An intelligent framework for the classification of the 12-lead ECG.

An intelligent framework has been proposed to classify an unknown 12-Lead electrocardiogram into one of a possible number of mutually exclusive and combined diagnostic classes. The framework segregates the classification problem into a number of bi-dimensional classification problems, requiring individual bi-group classifiers for each individual diagnostic class. The bi-group classifiers were generated employing Neural Networks (NN), combined with a combination framework containing an Evidential Reasoning framework to accommodate for any conflicting situations between the bi-group classifiers. A number of different feature selection techniques were investigated with the aim of generating the most appropriate input vector for the bi-group classifiers. It was found that by reducing the original input feature vector, the generalisation ability of the classifiers, when exposed to unseen data, was enhanced and subsequently this reduced the computational requirements of the network itself. The entire framework was compared with a conventional approach to NN classification and a rule based classification approach. The framework attained a significantly higher level of classification in comparison with the other methods; 80.0% compared with 66.7% for the rule based technique and 68.00% for the conventional neural approach.

Computerised electrocardiology employing bi-group neural networks.

A configuration of bi-group neural networks (BGNN) is proposed combined with an evidential reasoning framework to interpret 12-lead electrocardiograms for three mutually exclusive classes. A number of pre-processing feature selection techniques were investigated prior to application of the input feature vector to each individual BGNN. The network outputs were discounted within a belief interval of 1 based on their performance on test data prior to combination. It was found that the application of the feature selection techniques enhanced the individual performance of the BGNN, and subsequently enhanced the overall performance. The proposed framework was compared with conventional classification techniques of multi-output neural networks and linear multiple regression. The framework attained a higher level of classification in comparison with the other methods; 70.4% compared with 66.7% for both multi-output neural and statistical techniques.

Body surface ECG potential maps in acute myocardial infarction.

An algorithm for the early detection of acute myocardial infarction (MI) using body surface electrocardiographic potential mapping has been developed. The mapping system consists of a 64-hydrogel electrode harness applied rapidly to the anterior chest, from which electrocardiographic signals are stored on a memory card and processed by computer. At each of the 64 points, QRS and ST-T isointegrals and 10 other features of the QRST segment are measured. Using these measurements, new variables are derived that express the shape of the three-dimensional geometric surface of the map. The isointegrals, features, and shape variables are used in a variety of techniques to discriminate between MI and control subjects. Maps were recorded from 69 patients at initial presentation of chest pain suggestive of acute MI and from 80 healthy control subjects. Using a multiple logistic regression technique, 14 variables were identified that correctly classified 79 of the 80 control subjects (specificity, 98.8%) and 65 of the 69 MI patients (sensitivity, 94.2%). The algorithm based on these 14 variables was applied prospectively to maps recorded on a further 48 control subjects and 59 patients with acute MI. Of the MI patients, 31 had inferior, 13 inferoposterior, 10 anterior, 2 posterior, 1 lateral, 1 inferior with right bundle branch block, and 1 anterior non Q wave MI. The algorithm correctly classified all 48 control subjects (specificity, 100%) and 57 of the 59 MI patients (sensitivity, 96.6%). Marked differences in the three-dimensional geometric map surfaces between the control subjects and MI patients were demonstrated. Variables derived from these surfaces form the basis of an algorithm with a high sensitivity and specificity for the automated detection of acute MI. The design of adaptive algorithms and their application to patients with chest pain and atypical electrocardiographic changes, particularly ST depression, may lead to the earlier detection of MI and greater numbers of patients receiving thrombolytic therapy.

Portable cardiac mapping assessment of acute ischaemic injury.

A portable cardiac mapping system is used to improve the accuracy of diagnosis of acute ischaemic injury outside hospital. Patients presenting chest pain suggestive of myocardial infarction (MI) were mapped by attendant medical personnel operating from a mobile coronary unit. These first MI maps were compared against average normal maps using QRS and ST-T isointegral values. Discriminant function analysis performed on the parameters achieved a sensitivity of 90% and a specificity of 96%.

Discriminant function analysis of body surface potential maps in acute myocardial infarction.

Using a newly developed 64-electrode portable mapping device, QRS and ST-T isointegral maps were compared in 194 control subjects and 101 patients. One hundred ninety-four control subjects (mean age, 48 years; 120 men) with no history of cardiac disease were selected randomly and mapped. One hundred one patients (mean age, 62 years; 77 men) were mapped at presentation of chest pain suggestive of first myocardial infarction (MI); all patients had classic 12-lead electrocardiographic findings--46 with anterior and 55 with inferior MI. The diagnosis was confirmed in all cases by a significant rise in serial cardiac enzymes. The mean delay between onset of chest pain to map recording was 163 minutes. Of the 101 patients, 78 were first mapped outside the hospital. Applying discriminant function analysis to the isointegral measurements made on the control subjects and on the first map of MI patients achieved a correct classification of 97% of the control subjects (189 of 194) and 72% of the anterior (33 of 46) and 76% of the inferior (42 of 55) MI groups. This preliminary study suggests that discriminant function analysis, based on isointegral maps, not only provides a method of separating control subjects from MI patients but that it can also differentiate between types of infarct. Further studies are required to improve the predictive values of discriminant function and to extend the methodology to assess both the site and size of MI.