Neural networks for classification of ECG ST-T segments

Overview of featurization techniques used in traditional versus emerging deep learning-based algorithms for automated interpretation of the 12-lead ECG

Automated interpretation of the 12-lead ECG has remained an underpinning interest in decades of research that has seen a diversity of computing applications in cardiology. The application of computers in cardiology began in the 1960s with early research focusing on the conversion of analogue ECG signals (voltages) to digital samples. Alongside this, software techniques that automated the extraction of wave measurements and provided basic diagnostic statements, began to emerge. In the years since then there have been many significant milestones which include the widespread commercialisation of 12-lead ECG interpretation software, associated clinical utility and the development of the related regulatory frameworks to promote standardised development. In the past few years, the research community has seen a significant rejuvenation in the development of ECG interpretation programs. This is evident in the research literature where a large number of studies have emerged tackling a variety of automated ECG interpretation problems. This is largely due to two factors. Specifically, the technical advances, both software and hardware, that have facilitated the broad adoption of modern artificial intelligence (AI) techniques, and, the increasing availability of large datasets that support modern AI approaches. In this article we provide a very high-level overview of the operation of and approach to the development of early 12-lead ECG interpretation programs and we contrast this to the approaches that are now seen in emerging AI approaches. Our overview is mainly focused on highlighting differences in how input data are handled prior to generation of the diagnostic statement.

Acquaintance to Artificial Neural Networks and use of artificial intelligence as a diagnostic tool for tuberculosis: A review

Tuberculosis [TB] has afflicted numerous nations in the world. As per a report by the World Health Organization [WHO], an estimated 1.4 million TB deaths in 2015 and an additional 0.4 million deaths resulting from TB disease among people living with HIV, were observed. Most of the TB deaths can be prevented if it is detected at an early stage. The existing processes of diagnosis like blood tests or sputum tests are not only tedious but also take a long time for analysis and cannot differentiate between different drug resistant stages of TB. The need to find newer prompt methods for disease detection has been aided by the latest Artificial Intelligence [AI] tools. Artificial Neural Network [ANN] is one of the important tools that is being used widely in diagnosis and evaluation of medical conditions. This review aims at providing brief introduction to various AI tools that are used in TB detection and gives a detailed description about the utilization of ANN as an efficient diagnostic technique. The paper also provides a critical assessment of ANN and the existing techniques for their diagnosis of TB. Researchers and Practitioners in the field are looking forward to use ANN and other upcoming AI tools such as Fuzzy-logic, genetic algorithms and artificial intelligence simulation as a promising current and future technology tools towards tackling the global menace of Tuberculosis. Latest advancements in the diagnostic field include the combined use of ANN with various other AI tools like the Fuzzy-logic, which has led to an increase in the efficacy and specificity of the diagnostic techniques.

Prediction models in the design of neural network based ECG classifiers: a neural network and genetic programming approach

BACKGROUND

Classification of the electrocardiogram using Neural Networks has become a widely used method in recent years. The efficiency of these classifiers depends upon a number of factors including network training. Unfortunately, there is a shortage of evidence available to enable specific design choices to be made and as a consequence, many designs are made on the basis of trial and error. In this study we develop prediction models to indicate the point at which training should stop for Neural Network based Electrocardiogram classifiers in order to ensure maximum generalisation.

METHODS

Two prediction models have been presented; one based on Neural Networks and the other on Genetic Programming. The inputs to the models were 5 variable training parameters and the output indicated the point at which training should stop. Training and testing of the models was based on the results from 44 previously developed bi-group Neural Network classifiers, discriminating between Anterior Myocardial Infarction and normal patients.

RESULTS

Our results show that both approaches provide close fits to the training data; p = 0.627 and p = 0.304 for the Neural Network and Genetic Programming methods respectively. For unseen data, the Neural Network exhibited no significant differences between actual and predicted outputs (p = 0.306) while the Genetic Programming method showed a marginally significant difference (p = 0.047).

CONCLUSIONS

The approaches provide reverse engineering solutions to the development of Neural Network based Electrocardiogram classifiers. That is given the network design and architecture, an indication can be given as to when training should stop to obtain maximum network generalisation.

