the Data Sets-UCI Machine Learning Repository of the University of California, Irvine. The study dataset comprised ECG, PPG, and arterial BP (ABP) signals from the PhysioNet MIMIC II dataset. The original signals were processed by removing noise and artifacts. The aforementioned dataset contains 12,000 records in a hierarchical data format, with each record containing three signals, namely 125-Hz ECG signals from channel II (ECG lead II), 125-Hz PPG signals from the fingertip, and 125-Hz invasive ABP signals. To validate the stability and performance of the developed model, ten-fold cross-validation was conducted. The mean absolute error (MAE) (standard deviation (SD)) values of the developed model for predicting SBP, DBP, and HR were 2.24 mmHg (3.59 mmHg), 1.40 mmHg (2.56 mmHg), and 0.84 bpm (2.23 bpm), respectively. In addition, the estimated SBP and DBP values satisfied the standards of the British Hypertension Society and the Association for the Advancement of Medical Instrumentation. Compared with the methods proposed in other studies, the deep learning model developed in this study required a lower number of layers to provide accurate SBP, DBP, and HR estimations. The results of this study confirmed the effectiveness of the proposed deep learning architecture.

Classification of Obstructive Sleep Apnoea from single-lead ECG signals using convolutional neural and Long Short Term Memory networks