Performance of the automated complete Selvester QRS scoring system in normal subjects and patients with single and multiple myocardial infarctions.

The automated version of the complete Selvester QRS scoring system for estimation of myocardial infarct size was evaluated in 1,344 normal subjects, 706 patients with a single myocardial infarction (366 with inferior infarction, 277 with anterior infarction and 63 with posterolateral infarction) and 131 patients with combined inferior and anterior infarction. The presence and location were determined by angiographic and ventriculographic criteria. The performance of the overall 32-point system, each of the 19 criteria and the 13 criteria sets and each of the 35 criteria within the 13 sets was examined. The mean point scores were 1.7 for normal subjects, 3.7 for posterolateral infarction, 4.1 for inferior infarction, 6.3 for anterior infarction and 6.9 for multiple infarcts. A score greater than 4 yielded a sensitivity of 67% for anterior infarction, 41% for inferior infarction, 32% for posterolateral infarction and 72% for multiple infarcts. However, 7 of 32 criteria failed to achieve 95% specificity and 10 of 35 criteria in criteria sets had a sensitivity that was even lower than their false positive rate. The automated Selvester QRS scoring system currently has limitations that are attributable to development of the original system, which used manual scoring techniques and established criteria limits from middle-aged men. Future automated analysis should use gender- and age-dependent criteria limits.

Agreement between artificial neural networks and experienced electrocardiographer on electrocardiographic diagnosis of healed myocardial infarction.

OBJECTIVES: The purpose of this study was to compare the diagnoses of healed myocardial infarction made from the 12-lead electrocardiogram (ECG) by artificial neural networks and an experienced electrocardiographer. BACKGROUND: Artificial neural networks have proved of value in pattern recognition tasks. Studies of their utility in ECG interpretation have shown performance exceeding that of conventional ECG interpretation programs. The latter present verbal statements, often with an indication of the likelihood for a certain diagnosis, such as "possible left ventricular hypertrophy." A neural network presents its output as a numeric value between 0 and 1; however, these values can be interpreted as Bayesian probabilities. METHODS: The study was based on 351 healthy volunteers and 1,313 patients with a history of chest pain who had undergone diagnostic cardiac catheterization. A 12-lead ECG was recorded in each subject. An expert electrocardiographer classified the ECGs in five different groups by estimating the probability of anterior myocardial infarction. Artificial neural networks were trained and tested to diagnose anterior myocardial infarction. The network outputs were divided into five groups by using the output values and four thresholds between 0 and 1. RESULTS: The neural networks diagnosed healed anterior myocardial infarctions at high levels of sensitivity and specificity. The network outputs were transformed to verbal statements, and the agreement between these probability estimates and those of an expert electrocardiographer was high. CONCLUSIONS: Artificial neural networks can be of value in automated interpretation of ECGs in the near future.

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.

Computer analysis of the orthogonal electrocardiogram and vectorcardiogram in 257 patients with aortic valve disease.

In most electrocardiographic studies of left ventricular hypertrophy the hypertrophy has had different causes. This study examined the performance of the Frank orthogonal electrocardiogram in 257 patients with aortic valve disease, 90 with predominant aortic stenosis and the remaining 167 with predominant aortic insufficiency. Optimal measurements suggested for routine electrocardiographic diagnosis of left ventricular overload resulted in recognition of 64 percent of cases of aortic stenosis and 74 percent of cases of aortic insufficiency with a 6 percent false positive rate. Multivariate analysis techniques improved performance: 82 percent of the aortic stenosis records and 78 percent of the aortic insufficiency records were correctly recognized as showing left ventricular overload; the false positive rate was 5 percent. Although many electrocardiographic measurements were significantly correlated with the aortic valve gradient, correlations were not high enough for use in predicting the severity of the aortic stenosis. Comparison of electrocardiographic variables in aortic stenosis and aortic insufficiency demonstrated a wide overlap between groups, and the electrocardiographic changes of aortic stenosis could not be differentiated from those of aortic insufficiency.

Specificity and sensitivity of QRS criteria for diagnosis of single and multiple myocardial infarcts.

A subset of 3 criteria from the complete Selvester scoring system has been proposed earlier for electrocardiographic screening of healed myocardial infarcts. This subset yielded 95% specificity and high sensitivity for single anterior and inferior infarcts. In the present study, an automated version of these criteria was applied to 1,344 electrocardiograms from normal subjects (473 normal subjects as determined by cardiac catheterization and 871 apparently normal subjects by history and physical examination), to 706 from subjects with single myocardial infarction, and to 131 from subjects with combined anterior and inferior myocardial infarcts. Of the single infarcts, 366 had inferior, 277 anterior and 63 posterolateral locations. Presence and location of infarcts were judged from left ventriculograms and coronary angiograms. Overall specificity was only 86%, whereas overall sensitivity for the infarct population was 77%. Specificity was lower in men than in women; it was also lower in older than in younger subjects. One of the screening criteria (R greater than or equal to 40 ms in V1) may possibly be eliminated to augment specificity; this can be done with only minor loss of sensitivity. Differences in wave form measurements between the manual and computer methods account for a large part of the deterioration of specificity in this study compared with previously published results. Computer application of the screening criteria requires altered criteria limits in comparison with those used in manual application. Probably sex- and age-dependent criteria limits should be used.

