Search

Serial ECG management and control of the comparison process should be dynamically driven by patient data. The functional architecture requirements are specified for a computerized ECG processing system integrating serial analysis. The concepts of reference investigations and temporal events are first explained. A conceptual reference model is proposed which stratifies the serial analysis process into a few generic strategies which will be instantiated into several types of serial comparison processing tasks as a function of the clinical evolution of the patient. This model is an extension of the OEDIPE core data model developed and implemented during the AIM #2026 project.

A new method to assess the relative difficulty of a medical diagnosis and the efficacy of a diagnostic procedure is described. The cases within a population are classified by N independent interpreters, i.e., medical experts or computer programs. The percentages of correctly diagnosed cases by N (unanimity), N-1, N-2, ..., zero interpreters, respectively, are calculated. These N + 1 percentages are presented in a histogram. A cumulative plot is derived from the "unanimity" value. The ratio "unanimity value"/"zero value" is proposed as an index of diagnostic performance. An application to the CSE database is presented which is composed of 1,220 electrocardiographic tracings interpreted by 13 cardiologists and 15 computer programs. The cumulative plots, the performance index and Log (index) values, established for eight different diagnostic statements, clearly show the relative degree of diagnostic difficulty for each of them and the particular efficacy of each approach, i.e., electrocardiography versus vectorcardiography, and computerized versus human interpretation.

The prerequisite for future development and assessment of serial electrocardiographic analysis programs is the availability of easy to use database management systems for storage and retrieval of electrocardiographic data and results and efficient means for electronic interchange of digital electrocardiographic data. The basic aim of project 2026 of the European program for advanced informatics in medicine is to encourage "open European data interchange and processing for electrocardiography" (OEDIPE) and to provide objective decision support techniques in the field of quantitative electrocardiology. The final objective is to develop demonstration systems for the follow-up evaluation of (1) selected populations with heart disease, (2) integrating serial analysis, (3) decision support, (4) open databases, and (5) communication protocols, which will foreshadow large-scale monitoring of ambulatory or critically ill patients at home, during their transportation in an emergency vehicle, or during their stay in a hospital. The first-year objectives and achievements of the OEDIPE project are briefly described. Emphasis is put on the management of serial electrocardiographic records and data and on the control strategies for the management of the serial analysis process.

Performance requirements and national, as well as international, standards for computer-assisted electrocardiography are behind today's technology. Upgrading is urgently needed from a legal point of view, as well as from a technologic point of view. In Europe a new directive on medical devices requires that instruments with a measurement function have to provide the appropriate measurement accuracy and that manufacturers must specify its measurement accuracy. This study addresses the problem. A description for a new testing system for computer-assisted electrocardiography is given. Also, new tools, for example, the Conformance Testing Services database, and minimum performance requirements for testing computer-assisted electrocardiographs are described and discussed.

This study illustrates the use of the neural network approach in the problem of diagnostic classification of resting 12-lead electrocardiograms. A large electrocardiographic library (the CORDA database established at the University of Leuven, Belgium) has been utilized in this study, whose classification is validated by electrocardiographic-independent clinical data. In particular, a subset of 3,253 electrocardiographic signals with single diseases has been selected. Seven diagnostic classes have been considered: normal, left, right, and biventricular hypertrophy, and anterior, inferior, and combined myocardial infarction. The basic architecture used is a feed-forward neural network and the backpropagation algorithm for the training phase. Sensitivity, specificity, total accuracy, and partial accuracy are the indices used for testing and comparing the results with classical methodologies. In order to validate this approach, the accuracy of two statistical models (linear discriminant analysis and logistic discriminant analysis) tuned on the same dataset have been taken as the reference point. Several nets have been trained, either adjusting some components of the architecture of the networks, considering subsets and clusters of the original learning set, or combining different neural networks. The results have confirmed the potentiality and good performance of the connectionist approach when compared with classical methodologies.

