There are MANY papers related to automated ECG analysis. A few recent abstracts of relevant papers appear below.
Ventricular electrograms and intramyocardial pressure (IMP) signals were recorded in 11 dogs during sinus rhythm, during paced ventricular tachycardia (VT), and at the onset of and during ventricular fibrillation (VF). The autocorrelation function (ACF) and the probability density function (PDF) of short episodes of the electrograms were analyzed off-line on a digital computer. Peak-to-peak values of the IMP were calculated during sinus rhythm and during VT/VF. An algorithm was developed to recognize VT and VF using the ACF, the PDF and the IMP as input signals. Results show that in case of sinus rhythm, all detection methods are reliable (recognition rate of 100%). In case of VT with hemodynamic impairment the ACF is slightly better (66.6%) than the PDF (44.4%). The onset of VF is sensed in 81.8% of all episodes with the ACF and in 63.6% with the PDF. During VF, this improves, respectively, to 92.3 and 69.2%. In all previous cases, the IMP signal was 100% reliable. It is concluded that in this arrhythmia model, the sensitivity of an automatic VT/VF detection system was increased by combining ECG processing with analysis of a hemodynamic parameter.
A class of algorithms has been developed which detects QRS complexes in the electrocardiogram (ECG). The algorithms employ nonlinear transforms derived from multiplication of backward differences (MOBD). The algorithms are evaluated with the American Heart Association ECG database, and comparisons are made with the algorithms reported by Okada (1979) and by Hamilton and Tompkins (1986). The MOBD algorithms provide a good performance tradeoff between accuracy and response time, making this type of algorithm desirable for real-time microprocessor-based implementation.
The proposed work suggest an automatic test setup for the measurement of various parameters of ECG signals based on a recursive segmentation technique. This recursive segmentation technique is a time domain analysis and is preferred over frequency domain analysis, This relative inefficiency of the frequency domain analysis for electrophysiological signals is to a large extent due to the fluctuations of the biorhythms. The variation in frequency and amplitude of the signal induce an averaging effect that tends to mask out certain elementary phenomena. This difficulty is overcome in recursive segmentation techniques by assigning an adaptive time base and measuring the various parameters with respect to this time reference.
Presents an algorithm for automatically locating the waveform boundaries (the onsets and ends of P, and QRS, and T waves) in multilead ECG signals (the 12 standard leads and the orthogonal XYZ leads). Given these locations, features of clinical importance (such as the RR interval, the PQ interval, the QRS duration, the ST segment, and the QT interval) may be measured readily. First, a multilead and RS detector locates each beat, using a differentiated and low-pass filtered ECG signal as input. Next, the waveform boundaries are located in each lead. The leads in which the detected electrical activity is of longest duration are used for the final determination of the waveform boundaries. The performance of the authors' algorithm has been evaluated using the CSE multilead measurement database. In comparison with other algorithms tested by the CSE, their algorithm achieves better agreement with manual measurements of the T-wave end and of interval values, while its measurements of other waveform boundaries are within the range of the algorithm and manual measurements obtained by the CSE.
Since the QT interval changes with time under the influence of heart rate and autonomic input, it needs to be analyzed on a long-term basis. The authors present a method for automatic QT measurement based on linear regression analysis, by which the end of the T wave is determined as the crossing point between the baseline and the tangent to the last branch of the T wave itself. The automatic QT measurement is coupled with the analysis of the heart rate variability when examining 24-h Holter recordings. The focus is on the characteristics of the algorithm for measuring the QT interval and the results of its validation. This procedure proved to be accurate when compared with manual measurements and, to some extent, was little affected by the sampling rate. This feature could substantially reduce the amount of data to be analyzed and quicken the procedure without appreciably affecting the quality of the analysis.
Synthesis of the 12-lead electrocardiogram (ECG) has been investigated recently as a method for improving patient monitoring in situations where three leads are recorded routinely. The authors present a recent study which analyzes ECG-estimation accuracy by comparing twelve diagnostic parameters from the recorded and synthesized QRS and ST-T segments of 240 patients. The results reveal strong correlations between the synthesized and recorded measurements. These correlations are present for some measurements despite statistically-significant absolute differences between the two sets. Spurious deflections were observed at the beginning of the QRS complex in leads V/sub 1/ or V/sub 3/ in a small number of cases. The measurements taken from the Q and R waves in leads V/sub 1/ and V/sub 3/ indicate that these spurious deflections cause erroneous measurements in about 9 to 17% of the patients.
The authors investigate the contribution to the prediction of ventricular tachycardia (VT) of a new time-scale technique suited to transient signal detection: wavelet analysis (WA). Wavelet Transformation (WT) is obtained by expanding the signal on a set of functions resulting from the translation (time) and the dilatation (scale) of a so-called 'analyzing wavelet'. It provides a bidimensional representation of the signal in function of time and scale. Bipolar X, Y, Z signals were acquired using a standard, commercially available signal-averaging system, in groups of 10 patients: myocardial infarction (MI) with VT, MI without VT, and healthy subjects. WT was based on J. Morlet wavelets. Contour maps allowed the localization of short, low energy transient signals, even within the QRS complex. Patients with VT displayed the most disturbed contour plots. It is concluded that WA, combined with an adequate graphic representation, provides an accurate characterization of patients prone to VT. These preliminary results are very promising as regards the potential improvements which may be obtained by searching for an optimal analyzing wavelet.
The combination of two techniques of pattern recognition i.e., cluster analysis and neural networks, is investigated in the specific problem of the diagnostic classification of 12-lead electrocardiograms (ECGs). For this study a previously used database, established at the University of Leuven, has been employed. Sensitivity, specificity, and total and partial accuracy were the indices used for the assessment of the performance. Several neural networks have been obtained by either varying the training set (considering clusters of the original learning set) or adjusting some components of the architecture of the networks. The combination of different neural networks has shown satisfactory performances in the diagnostic classification task.
Detecting arrhythmias from the electrocardiogram (ECG) is of great importance for the continued development of intelligent cardiovascular monitors (ICM). An ICM's main goal is to present to the clinician a 'high-level' analysis of the patient's condition (e.g. the patient is slightly hypovolemic) based upon 'low-level' physiologic signals (e.g blood pressure, heart rate, etc.). The authors report on a parallel implementation of a multi-state Kalman filtering algorithm, within a prototype ICM, to help detect ECG arrhythmias. Preliminary test results show that the parallel, multi-state implementation performed exactly as the original sequential version. Several different rhythm disturbances were correctly identified after 3-5 beats. The authors conclude that the parallel implementation of the multi-state Kalman filter provides a faster and still reliable means of accurately detecting ECG arrhythmias in real-time.
An expert system for comprehensive electrocardiographic diagnosis has been developed. The system provides ECG (electrocardiogram) processing by the HES ECG program, a database with validated ECGs and their accompanying clinical findings, and a knowledge base containing established rule sets for ECG classification (Cardiac Infarction Injury Score, Romhilt Estes Hypertrophy score, the Selvester score system, etc.). For access to the database a group definition language has been developed; and a rule definition language is provided for creation of rule sets and diagnostic classification algorithms. Rule sets may be applied to single cases or groups of cases (batch processing). Reasoning is provided by using the backward chaining inference engine, which generates a structured reasoning path.