A new approach to adaptive threshold based method for QRS detection with fuzzy clustering

Arrhythmia classification using multirate processing metaheuristic optimization and variational mode decomposition

A review of different ECG classification/detection techniques for improved medical applications

Electrocardiogram (ECG) signal has been widely studied for biometric recognition (or subject identification) applications over the past decade. Many techniques have been offered as well, from traditional computing to machine learning techniques. However, most studies evaluate their proposed models by independently separating the datasets between healthy (normal) and unhealthy (abnormal/arrhythmia) heart conditions. Some studies even used the ECG signal datasets from subjects only under healthy heart conditions. We believe that developing an ECG-based biometric system that includes the heart, both under healthy and unhealthy conditions, is highly needed since the users might be in both conditions. To the best of our knowledge, this is the first study in the literature, which considers the combination of the dataset of ECG signals from subjects under the healthy heart and unhealthy (arrhythmia) conditions in training, validation, and even testing processes. We combined the datasets from MIT-BIH Normal Sinus Rhythm, MIT-BIH Arrhythmia, and St Petersburg INCART 12-lead Arrhythmia databases obtained from Physionet. From the combination of those databases, we obtained a total of 32,628 datasets from 89 subjects which were later modeled using a machine learning technique, i.e., one-dimensional convolutional neural network (CONV1D). This study aims to build a light-weight yet reliable ECG-based biometric system by implementing only two layers of the convolutional neural network into those joined datasets. We evaluated the model's performance with k-fold cross-validation (k = 5) and yielded the accuracy and F1-scores of 99.8% And 99.8% respectively. We also proved that our model is a good fitting (not under/over-fitting) by providing the visualization plot of accuracy and loss. These findings show that our model is robust enough for ECG-based subject identification, which can be implemented for healthy and unhealthy heart conditions. 

Implementation of One-Dimensional Convolutional Neural Network for Individual Identification Based on ECG Signal

Accurate QRS detection is an important first step for almost every electrocardiogram (ECG) signal analysis. However, detecting QRS is difficult, not only because of the large variety, but also as a result of interference caused by various types of noise. This paper employs a hybrid feature extraction technique of ECG signal for the detection of cardiac abnormalities. Noise removal and accurate QRS detection play a major role in the analysis of ECG signals. In this paper various types of noises such as additive white Gaussian noise, baseline wander and power line interference is eliminated to enhance the signal quality. This study proposes an improved QRS complex detection algorithm based on the combination of empirical mode decomposition-discrete wavelet transform (EMD-DWT) with threshold and compared with ordinary discrete wavelet transform. The system efficacy and performance have been evaluated using accuracy, sensitivity (Se), positive predictive value (PPV) and detection error rate (DER). The results show the high accuracy of the proposed EMD-DWT algorithm, which attains a detection error rate of 1.1233%, a sensitivity of 99.28%, and a positive predictive value of 99.99%, evaluated using the MIT-BIH arrhythmia database. The proposed algorithm improves the accuracy of QRS detection compared to state-of-art methods.

Pinjala N. Malleswari

Papers

A hybrid EMD-DWT based algorithm for detection of QRS complex in electrocardiogram signal

An Improved Denoising of Electrocardiogram Signals Based on Wavelet Thresholding