In the SCP-ECG project (AIM #1015) of the pilot AIM Programme a standard communications protocol was developed for computerized electrocardiography, consisting of standards for the interchange, encoding and storage of digital ECG data. Part of this work is continued under the main AIM Programme (January 1992- December 1994) as AIM Project #2026, entitled "Open European Data Interchange and Processing for Computerized Electrocardiography (OEDIPE)" and was taken up by the CEN/TC 251 Committee as a priority item, for implementation into an official European Standard in Health Care Informatics. The objectives and implementation of this Standard are described in the present paper.

In the international project "Common Standards for Quantitative Electrocardiography" (CSE), diagnostic results of different computer programs for the interpretation of the electrocardiogram (ECG) and of the vectorcardiogram (VCG) were combined, and it was shown that the "combined program" performs better than each program separately. Because the program MEANS (Modular ECG Analysis System) comprises 2 different classification programs--one for the ECG, the other for the VCG--this allowed investigation of whether the combined interpretations would yield a better diagnostic result than either one separately. This approach requires that a VCG always be recorded in addition to the ECG. To circumvent this complication, the VCG was reconstructed from the simultaneously recorded ECG leads. This reconstructed VCG was then interpreted by the VCG classification program, whereupon the diagnostic interpretations of the ECG and the reconstructed VCG were combined. For the validation, the CSE database of documented ECGs and VCGs (n = 1,220) was used. The combination of the ECG and VCG interpretations yielded a better diagnostic result than each interpretation program separately (total accuracy 74.2% (ECG + VCG) vs 69.8% (ECG) and 70.2% (VCG), p less than 0.001 in both cases). The results for the reconstructed VCG (total accuracy 70.5%) are comparable to those for the ECG and the VCG (p greater than 0.10 in both cases). The performance of the combined interpretations of ECG and reconstructed VCG (total accuracy 73.6%) is approximately the same as that of the combined ECG and VCG (p greater than 0.10). Thus, the performance of an ECG computer program can be improved by incorporating both ECG and VCG classificatory knowledge, using only the ECG itself.

Information and decision theory were applied to assess the interpretation results of 15 computer programs (9 electrocardiogram and 6 vectorcardiogram) and 9 cardiologists; 8 of whom analyzed the electrocardiogram and 5 the vectorcardiogram, using a database of 1,220 clinically validated cases. The study demonstrates that information content and utility indices, by providing a comprehensive view of diagnostic performance, can enhance the insight given by the more classic statistical performance measures.

In an international collaborative project aimed at producing common standards for quantitative electrocardiography, a standard communications protocol (SCP-ECG) was developed for computerized electrocardiography. The protocol consisted of standards for the interchange, encoding, and storage of digital ECG data. The work was performed in three distinct, but closely related, work-packages and in close collaboration with representatives from 13 manufacturers of computerized electrocardiographs from all over the world. The objectives and results of SCP-ECG are briefly described in this paper.

The objective of this study is to assess the performance of patient-specific segment-specific (PSSS) synthesis in QRST complexes using CAVIAR, a new method of the serial comparison for electrocardiograms and vectorcardiograms. A collection of 250 multi-lead recordings from the Common Standards for Quantitative Electrocardiography (CSE) diagnostic pilot study is employed. QRS and ST-T segments are independently synthesized using the PSSS algorithm so that the mean-squared error between the original and estimated waveforms is minimized. CAVIAR compares the recorded and synthesized QRS and ST-T segments and calculates the mean-quadratic deviation as a measure of error. The results of this study indicate that estimated QRS complexes are good representatives of their recorded counterparts, and the integrity of the spatial information is maintained by the PSSS synthesis process. Analysis of the ST-T segments suggests that the deviations between recorded and synthesized waveforms are considerably greater than those associated with the QRS complexes. The poorer performance of the ST-T segments is attributed to magnitude normalization of the spatial loops, low-voltage passages, and noise interference. Using the mean-quadratic deviation and CAVIAR as methods of performance assessment, this study indicates that the PSSS-synthesis algorithm accurately maintains the signal information within the 12-lead electrocardiogram.

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 of the cardiologists (median, 96.0 percent; P less than 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 less than 0.02), right ventricular hypertrophy (31.8 percent vs. 46.6 percent, P less than 0.01), anterior myocardial infarction (77.1 percent vs. 84.9 percent, P less than 0.001), and inferior myocardial infarction (58.8 percent vs. 71.7 percent, P less than 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 less than 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.

