Neural network versus electrocardiographer and conventional computer criteria in diagnosing anterior infarct from the ECG
11 Oct 1992-pp 667-670
TL;DR: In this article, the authors examined the performance of a neural network in an electrocardiogram (ECG) classification task and showed that the neural network showed a higher sensitivity than the conventional criteria, both having a specificity of 97%.
Abstract: The purpose of the present study was to examine the performance of a neural network in an electrocardiogram (ECG) classification task. ECGs recorded from 272 patients with anterior myocardial infarction and 479 subjects without myocardial infarction were studied. Fifteen QRS measurements of the leads V2-V4 were used as inputs to the network. The network was trained using 502 ECGs. Thereafter, a comparison of the network, conventional criteria and a human expert was performed using a test set of 249 ECGs. The neural network showed a higher sensitivity than the conventional criteria, both having a specificity of 97%. The performance of the human expert was the same as that of the neural network. It seems that neural networks could be used to improve the performance of some parts of ECG interpretation programs. >
Citations
More filters
TL;DR: The results show that the proposed MEES approach can successfully detect the MI pathologies and help localize different types of MIs.
Abstract: In this paper, a novel technique on a multiscale energy and eigenspace (MEES) approach is proposed for the detection and localization of myocardial infarction (MI) from multilead electrocardiogram (ECG). Wavelet decomposition of multilead ECG signals grossly segments the clinical components at different subbands. In MI, pathological characteristics such as hypercute T-wave, inversion of T-wave, changes in ST elevation, or pathological Q-wave are seen in ECG signals. This pathological information alters the covariance structures of multiscale multivariate matrices at different scales and the corresponding eigenvalues. The clinically relevant components can be captured by eigenvalues. In this study, multiscale wavelet energies and eigenvalues of multiscale covariance matrices are used as diagnostic features. Support vector machines (SVMs) with both linear and radial basis function (RBF) kernel and K-nearest neighbor are used as classifiers. Datasets, which include healthy control, and various types of MI, such as anterior, anteriolateral, anterioseptal, inferior, inferiolateral, and inferioposterio-lateral, from the PTB diagnostic ECG database are used for evaluation. The results show that the proposed technique can successfully detect the MI pathologies. The MEES approach also helps localize different types of MIs. For MI detection, the accuracy, the sensitivity, and the specificity values are 96%, 93%, and 99% respectively. The localization accuracy is 99.58%, using a multiclass SVM classifier with RBF kernel.
235 citations
TL;DR: This paper proposes a novel method of automated detection and localization of MI by using ECG signal analysis that can aid the physicians and clinicians in accurate and faster location of MIs, and thereby providing adequate time available for the requisite treatment decision.
Abstract: Identification and timely interpretation of changes occurring in the 12 electrocardiogram (ECG) leads is crucial to identify the types of myocardial infarction (MI). However, manual annotation of this complex nonlinear ECG signal is not only cumbersome and time consuming but also inaccurate. Hence, there is a need of computer aided techniques to be applied for the ECG signal analysis process. Going further, there is a need for incorporating this computerized software into the ECG equipment, so as to enable automated detection of MIs in clinics. Therefore, this paper proposes a novel method of automated detection and localization of MI by using ECG signal analysis. In our study, a total of 200 twelve lead ECG subjects (52 normal and 148 with MI) involving 611,405 beats (125,652 normal beats and 485,753 beats of MI ECG) are segmented from the 12 lead ECG signals. Firstly, ECG signal obtained from 12 ECG leads are subjected to discrete wavelet transform (DWT) up to four levels of decomposition. Then, 12 nonlinear features namely, approximate entropy ( E a x ), signal energy (?x), fuzzy entropy ( E f x ), Kolmogorov-Sinai entropy ( E k s x ), permutation entropy ( E p x ), Renyi entropy ( E r x ), Shannon entropy ( E s h x ), Tsallis entropy ( E t s x ), wavelet entropy ( E w x ), fractal dimension ( F D x ), Kolmogorov complexity ( C k x ), and largest Lyapunov exponent ( E L L E x ) are extracted from these DWT coefficients. The extracted features are then ranked based on the t value. Then these features are fed into the k-nearest neighbor (KNN) classifier one by one to get the highest classification performance by using minimum number of features. Our proposed method has achieved the highest average accuracy of 98.80%, sensitivity of 99.45% and specificity of 96.27% in classifying normal and MI ECG (two classes), by using 47 features obtained from lead 11 (V5). We have also obtained the highest average accuracy of 98.74%, sensitivity of 99.55% and specificity of 99.16% in differentiating the 10 types of MI and normal ECG beats (11 class), by using 25 features obtained from lead 9 (V3). In addition, our study results achieved an accuracy of 99.97% in locating inferior posterior infarction by using only lead 9 (V3) ECG signal. Our proposed method can be used as an automated diagnostic tool for (i) the detection of different (10 types of) MI by using 12 lead ECG signal, and also (ii) to locate the MI by analyzing only one lead without the need to analyze other leads. Thus, our proposed algorithm and computerized system software (incorporated into the ECG equipment) can aid the physicians and clinicians in accurate and faster location of MIs, and thereby providing adequate time available for the requisite treatment decision.
