Regardless of the class definition, this standard essentially recommends for performance evaluation the adoption of inter-patient scenario (i.e., training and test ECG beats are extracted from different patients), which is not usually adopted in most of the works published in the literature. Specifically, the AAMI standard defines five classes of interest: normal (N), ventricular (V), supraventricular (S), fusion of normal and ventricular (F) and unknown beats (Q). ĭespite these great efforts, it has been shown recently, , that automatic methods do not perform well if the recommendations of the Association for the Advancement of Medical Instrumentation (AAMI) for class labeling and results presentation are closely followed as a possible solution for standardization. Finally, the obtained features are used to learn the decision function of a classifier such as neural networks (NN), probabilistic NN, recurrent NN, support vector machines (SVMs), , least square SVM, path forest and Gaussian processes (GPs). Then feature reduction techniques such as principal component analysis (PCA), independent component analysis (ICA), and linear discriminant analysis (LDA) are usually applied to reduce the dimensionality of the feature representation. In general, the available feature representation methods include, but are not limited to, morphology, , temporal information, , wavelet transform, , high-order statistics (HOS), Hermite basis function, and hidden Markov modeling (HMM). Then several handcrafts features are calculated from these waveforms. After this step, the ECG waveforms, also known as PQRST, consisting mainly of P wave, QRS complex and T wave are extracted by means of segmentation. First, the ECG signals are enhanced by eliminating various kinds of noise and artifacts (i.e., baseline wanders, power line interference, and muscle contraction), ,.
Usually, these approaches are based on three main steps which are preprocessing, feature extraction and classification. In the last decades, several pattern recognition methods were developed for arrhythmia detection and classification, ,, ,. Therefore, the utilization of computer-based methods represents an important solution that can benefit cardiologists in the diagnosis. A careful inspection of its behavior is essential for detecting cardiac arrhythmias particularly in long-term recordings (usually over a period of 24 h). The Electrocardiogram (ECG) signal is a noninvasive test widely used for reflecting the underlying heart conditions. The results obtained show that the newly proposed approach provides significant accuracy improvements with less expert interaction and faster online retraining compared to state-of-the-art methods. Furthermore, we follow the recommendations of the Association for the Advancement of Medical Instrumentation (AAMI) for class labeling and results presentation. In the experiments, we validate the method on the well-known MIT-BIH arrhythmia database as well as two other databases called INCART, and SVDB, respectively. As ranking criteria, the method relies on the DNN posterior probabilities to associate confidence measures such as entropy and Breaking-Ties (BT) to each test beat in the ECG record under analysis. During the interaction phase, we allow the expert at each iteration to label the most relevant and uncertain ECG beats in the test record, which are then used for updating the DNN weights. After this feature learning phase, we add a softmax regression layer on the top of the resulting hidden representation layer yielding the so-called deep neural network (DNN).
To this end, we learn a suitable feature representation from the raw ECG data in an unsupervised way using stacked denoising autoencoders (SDAEs) with sparsity constraint. In this paper, we propose a novel approach based on deep learning for active classification of electrocardiogram (ECG) signals.