The clinical value of intracoronary electrography for the detection of myocardia ischaemia was assessed during coronary angioplasty and compared to a standard technique of surface ECG monitoring. In 73 patients undergoing single lesion angioplasty, an intracoronary electrogram and four representative surface ECG leads were obtained. During angioplasty of the left anterior descending artery leads, I, V3, V5, V6 were recorded. For the circumflex artery leads I, aVL, aVF, V6, and for the right coronary artery leads II, III, aVF, V6 were monitored. Eight patients were excluded due to transient intraventricular conduction disturbances during balloon inflation; 65 patients remained for further analysis. Out of a total of 154 balloon inflations (35 in the circumflex, 71 in the left anterior descending and 48 in the right coronary artery), the percentage that produced a greater than or equal to 1 mm ST segment elevation, the time to the appearance of a greater than or equal to 1 mm ST segment elevation and the maximal ST segment elevation were recorded. During inflations in the circumflex artery, the respective values of these three parameters were 20%, 22.6 +/- 11.5 s and 0.37 +/- 0.80 mm in V6, the most sensitive surface lead, versus 70% (P less than 0.001), 14.4 +/- 9.6 s (P less than 0.01) and 5.82 +/- 6.35 mm (P less than 0.0001) on the intracoronary electrogram.(ABSTRACT TRUNCATED AT 250 WORDS)

The value of the Selvester and Cardiac Infarction Injury Scores was assessed in a randomized, placebo-controlled study of intravenous recombinant tissue plasminogen activator (rt-PA) in 721 patients with acute myocardial infarction. Electrocardiograms (ECG) obtained at admission, 6 h and 10 to 22 days after the start of therapy were analysed. Patients with prior myocardial infarction or QRS duration greater than or equal to 120 ms were excluded, leaving 322 in the rt-PA group and 333 in the placebo group. Cumulative 72-h release of alpha-hydroxybutyrate dehydrogenase (HBDH) and global ejection fraction derived from angiography and nuclear scintigraphy were used as independent measures of infarct size. Predischarge results demonstrated a net benefit of rt-PA therapy, with the Selvester Score 11% lower (P less than 0.01) and the Cardiac Infarction Injury Score 5.4% lower (P = NS) in the rt-PA than the control group. Total enzyme release was reduced by 19.2% (P less than 0.001) in the rt-PA group. In patients with inferior infarction, neither enzyme release (r = 0.30 to 0.40) nor ejection fraction (r = 0.22 to 0.31) correlated well with the ECG indices of infarct size. In anterior infarction, the correlations were better, especially between the Selvester Score and enzyme release (r = 0.40 to 0.48) as well as ejection fraction (r = -0.48 to -0.67). It is concluded that ECG scoring systems, especially the Selvester Score, although imperfect are useful to assess thrombolytic therapy in clinical trials. However, their value for the management and assessment of thrombolytic therapy in individual patients is still limited.

To determine the ability of initial ST segment elevation and depression to predict infarct size limitation by thrombolytic therapy, data were analyzed in 721 patients with acute myocardial infarction who were admitted to a randomized, placebo-controlled study of intravenous recombinant tissue-type plasminogen activator. Patients with QRS duration of 120 msec or more or with previous history of myocardial infarction were excluded, leaving 322 in the treatment and 333 in the placebo group. Cumulative 72-hour release of alpha-hydroxybutyrate dehydrogenase and global ejection fraction as well as left ventricular wall motion derived from angiography were used as independent measures of infarct size. Electrocardiograms obtained at admission, 6 hours after start of therapy, and before discharge were analyzed. All ST measurements were made by hand at the J point and 60 msec after the J point. Patients with high ST segment elevation at admission (i.e., sum of ST elevation at 60 msec after the J point was 20 mm or more) had significantly larger infarction and higher hospital mortality when compared with those with lower (less than 20 mm) ST elevation. Reciprocal ST segment depression also showed a linear relation with infarct size and mortality, independent from ST elevation, both in anterior and inferior myocardial infarction. The sum of deviations measured at the J point and 60 msec after the J point differed significantly, especially in anterior myocardial infarction at admission (mean, 16 +/- 9 versus 23 +/- 11 mm). The prognostic value of one measurement was not, however, superior over the other. Treatment with recombinant tissue-type plasminogen activator was most effective in those with large ST deviations at admission, but patients with anterior infarction and smaller ST shifts also appeared to benefit from therapy. Results in individual patients were variable, and the overall correlation of initial ST shifts with enzymatic infarct size was rather low. In conclusion, the present study shows that the magnitude of initial ST elevation and also of reciprocal ST depression in the admission electrocardiogram is valuable for the management and assessment of thrombolytic therapy in patients with acute myocardial infarction.