Radial basis function neural network approach for the diagnosis of coronary artery disease based on the standard electrocardiogram exercise test

The purpose of the paper is the evaluation of a radial basis function neural network as a tool for computer aided coronary artery disease diagnosis based on the results of the traditional ECG exercise test. The research was performed using 776 data records from an exercise test (297 records from healthy patients and 479 from ill patients) confirmed by coronary arteriography results. Each record described the state of the patient, provided input data for the neural network, included the level and slope of an ST segment of a 12-lead ECG signal made at rest and after effort, heart rate, blood pressure, load during the test, and occurrence of coronary pain, coronary arteriography, correct output pattern for the neural network, and verified the existence (or not) of more than 50% stenosis of the particular coronary vessels. Radial basis function neural networks for coronary artery disease diagnosis were optimised by choosing the type of radial function, the method of training (setting the number of centres and their dimensions), and regularisation. The best network correctly recognised over 97% of cases from a 400-element test set, diagnosing not only the patients' condition (simple 'healthy/unhealthy' diagnosis), but also pointing out individual unhealthy/stenosed vessels.

ECG pattern recognition and classification using non-linear transformations and neural networks: a review

The most widely used signal in clinical practice is the ECG. ECG conveys information regarding the electrical function of the heart, by altering the shape of its constituent waves, namely the P, QRS, and T waves. Thus, the required tasks of ECG processing are the reliable recognition of these waves, and the accurate measurement of clinically important parameters measured from the temporal distribution of the ECG constituent waves. In this paper, we shall review some current trends on ECG pattern recognition. In particular, we shall review non-linear transformations of the ECG, the use of principal component analysis (linear and non-linear), ways to map the transformed data into n-dimensional spaces, and the use of neural networks (NN) based techniques for ECG pattern recognition and classification. The problems we shall deal with are the QRS/PVC recognition and classification, the recognition of ischemic beats and episodes, and the detection of atrial fibrillation. Finally, a generalised approach to the classification problems in n-dimensional spaces will be presented using among others NN, radial basis function networks (RBFN) and non-linear principal component analysis (NLPCA) techniques. The performance measures of the sensitivity and specificity of these algorithms will also be presented using as training and testing data sets from the MIT-BIH and the European ST-T databases.

An adaptive backpropagation neural network for real-time ischemia episodes detection: development and performance analysis using the European ST-T database

A supervised neural network (NN)-based algorithm was used for automated detection of ischemic episodes resulting from ST segment elevation or depression. The performance of the method was measured using the European ST-T database. In particular, the performance was measured in terms of beat-by-beat ischemia detection and in terms of the detection of ischemic episodes. The algorithm used to train the NN was an adaptive backpropagation (BP) algorithm. This algorithm drastically reduces training time (tenfold decrease in our case) when compared to the classical BP algorithm. The recall phase of the NN is then extremely fast, a fact that makes it appropriate for real-time detection of ischemic episodes. The resulting NN is capable of detecting ischemia independent of the lead used. It was found that the average ischemia episode detection sensitivity is 88.62% while the ischemia duration sensitivity is 72.22%. The results show that NN can be used in electrocardiogram (ECG) processing in cases where fast and reliable detection of ischemic episodes is desired as in the case of critical care units (CCU's).

ECG processing techniques based on neural networks and bidirectional associative memories

Two ECG processing techniques are described for the classification of QRSs, PVCs and normal and ischaemic beats. The techniques use neural network (NN) technology in two ways. The first technique, uses nonlinear ECG mapping preprocessing and subsequently for classification uses a shrinking algorithm based on NNs. This technique is applied to the QRS/PVC problem with good result. The second technique is based on the Bidirectional Associative Memory (BAM) NN and is used to distinguish normal from ischaemic beats. In this technique the ECG beat is treated as a digitized image which is then transformed into a bipolar vector suitable for input in the BAM. The results show that this method, if properly calibrated, can result in a fast and reliable ischaemic beat detection algorithm.