181 citations
TL;DR: The proposed K-nearest neighbor (KNN) classifier due to its simplicity and high accuracy over the PTB database can be very helpful in correct diagnosis of myocardial infarction in a practical scenario.
Abstract: This paper presents automatic detection and localization of myocardial infarction (MI) using K-nearest neighbor (KNN) classifier. Time domain features of each beat in the ECG signal such as T wave amplitude, Q wave and ST level deviation, which are indicative of MI, are extracted from 12 leads ECG. Detection of MI aims to classify normal subjects without myocardial infarction and subjects suffering from Myocardial Infarction. For further investigation, Localization of MI is done to specify the region of infarction of the heart. Total 20,160 ECG beats from PTB database available on Physio-bank is used to investigate the performance of extracted features with KNN classifier. In the case of MI detection, sensitivity and specificity of KNN is found to be 99.9% using half of the randomly selected beats as training set and rest of the beats for testing. Moreover, Arif-Fayyaz pruning algorithm is used to prune the data which will reduce the storage requirement and computational cost of search. After pruning, sensitivity and specificity are dropped to 97% and 99.6% respectively but training is reduced by 93%. Myocardial Infarction beats are divided into ten classes based on the location of the infarction along with one class of normal subjects. Sensitivity and Specificity of above 90% is achieved for all eleven classes with overall classification accuracy of 98.8%. Some of the ECG beats are misclassified but interestingly these are misclassified to those classes whose location of infarction is near to the true classes of the ECG beats. Pruning is done on the training set for eleven classes and training set is reduced by 70% and overall classification accuracy of 98.3% is achieved. The proposed method due to its simplicity and high accuracy over the PTB database can be very helpful in correct diagnosis of MI in a practical scenario.
173 citations
TL;DR: A novel algorithm based on a convolutional neural network (CNN) is proposed for myocardial infarction detection via multilead electrocardiogram (ECG) via beat segmentation algorithm utilizing multileads, and fuzzy information granulation is adopted for preprocessing.
Abstract: In this paper, a novel algorithm based on a convolutional neural network (CNN) is proposed for myocardial infarction detection via multilead electrocardiogram (ECG). A beat segmentation algorithm utilizing multilead ECG is designed to obtain multilead beats, and fuzzy information granulation is adopted for preprocessing. Then, the beats are input into our multilead-CNN (ML-CNN), a novel model that includes sub two-dimensional (2-D) convolutional layers and lead asymmetric pooling (LAP) layers. As different leads represent various angles of the same heart, LAP can capture multiscale features of different leads, exploiting the individual characteristics of each lead. In addition, sub 2-D convolution can utilize the holistic characters of all the leads. It uses 1-D kernels shared among the different leads to generate local optimal features. These strategies make the ML-CNN suitable for multilead ECG processing. To evaluate our algorithm, actual ECG datasets from the PTB diagnostic database are used. The sensitivity of our algorithm is 95.40%, the specificity is 97.37%, and the accuracy is 96.00% in the experiments. Targeting lightweight mobile healthcare applications, real-time analyses are performed on both MATLAB and ARM Cortex-A9 platforms. The average processing times for each heartbeat are approximately 17.10 and 26.75 ms, respectively, which indicate that this method has good potential for mobile healthcare applications.