A new method is presented, different from usual methods, for the discussion of the results of computer ECG measurement programs, based on a new graphical evaluation method. The proposed "scatter-graphs" both highlight the main program results and facilitate the comparison between various wave-recognition algorithms. They allow the distinction between the reliability of an algorithm, which is its capacity to provide a maximum of measurements with a minimum amount of errors, and its precision, i.e., the standard deviation of the differences between its point estimates and the reference. The method proves to be a powerful tool for the discussion of individual, as well as the median program results in CSE Measurement Study. It allows to highlight limitations induced in performance assessment by the variability of the reference itself, and to state that the median program is at least as close to the "Gold Standard" as the median of the referees.

Computer processing of electrocardiograms (ECGs) has over the last 15 years increased rapidly. Still, there are at present no standards for computer ECG interpretation. Different techniques are used not only for measurement and interpretation, but also for transmission and storage of data. In order to fill these gaps, a large international project, sponsored by the European Commission, was launched in 1980 to develop "Common Standards for Quantitative Electrocardiography (CSE)". The main objective of the first CSE study was to reduce the wide variation in wave measurements currently obtained by ECG computer programs. The second study was started in 1985 and aimed at the assessment and improvement of diagnostic classification of ECG interpretation programs. To this end reference libraries of well documented ECGs have been developed and comprehensive reviewing schemes devised for the visual and computer analysis of ECGs. This task was performed by a board of cardiologists in a Delphi review process, and by 9 VCG and 10 standard 12-lead programs developed by university research groups and by industry. A third action was started in June 1989 to harmonize acquisition, encoding, interchange and storing of digital ECG data. The action thus performed have become internationally recognized milestones for the standardization of quantitative electrocardiography.

The field of automated ECG analysis was one of the earliest topics in Medical Informatics and may be regarded as a model both for computer-assisted medical diagnosis and for evaluating medical diagnostic programs. The CSE project has set reference standards of two kinds: In a broad sense, a standard how to perform a comprehensive evaluation study, in a narrow sense, standards as specific references for evaluating computer ECG programs. The evaluation methodology used within the CSE project is described as a basis for presentation of results which are published elsewhere in this issue.

In an international project investigators from 25 institutes are trying to establish a common reference library and evaluation methods for testing the diagnostic performance of various ECG computer programs and of cardiologists, based on ECG-independent clinical information. A first set of 500 validated ECGs was collected and analyzed by fifteen different computer programs and nine cardiologists, seven of who analysed the ECG and five the VCG. A coding scheme was used to map individual diagnostic statements onto a common set. Combined program and referee results were obtained by weighted averaging. Preliminary results indicate that the classification accuracy of several programs can still be improved. However, it was also apparent that the results of the best 12-lead ECG computer programs proved to be almost as accurate as the best of seven cardiologists in classifying seven main disease categories, i.e., normal, left, right and biventricular hypertrophy, anterior, inferior and combined myocardial infarction. Evaluation of rhythm statements and conduction disturbances was not included in the study. The data collection is still being pursued in order to reach over 1,000 cases. In this way a common diagnostic database is being established for comparative testing of diagnostic computer programs. This should lead to consumer protection and improve the accuracy and reliability of computerized electrocardiography.