Modelling ECG signals with hidden Markov models

In this paper, we have studied the use of continuous probability density function hidden Markov models for the ECG signal analysis problem. Our previous work has focused on syntactic pattern recognition methods in signal processing. Hidden Markov model is basically a non-deterministic probabilistic finite state machine, which can be constructed inductively. It has been widely used in speech recognition and DNA modelling. We have found that hidden Markov models are very suitable for ECG recognition and analysis problems and that they are able to model accurately segmented ECG signals.

Performance of a neural network trained to make third-molar treatment-planning decisions

The authors developed and tested 12 neural networks of different architectures to make lower-third-molar treatment-planning decisions, using a software-based neural network (Neudesk 1.2, Neural Computer Sciences, Southampton, UK). Network training was undertaken using clinical histories from 119 patients (with 238 lower third molars) referred for treatment planning (79 females and 40 males, mean age 25 years) together with output data consisting of actual treatments planned by a senior oral surgeon. Both the input clinical data and the consultant decisions were treated on a tooth-wise basis and were coded to numerical values. Binary data (e.g., present/absent) were coded to 1 and 0, while quantitative data (e.g., age) were scaled to fall between 0 and 1. A network based on the optimal architecture was trained and then interrogated with test data derived from a further 174 patients (119 females and 55 males, mean age 26 years) with 348 lower third molars. Network decisions were dichotomized with a threshold of 0.8. With no knowledge of the network decisions, the senior oral surgeon indicated his preferred treatments. The teeth were then assigned to "gold-standard" categories of indications present or absent based on National Institutes of Health consensus criteria. Against this, the network achieved a sensitivity of 0.78, which was slightly inferior to that of the oral surgeon (0.88), although this difference was not significant, and a specificity of 0.98, compared with 0.99 for the oral surgeon (p = NS). Agreement between the oral surgeon and network decisions was very high (kappa = 0.850). This study demonstrates that it is possible to train a neural network to provide reliable decision support for lower-third-molar treatment planning.

Possibilities of using neural networks for ECG classification

Some characteristics of the neural network approach have been tested and validated for the particular problem of diagnostic classification in the field of computerized electrocardiography. Two different databases have been used for the evaluation process: CORDA, developed by the Medical Informatics Department of the University of Leuven, and ECG-UCL, developed by the Cliniques Universitaires Saint-Luc, Université Catholique de Louvain. Electrocardiographic signals classified on the basis of electrocardiographic independent clinical data, with a single diagnosis and no conduction abnormalities, have been considered. Seven diagnostic classes have been taken into account, including the different locations of ventricular hypertrophy and myocardial infarction. Two architectures of neural networks have been analyzed in detail considering three aspects: the normalization process, pruning techniques, and fuzzy preprocessing by the use of radial basis functions. The comparison of the results obtained with the two databases will be discussed in detail.

Artificial neural networks for decision support in clinical medicine

Connectionist models such as neural networks are alternatives to linear, parametric statistical methods. Neural networks are computer-based pattern recognition methods with loose similarities with the nervous system. Individual variables of the network, usually called 'neurones', can receive inhibitory and excitatory inputs from other neurones. The networks can define relationships among input data that are not apparent when using other approaches, and they can use these relationships to improve accuracy. Thus, neural nets have substantial power to recognize patterns even in complex datasets. Neural network methodology has outperformed classical statistical methods in cases where the input variables are interrelated. Because clinical measurements usually derive from multiple interrelated systems it is evident that neural networks might be more accurate than classical methods in multivariate analysis of clinical data. This paper reviews the use of neural networks in medical decision support. A short introduction to the basics of neural networks is given, and some practical issues in applying the networks are highlighted. The current use of neural networks in image analysis, signal processing and laboratory medicine is reviewed. It is concluded that neural networks have an important role in image analysis and in signal processing. However, further studies are needed to determine the value of neural networks in the analysis of laboratory data.