Effects of atrial fibrillation and atrial flutter on the signal-averaged electrocardiogram.

In conclusion, atrial flutter can create significant errors in the automated time-domain analysis of the SAECG that are only apparent when the study is repeated in sinus rhythm, thus lowering the predictive accuracy of the technique in patients with atrial flutter. Atrial fibrillation rarely creates problems with time-domain analysis of the SAECG. These findings suggest that, unless the performance of a specific signal-averaging device has been evaluated in patients with atrial flutter and found to have acceptable error rates, patients with atrial flutter should not have SAECGs performed for postinfarction risk assessment.

An ischemic index from the electrocardiogram to select patients with low left ventricular ejection fraction for coronary artery bypass grafting.

Patients who have a decreased left ventricular (LV) ejection fraction (EF) may be denied coronary by-pass grafting (CABG) because it is assumed that improvement in function is unlikely. If the low LVEF were due to myocardial necrosis, this assumption would be valid. If the dysfunction were due to both necrosis and ischemia, however, then improvement may be possible with CABG. A method capable of identifying such patients would be useful. In this study, an "ischemic index" was determined for 37 patients based on the difference between the presurgical LVEF estimated from the standard 12-lead electrocardiogram by the Selvester QRS score (indicating the extent of dysfunction due to necrosis) and the presurgical LVEF measured from resting multigated radionuclide angiography (indicating dysfunction due to both necrosis and ischemia). It was hypothesized that a high ischemia index, that is, a large discrepancy between estimated and measured LVEF, would be associated with an improved post-surgical measured LVEF. The results showed that patients with an ischemic index of less than or equal to 0 had a mean decrease in LVEF of 8% (p = 0.02) and those with an index between 0 and 10 had no mean change. Patients with an index greater than or equal to 11, however, had a mean increase of 5% (p = 0.02), suggesting that depressed LVEF may improve following CABG among patients whose low function is due primarily to reversible ischemia as indicated by a high ischemic index.

Development and validation of an automated method of the Selvester QRS scoring system for myocardial infarct size.

The Selvester QRS score for estimation of myocardial infarct (MI) size from the standard 12-lead electrocardiogram (ECG) has not yet achieved wide recognition as a valuable tool in the routine assessment of the MI patient, primarily because of the practical limitations inherent to manual application. This study examined the ECGs of 438 patients (105 normal subjects, 161 with "possible" MI and 172 with "definite" MI based on data from cardiac catheterization) to develop software for an automated method of the Selvester system in attempts to overcome the manual constraints. After a comprehensive validation process involving extensive interactions between the manual scorer and the software developer, an automated method of the Selvester system was generated that had a high correlation with manual application (r = 0.94) and was superior regarding time, training, reader bias, reproducibility and precision of measurement. These results indicate that an automated version of the Selvester QRS scoring system would resolve many of the limitations of manual application and would provide a reliable, technically accurate estimate of MI size that could be incorporated into ECG diagnostic programs and used in standard digital ECG machines.

Establishment of signal-averaged electrocardiographic criteria with Frank XYZ leads and spectral filter used alone and in combination with ejection fraction to predict inducible ventricular tachycardia in coronary artery disease.

Signal-averaged electrocardiographic criteria are reported for corrected Frank XYZ leads and a spectral filter. The new criteria were used alone and in combination with ejection fraction to predict inducibility of ventricular tachycardia (VT) at electrophysiologic testing. Signal-averaged electrocardiographic criteria were developed in 87 control subjects and validated in 182 patients (aged 63 +/- 10 years) with coronary artery disease and QRS duration less than 118 ms. Patients underwent electrophysiologic testing in which up to 3 extra-stimuli were used during 2 paced drives from 2 right ventricular sites. A positive finding was monomorphic VT lasting 30 seconds or needing intervention. An ejection fraction less than 40% was considered abnormal. Signal-averaged electrocardiographic variables that best characterized control subjects and separated patients with and without inducible VT were filtered QRS duration less than 120 ms, low-amplitude signal duration less than 38 ms and root-mean-square voltage greater than 20 muv. With these criteria, signal-averaged electrocardiographic and ejection fraction sensitivities were 87 and 45%, respectively, and specificities were 65 and 77%, respectively. Combining signal-averaged electrocardiography with ejection fraction improved the predictive accuracy. In conclusion, diagnostic criteria for signal-averaged electrocardiography with use of Frank XYZ leads and a spectral filter produced results similar to those reported for use of bipolar XYZ leads and a Butterworth filter. Signal-averaged electrocardiography was a better predictor of VT than was ejection fraction.