An important feature of an ECG analysis program is its ability to provide reliable measurements under various operating conditions, e.g., on noise-free and noisy ECGs. Therefore, within the European cooperative project "Common Standards For Quantitative Electrocardiography" (CSE), the accuracy and stability of ECG measurements obtained by several computer programs has been compared. To investigate the stability of measurements two sets of 10 ECGs with and without seven different high- and low-frequency types of noise--altogether 160 electrocardiograms and 160 vectorcardiograms--have been analyzed by eight electrocardiographic and five vectorcardiographic computer programs. The stability of measurement was tested with respect to results obtained for the noise-free recordings. In a previous paper, the influence of noise on wave boundary recognition has been reported. In the present paper, the effect of noise on amplitude measurements and on problems of waveform definitions within the QRS complex are described. The results indicate that programs analyzing an averaged beat exhibit less variability than programs which measure every complex or a selected beat. Comparability and stability of measurements could be improved if a standardized procedure for amplitude references were to be introduced. In addition, the stability of QRS waveform labelling could be improved if waveforms' minimum amplitude and duration were to be validated against the noise level which itself should be determined by a standardized procedure.

In the international cooperative project entitled "Common Standards for Quantitative Electrocardiography" (CSE) systematic noise tests have been performed in order to compare measurement results of electrocardiographic computer programs under degraded operational conditions and to develop recommendations for preprocessing and measurement strategies. The influence of seven different high- and low-frequency noise types on the recognition of P, QRS, and T wave onsets and offsets was investigated. The analysis was performed on 160 electrocardiograms derived from two sets of 10 cases each, by eight electrocardiographic and six vectorcardiographic computer programs. The stability and precision of these programs were tested with respect to the results obtained (1) in the noise-free recordings and (2) by a group of five cardiologists who have analyzed the recordings previously in a Delphi reviewing process. Increasing levels of high-frequency noise shifted the onsets and offsets of most programs outward. Programs analyzing an averaged beat showed significantly less variability than programs which measure every complex or a selected beat. On the basis of the findings of the present study, a measurement strategy based on selective averaging is recommended for diagnostic ECG computer programs. However, averaging should be performed only if proper alignment and precise waveform comparison have been performed beforehand in order to exclude dissimilar complexes.

The authors describe an extension of McNemar's test that can be used to compare diagnostic performance when multiple statements are obtained from computer analysis or visual interpretation of the ECG. If two or more diagnostic statements are made, by definition only one can be correct for the cases in the CSE pilot database, which were selected for single, clinically well documented, abnormalities. If one statement stood out from others as being made with the highest degree of certainty, that was accepted as the single interpretation, right or wrong. However, when two or more statements were made with the same degree of certainty and only one statement was correct, then in the previous application of McNemar's test that interpretation was given credit for being correct. In the extension of the method presented in this article, such an interpretation is given partial credit for the one correct statement and partial discredit for any incorrect statement, thereby reporting the results more properly in the sensitivity and specificity statistics for the different diagnostic categories.

As a measure for performance of ECG computer programs, systematic noise tests have been performed within the European project Common Standards For Quantitative Electrocardiography. Ten original and ten artificial VCG and ECG records (the artificial records were constructed from one beat of the original ones) without and with added noise were processed by eight ECG programs and by five VCG programs. Three lowpass filtered high frequency noise levels, line frequency noise and three baseline noise levels were added to the original records. The processing results for the noise-free and for the noisy records have been analyzed in the CSE coordinating center. Results presented in this study refer to the amplitude differences obtained after comparison of results from noise-free and noisy ECG records. Increasing high frequency noise results in overestimation of peak amplitudes and increased variability of measurements.

An analysis was made of clinical and electrocardiographic prognostic determinants of multiple vessel disease in 100 men, aged under 45 years, who survived a myocardial infarction. All patients underwent selective coronary arteriography within 1 year after sustaining a myocardial infarction. Multivessel disease was present in 64 patients; 33 patients had single vessel disease and 3 had either normal coronary arteries or minimal lesions. Exercise stress testing, electrocardiographic location of the infarction, total serum cholesterol and clinical features including body build, arterial blood pressure, smoking habits, family history of coronary artery disease and the presence of angina pectoris either before or after the acute event proved to be poor predictors of multiple vessel disease. Only 74 percent of the patients were correctly classified by a discriminant function analysis. Thus, for prognostic reasons, coronary arteriography seems warranted in young patients after acute myocardial infarction, even in the absence of residual angina or multiple risk factors.