Artificial neural networks for recognition of electrocardiographic lead reversal

Misplacement of electrodes during the recording of an electrocardiogram (ECG) can cause an incorrect interpretation, misdiagnosis, and subsequent lack of proper treatment. The purpose of this study was twofold: (1) to develop artificial neural networks that yield peak sensitivity for the recognition of right/left arm lead reversal at a very high specificity; and (2) to compare the performances of the networks with those of 2 widely used rule-based interpretation programs. The study was based on 11,009 ECGs recorded in patients at an emergency department using computerized electrocardiographs. Each of the ECGs was used to computationally generate an ECG with right/left arm lead reversal. Neural networks were trained to detect ECGs with right/left arm lead reversal. Different networks and rule-based criteria were used depending on the presence or absence of P waves. The networks and the criteria all showed a very high specificity (99.87% to 100%). The neural networks performed better than the rule-based criteria, both when P waves were present (sensitivity 99.1%) or absent (sensitivity 94.5%). The corresponding sensitivities for the best criteria were 93.9% and 39.3%, respectively. An estimated 300 million ECGs are recorded annually in the world. The majority of these recordings are performed using computerized electrocardiographs, which include algorithms for detection of right/left arm lead reversals. In this study, neural networks performed better than conventional algorithms and the differences in sensitivity could result in 100,000 to 400,000 right/left arm lead reversals being detected by networks but not by conventional interpretation programs.

Evaluation of new self-learning techniques for the generation of criteria for differentiation of wide-QRS tachycardia in supraventricular tachycardia and ventricular tachycardia

This study presents a comparison of three different methods for differentiating between supraventricular and ventricular tachycardias with wide-QRS complex. One set of criteria, derived using classical statistical techniques, was compared with two new self-learning computer techniques: the artificial neural networks and the induction algorithm approach. By analyzing the results obtained in an independent test set, using these new techniques, the criteria defined by the classical method could be improved.

Artificial neural networks for the diagnosis of atrial fibrillation

Different forms of artificial intelligence have been applied to pattern recognition in medicine. Recently, however, a relatively new technique involving software-based neural networks has become more readily available. Deterministic logic is currently applied to rhythm analysis in computer-assisted ECG interpretation methods developed in the University of Glasgow. The aim of the present study is to compare an artificial neural network with deterministic logic for separating sinus rhythm (SR) with supraventricular extrasystoles (SVEs) and/or ventricular extra-systoles (VEs) from atrial fibrillation (AF) at a particular point in the diagnostic logic of the Glasgow Program. A total of 2363 ECGs with 1495 AF and 868 SR + (SVEs and/or VEs) are used for training and testing a variety of neural networks, and the optimum design is selected. Methods for combining the results of the neural-network classification and the deterministic interpretation are also developed. A further 717 ECGs are used to test the selected network. The results show that the use of an artificial neural network can improve the sensitivity of reporting AF from 88.5% using the deterministic approach to 92%, without sacrificing specificity (92.3%).

Artificial neural networks for the electrocardiographic diagnosis of healed myocardial infarction

Artificial neural networks are computer-based expert systems that learn by example, in contrast to the currently used rule-based electrocardiographic interpretation programs. For the purpose of this study, 1,107 electrocardiograms (ECGs) from patients who had undergone cardiac catheterization were used to train and test neural networks for the diagnosis of myocardial infarction. Different combinations of QRS and ST-T measurements were used as input to the neural networks. In a learning process, the networks automatically adjusted their characteristics to correctly diagnose anterior or inferior wall myocardial infarction from the ECG. Two thirds of the ECGs were used in this process. Thereafter, the performance of the networks was studied in a separate test set, using the remaining third of the ECGs. The results from the networks were also compared with that of conventional electrocardiographic criteria. The sensitivity for the diagnosis of anterior myocardial infarction was 81% for the best network and 68% for the conventional criteria (p < 0.01), both having a specificity of 97.5%. The corresponding sensitivities of the network and the criteria for the diagnosis of inferior myocardial infarction were 78% and 65.5% (p < 0.01), respectively, compared at a specificity of 95%. The results indicate that artificial neural networks may be of interest in the attempt to improve computer-based electrocardiographic interpretation programs.

Application of neural networks to the classification of giant cell arteritis

OBJECTIVE

Neural networks are a group of computer-based pattern recognition methods that have recently been applied to clinical diagnosis and classification. In this study, we applied one type of neural network, the backpropagation network, to the diagnostic classification of giant cell arteritis (GCA).