121 citations
TL;DR: A novel Multiple-Feature-Branch Convolutional Neural Network (MFB-CNN) is proposed for automated MI detection and localization using ECG, based on deep learning framework, which can achieve a good performance in MI diagnosis.
108 citations
References
More filters
01 Jan 1988
TL;DR: This chapter contains sections titled: The Problem, The Generalized Delta Rule, Simulation Results, Some Further Generalizations, Conclusion.
Abstract: This chapter contains sections titled: The Problem, The Generalized Delta Rule, Simulation Results, Some Further Generalizations, Conclusion
17,604 citations
TL;DR: It is shown that some but not all computer programs for the interpretation of ECGs perform almost as well as cardiologists in identifying seven major cardiac disorders.
Abstract: Background. Computer programs for the interpretation of electrocardiograms (ECGs) are now widely used. However, a systematic assessment of various computer programs for the interpretation of ECGs has not been performed. Methods. We undertook a large international study to compare the performance of nine electrocardiographic computer programs with that of eight cardiologists in interpreting ECGs in 1220 clinically validated cases of various cardiac disorders. ECGs from the following groups were included in the sample: control patients (n = 382); patients with left ventricular hypertrophy (n = 183), right ventricular hypertrophy (n = 55), or biventricular hypertrophy (n = 53); patients with anterior myocardial infarction (n = 170), inferior myocardial infarction (n = 273), or combined myocardial infarction (n = 73); and patients with combined infarction and hypertrophy (n = 31). The interpretations of the computer programs and the cardiologists were compared with the clinical diagnoses made independently of the ECGs, and the computer interpretations were compared with those of the cardiologists. Results. The percentage of ECGs correctly classified by the computer programs (median, 91.3 percent) was lower than that for the cardiologists (median, 96.0 percent; P < 0.01). The median sensitivity of the computer programs was also significantly lower than that of the cardiologists in diagnosing left ventricular hypertrophy (56.6 percent vs. 63.9 percent, P < 0.02), right ventricular hypertrophy (31.8 percent vs. 46.6 percent, P < 0.01), anterior myocardial infarction (77.1 percent vs. 84.9 percent, P < 0.001), and inferior myocardial infarction (58.8 percent vs. 71.7 percent, P < 0.0001). The median total accuracy level (the percentage of correct classifications) was 6.6 percent lower for the computer programs (69.7 percent) than for the cardiologists (76.3 percent; P < 0.001). However, the performance of the best programs nearly matched that of the most accurate cardiologists. Conclusions. Our study shows that some but not all computer programs for the interpretation of ECGs perform almost as well as cardiologists in identifying seven major cardiac disorders.
516 citations
TL;DR: An artificial neural network trained to identify myocardial infarction in adult patients presenting to an emergency department may be a valuable aid to the clinical diagnosis of myocardia infarctions; however, this possibility must be confirmed through prospective testing on a larger patient sample.
Abstract: ▪Objective:To validate prospectively the use of an artificial neural network to identify myocardial infarction in patients presenting to an emergency department with anterior chest pain. ▪...
461 citations
TL;DR: Neural networks can be used for classification of ST-T segments if carefully incorporated into a conventional ECG interpretation program and may well be of value for automatedECG interpretation in the near future.
58 citations
23 Sep 1990
TL;DR: The performance of the neural network approach in the diagnostic classification of 12-lead electrocardiograms (ECG) is investigated and shows a comparable behavior with the two statistical methods.
Abstract: The performance of the neural network approach in the diagnostic classification of 12-lead electrocardiograms (ECG) is investigated. For this study a validated ECG database established at the University of Leuven is used. Previous results obtained from the same database to derive two classifiers based on statistical models (linear discriminant analysis and logistic discriminant analysis) are taken as reference points in the evaluation. A simple neural network architecture is chosen: the feed-forward structure with the use of the back-propagation algorithm. Sensitivity, specificity, total and partial accuracy are the indices used for the assessment of the performance. The results show a comparable behavior with the two statistical methods. >
28 citations