The development of a data base is described which can be used as common reference for ECG computer programs analyzing 12 or 15 simultaneously recorded leads. This data base is an extension of the 3-lead CSE reference library, described previously.

The use of computers for the interpretation of the resting electrocardiogram (ECG) is critically reviewed. After a short description of the motives for the application of computers in this field, the processes that are sequentially executed in an ECG data processing system are critically analyzed, i.e., the acquisition and transmission of signals; the wave recognition and measurement phase; the parameter extraction, rhythm analysis, and classification phases; and, finally, the reporting of results as well as serial comparison. Emphasis is given on the evaluation and standardization of measurement and interpretation results of various ECG computer programs, based on the experience of the author as project leader of the International Cooperative Project "Common Standards for Quantitative Electrocardiography" (CSE). An extensive list of references is provided.

As a result of an international cooperative project entitled "Common Standards for Quantitative Electrocardiography" (CSE), an ECG reference data base has been established with the aim of standardizing computer-derived ECG measurements. The objective of the project is to reduce the wide variation in wave measurements currently obtained by ECG analysis programs. A library of 250 ECGs with selective ECG abnormalities was established and a comprehensive reviewing scheme was devised for the visual determination of the onsets and offsets of P, QRS, and T. This task was performed by a board of cardiologists on highly amplified, selected complexes from the library. A subset was examined in order to study beat-to-beat and intraobserver variability. By using a modified Delphi approach, individual outlying point estimates were eliminated in four successive rounds. In this way final referee estimates were obtained which proved to be highly reproducible and precise. A reference library has thereby been developed which allows testing of the performance of ECG measurement programs and is a useful instrument in establishing recommendations for more precise measurement rules and definitions.

The performance of logistic (LOG) and linear discriminant analysis (LDA) has been studied, both for the conventional 12-lead electrocardiogram (ECG) and the orthogonal Frank 3-lead electrocardiogram (VCG), using a large validated data base. Classification rules were derived form a learning set (N = 2446) and applied to a test set (N = 820) to differentiate between normal, left, right and biventricular hypertrophy, anterior, inferior and combined myocardial infarction (MI). Total accuracy of LOG, assuming no normal distribution and using population proportions as prior probabilities, was up to 3% higher than that of LDA, depending on the number of variables used. The 12- and 3-lead LOG and LDA formulas resulted in very similar accuracy rates, i.e., between 67 and 70% for the seven-group and between 77 and 84% for the five-group analysis. LDA posterior probabilities were systematically more extreme than LOG ones. Correct classification of normals' specificity by LDA was 5 to 9% higher, but sensitivity for different groups was 1.5 to 10% lower than by LOG, with sample size proportions as priors. Specificity could be improved by changing the priors at the cost of lower sensitivity and vice versa, both for the LDA and LOG models. Classification results at 95% specificity were only slightly different, except for anterior MI where LOG scored 6% better. Other measures of performance demonstrated that the LDA model was overconfident and that the LOG model fitted better the real class membership of the patients. In conclusion, logistic ECG and VCG models improve the total accuracy of classification by about 1 to 3% when compared to LDA. More importantly, reliability of classification represents the improvement we want to emphasize. These methods may enhance the diagnostic utility of the ECG and VCG in routine practice.

Controversy exists over the classification ability of the standard 12-lead electrocardiogram (EGG) and the vectorcardiogram (VCG). In this study the diagnostic information content and classification performance of the ECG and VCG were examined using multivariate statistical techniques and a large validated data base of 3,266 cases. Logistic classification models were developed to differentiate between 7 diagnostic entities: normal (n = 538), left (n = 557), right (n = 323) and biventricular (n = 437) hypertrophy, and anterior (n = 390), inferior (n = 657) and combined (n = 364) myocardial infarction. The models were obtained from a learning sample (n = 2,446) using an optimal set of computer derived ECG and VCG measurements. They were subsequently applied to a test sample (n = 820). In the learning sample, the discrimination models resulted in a total correct classification rate of 69.6% for the ECG and 69.4% for the VCG. The total accuracy rate was slightly lower in the test set: 66.3% for the ECG and 67.1% for the VCG. The combined use of the best ECG and VCG variables did not increase total diagnostic accuracy. When cases with biventricular hypertrophy and combined infarction were deleted, accuracy rates of more than 80% were achieved for both lead systems. Differences in the classification rates for the subgroups were not statistically significant. Thus, the conventional 12-lead ECG is as good as the VCG for the differential diagnosis of 7 main entities, provided identical procedures are used in the design of the classifiers.(ABSTRACT TRUNCATED AT 250 WORDS)