METHODS

The analysis was performed on the 807 cases in the vasculitis database of the American College of Rheumatology. Classification was based on the 8 clinical criteria previously used for classification of this data set: 1) age > or = 50 years, 2) new localized headache, 3) temporal artery tenderness or decrease in temporal artery pulse, 4) polymyalgia rheumatica, 5) abnormal result on artery biopsy, 6) erythrocyte sedimentation rate > or = 50 mm/hour, 7) scalp tenderness or nodules, and 8) claudication of the jaw, of the tongue, or on swallowing. To avoid overtraining, network training was terminated when the generalization error reached a minimum. True cross-validation classification rates were obtained.

RESULTS

Neural networks correctly classified 94.4% of the GCA cases (n = 214) and 91.9% of the other vasculitis cases (n = 593). In comparison, classification trees correctly classified 91.6% of the GCA cases and 93.4% of the other vasculitis cases. Neural nets and classification trees were compared by receiver operating characteristic (ROC) analysis. The ROC curves for the two methods crossed, indicating that the better classification method depended on the choice of decision threshold. At a decision threshold that gave equal costs to percentage increases in false-positive and false-negative results, the methods were not significantly different in their performance (P = 0.45).

CONCLUSION

Neural networks are a potentially useful method for developing diagnostic classification rules from clinical data.

Use of artificial neural networks within deterministic logic for the computer ECG diagnosis of inferior myocardial infarction

An investigation into the use of software-based artificial neural networks for the electrocardiographic (ECG) detection of inferior myocardial infarction was made. A total of 592 clinically validated subjects, including 208 with inferior myocardial infarction, 300 normal subjects, and 84 left ventricular hypertrophy cases, were used in this study. A total of 200 ECGs (100 from patients with inferior myocardial infarction and 100 from normal subjects) were fed to 66 supervised feedforward neural networks for training using a back-propagation algorithm. QRS and ST-T wave measurements were used as the input parameters for the neural networks. The best performing network using QRS measurements only and the best using QRS and ST-T data were selected by assessing a test set of 292 ECGs (108 from patients with inferior myocardial infarction, 84 from patients with left ventricular hypertrophy, and 100 from normal subjects). These two networks were then implanted separately into the deterministic Glasgow program for further study. After the implementation, it was found necessary to include a small inferior Q criterion to improve the specificity of reporting inferior myocardial infarction, thereby producing a small loss of sensitivity as compared with use of the network alone. The use of an artificial neural network within the deterministic logic performed better than either alone in the diagnosis of inferior myocardial infarction, producing a 20% gain in sensitivity with 2% loss in overall specificity compared with the original deterministic logic.

Development of ECG criteria to diagnose the number of narrowed coronary arteries in rest angina using new self-learning techniques

Recently, an evaluation of the value of the resulting electrocardiogram recorded during chest pain for identifying high-risk patients with three-vessel or left main stem coronary artery disease has resulted in the definition of one characteristic pattern: ST-segment depression in leads I, II, and V4-V6 and elevation in lead aVR. This study evaluated the generation of such criteria using two self-learning techniques: neural networks and induction algorithms. In 113 patients, five variables, including the amount of ST elevation, the number of leads with abnormal ST-segments, and this above-mentioned characteristic sign, were correlated with the number of narrowed vessels. All patients were randomly subdivided into a training (n = 63) and test set (n = 50), stratified for both this characteristic sign and for the vessel involved. Using the learning set, the neural network and the induction algorithm were trained separately to identify (1) pure left main stem disease and (2) three-vessel disease and left main stem disease. The neural network was trained for 1,000 runs. The induction algorithm was trained, allowing all variables to be used in any order. The experiments were repeated after adding weight factors to promote the recognition of the more severe cases. Subsequently, the ST elevation in all 12 leads was added to the training and test sets, once with and once without the polarity of the ST deviation. Altogether, 18 different combinations were evaluated. Basically, the neural network and the induction algorithm approach misclassified the same cases in corresponding test combinations.(ABSTRACT TRUNCATED AT 250 WORDS)