In an international project investigators from 21 institutes are trying to establish a common reference library and evaluation methods for testing the diagnostic performance of various ECG computer programs using ECG independent clinical information. Preliminary results indicate that the classification accuracy of different programs varies widely.

In an effort to standardize and evaluate the performance of electrocardiographic computer measurement programs, a 15 lead reference library has been developed based on simultaneously recorded standard 12 lead and orthogonal XYZ lead data. A set of 250 electrocardiograms (ECGs) with selected abnormalities was analyzed by a group of five referee cardiologists and 11 different 12 lead and 6 XYZ computer programs. Attention was focused on the exact determination of the onsets and offsets of P, QRS and T waves. The referees performed their task on highly amplified, selected complexes from the library in a two round process. Median results of the referees coincided best with the median derived from all programs. An analysis of stability proved that the combined program median was a robust reference. However, some individual program results were widely divergent. Paired t tests demonstrated earlier onset for P and QRS (p less than 0.001), as well as later offset for P and T waves in the median 12 lead than in the XYZ results. Significant differences also existed among results obtained by programs analyzing all standard ECG leads at one time, the so-called multilead programs, and those obtained by the conventional standard three lead analysis programs. As a consequence, the derived P, PR, QRS and QT interval measurements varied quite widely among the various programs. Significant differences were also observed among measurements of Q, R and S duration. Some programs showed Q waves that were on the average 6 ms (p less than 0.001) longer than those of others. This may significantly influence diagnostic performance.(ABSTRACT TRUNCATED AT 250 WORDS)

We present a new graphical evaluation method aimed at displaying the main characteristics of computer ECG measurement program results. The proposed graphs provide information on the number of rejected measurements and gross program errors, and on the precision of the results, i.e. the scatter of the differences between fiducial point estimates given by the programs and the reference. 'Scatter-graphs' concerning P, QRS and T wave boundary recognition from the programs participating in the European Concerted Action on 'Common Standards for Quantitative Electrocardiography (C.S.E.)' are analysed and discussed (10 standard 12-lead ECG programs and 9 Vectorcardiographic (VCG) programs from 10 European, 5 American and 1 Japanese Centers). The method proved to be a powerful tool for a direct and global qualitative comparison of the efficiency of the algorithms elaborated by the program developers, and allowed some of the strategies adopted in these programs to be highlighted.

A large international co-operative project, sponsored by the European Commission, was launched in 1980 aimed at developing common standards for quantitative electrocardiography. The first and main objective of the project was to reduce the wide variation in wave measurements currently obtained by electrocardiographic computer programs. To this end a reference library was developed and a comprehensive reviewing scheme was devised for the visual determination of the onsets and offsets of P, QRS and T. This task was performed by a board of cardiologists on highly amplified recordings, in an interactive four round Delphi-type analysis. The reference library, so obtained, has become an internationally recognized yardstick for the evaluation and improvement of ECG measurement programs. It has been used to test the performance of 9 VCG and 10 standard 12-lead programs. The library proved to be a useful instrument in the establishment of recommendations for more precise measurement rules and definitions. Records with added noise and multi-lead ECGs were subsequently analysed to meet specific objectives. The project was expanded in 1984 towards testing and improvement of diagnostic criteria and classification programs.

In an effort to standardize terminology and criteria for clinical electrocardiography, and as a follow-up of its work on definitions of terms related to cardiac rhythm, an Ad Hoc Working Group established by the World Health Organization and the International Society and Federation of Cardiology reviewed criteria for the diagnosis of conduction disturbances and pre-excitation. Recommendations resulting from these discussions are summarized for the diagnosis of complete and incomplete right and left bundle branch block, left anterior and left posterior fascicular block, nonspecific intraventricular block, Wolff-Parkinson-White syndrome and related pre-excitation patterns. Criteria for intraatrial conduction disturbances are also briefly reviewed. The criteria are described in clinical terms. A concise description of the criteria using formal Boolean logic is given in the Appendix. For the incorporation into computer electrocardiographic analysis programs, the limits of some interval measurements may need to be adjusted.

To allow an exchange of measurements and criteria between different electrocardiographic (ECG) computer programs, an international cooperative project has been initiated aimed at standardization of computer-derived ECG measurements. To this end an ECG reference library of 250 ECGs with selected abnormalities was established and a comprehensive reviewing scheme was devised for the visual determination of the onsets and offsets of P, QRS, and T waves. This task was performed by a group of cardiologists on highly amplified, selected complexes from the library of ECGs. With use of a modified Delphi approach, individual outlying point estimates were eliminated in four successive rounds. In this way final referee estimates were obtained that proved to be highly reproducible and precise. This reference data base was used to study measurement results obtained with nine vectorcardiographic and 10 standard 12-lead ECG analysis programs. The medians of program determinations of P, QRS, and T wave onsets and offsets were close to the final referee estimates. However, an important variability could be demonstrated between measurements from individual programs and mean differences from the referee estimates amounted to 10 msec for QRS for certain programs. In addition, the variances of all programs with respect to the referee point estimates were variable. Some programs proved to be more accurate and stable when the data from high- vs low-noise recordings were analyzed. Average Q wave durations calculated from ECGs for which programs agreed on the presence of a Q or QS wave differed by more than 8 msec in several program-to-program comparisons. Such differences may have important consequences with respect to diagnostic performance. Various factors that might explain these differences have been determined. The present study demonstrates that to allow an exchange of results and diagnostic criteria between different ECG computer programs, definitions, minimum wave requirements, and measurement procedures urgently need to be standardized.

Reproducibility of the AVA program (version 3.5) was tested (1) by analyzing the same analog tracing 10 times in 150 patients, and (2) by studying the influence of 12 different sets of prior probabilities (PRP), using the digital data from a total of 2718 patients. The same QRS-T diagnosis plus or minus 8 percentage points was found in patients with high posterior probabilities. Variability for the P wave diagnosis was twice as high as for QRS-T results. Failure to select the appropriate set of PrP markedly influenced reproducibility. This was especially true for right ventricular hypertrophy and pulmonary emphysema. 'Mispriorizing' in a subgroup of 199 patients with well-documented myocardial infarction lowered the total number of correct ECG diagnoses on the average between 4.8 and 35%. These results are inherent to Bayes' theorem where PrP are used to achieve minimum assignment error in populations, but where Prp errors may result in wide classification discrepancies. Without errors in PrP reproducibility by the VA program was found to be very high.

In order to test the diagnostic performance of various ECG computer programs a reference library of ECGs is being established and evaluation methods are being developed in an international co-operative project. A pilot study was undertaken in which 250 validated electrocardiograms (ECG) and vectorcardiograms (VCG) comprising seven diagnostic groups i.e., normal, left, right and bi-ventricular hypertrophy, anterior, inferior and combined infarction have been analysed independently by 11 different computer programs as well as by six cardiologists. A coding scheme was applied to assign individual diagnostic statements to a common set and to obtain combined program and cardiologist interpretation results. Preliminary results indicate that the accuracy of classification by different programs varies widely. Total accuracy varied between 57.2% and 75.8% (median 69.4%). The cardiologists had a higher accuracy (median 74.3%) than the majority of programs, at least when using the standard ECG. As it is considered premature to stress individual program results, in view of the current sample size, the enhancement in diagnostic accuracy obtained by combining interpretation results is highlighted. Indeed combined cardiologist and program results demonstrated the highest accuracy i.e., respectively 78.7% and 76.1%, higher than the result of any individual reader or program. The combined result of the five most accurate programs was 78.4%, that of the six least accurate was 71.5%, which is again higher than the respective individual components. These findings demonstrate that the combination of expert knowledge of computer programs can, similar to panel review and group analysis in clinical practice, enhance diagnostic accuracy.