Evaluation of changes in standard electrocardiographic QRS waveforms recorded from activity-compatible proximal limb lead positions.

Proximal limb lead positions are currently used for activity-compatible electrocardiographic monitoring of myocardial ischemia. Two previously described systems for alternate limb lead placement were studied in patients with and without QRS evidence of healed anterior or inferior myocardial infarction. An innovative method was used to simultaneously record 6 standard and 6 modified limb leads, and 3 standard and 3 modified precordial leads on a standard digital electrocardiograph. Both alternate lead placement systems showed rightward frontal plane axis shift and diminished Q-wave durations in lead aVF compared with those of their simultaneous standard controls. Furthermore, potential differences between the standard distal limb lead sites and 5 more proximal sites were explored along each limb. Differences along the left arm were accentuated relative to those along the right arm owing to differences in proximity of the arms to the myocardium. Along the lower limb, and anterior site showed less deviation from standard than did a more lateral site. It is imperative that recordings from alternate sites be labeled accordingly so that their output cannot be confused with that obtained from standard sites.

Simultaneous ST-segment measurements using standard and monitoring-compatible torso limb lead placements at rest and during coronary occlusion.

Electrocardiographic recordings used to assess ST-segment deviation are performed using both standard and torso limb lead positions, where bony prominences give more artifact-free signal. Whereas significant QRS artifact can be introduced by such changes in lead location, the impact on ST-segment measurements has never been assessed. Digital electrocardiographic recordings were performed in 29 patients throughout elective angioplasty balloon inflation in the left anterior descending (n = 12), right coronary (n = 14), and circumflex (n = 3) arteries. In all cases, unipolar leads V1, V4, and V6 were affixed to the torso lead positions, allowing reconstruction of simultaneously acquired standard and modified 9-lead electrocardiograms (ECGs). ST levels in the 26 patients who had ST deviation during angioplasty were compared at both baseline and peak ischemia of up to 1,046 microV in the anterior, and 551 microV in the inferior leads. Differences in recorded ST levels for modified versus standard lead locations were all < 100 microV, even at peak ischemia. Although ST-segment elevation in the inferior leads appeared to show slightly more pronounced differences between lead sets than did anterior elevation, all differences were < 100 microV. Thus, measurement of ST-segment levels appears unlikely to be importantly affected by the intermixture of ECGs recorded with standard lead positions and ECGs recorded with monitoring-compatible lead positions on the torso. Recalibration of ST-segment measurements may be necessary for meticulous quantification of ischemia, infarct size, or other measurements that might be affected by variations < 100 microV.

Importance of accounting for the variability of electrocardiographic data among diagnostically similar patients with inferior wall healed myocardial infarction.

The authors describe a method to account for patient-to-patient variability in electrocardiographic data. The method yielded criteria for healed inferior myocardial infarction with diagnostic performances better than those of traditional electrocardiographic parameters.

Left ventricular to right atrial defect and tricuspid insufficiency secondary to nonpenetrating cardiac trauma.

We present a report of a 28-year-old male patient who developed a left ventricular to right atrial communication associated with tricuspid insufficiency secondary to nonpenetrating cardiac trauma. The patient's heart block was clarified by serial observation of electrocardiograms and the use of HIS bundle electrocardiography. The patient's fistula was surgically repaired, and 20 months postinjury he is asymptomatic and pacemaker dependent. Echocardiography may aid in the early diagnosis of a left to right shunt and/or tricuspid insufficiency in the setting of possible cardiac trauma.

Electrode position effects on Frank lead electrocardiograms.

Frank lead electrocardiograms were recorded from 149 normal and abnormal adult males using four different electrode placements. All chest electrodes were placed at: (1) the fourth intercostal space level, (2) the fifth intercostal space level, (3) the fourth intercostal space level with V4 substituted for C, and (4) the fifth intercostal space level with V4 substituted for C. Differences in mean values of many commonly used amplitudes and orientations were not statistically significant among the four recording methods, but amplitude differences for individual subjects were often large and difficult to predict. When V4 is substituted for C, as commonly done in some laboratories, Rx decreased and Rz increased by more than 10 per cent in about 40 per cent of the cases. In about 70 per cent of the cases, Rx and Rz changed significantly when electrode level was shifted from the fifth to the fourth intercostal space. For these 70 per cent, it does not appear possible to accurately predict increase or decrease of Rx, Rz, or QRSm. Analysis programs which depend on individual amplitude measurements are likely to be significantly affected by electrode placement. It is suggested that criteria for analysis programs developed using a specified version of the Frank system should ideally be applied only to electrocardiograms recorded in the same manner.

Computer-assisted ECG interpretation for confirmation of premature atrial complexes.

UNLABELLED: Current programs for computerized ECG analysis are not interactive. We developed custom software for computer-assisted ECG interpretation that functioned interactively with an observer directing the computerized process. The software was first used for recognition of PACs superimposed on ST-T waveforms. The interactive process included 6 steps. 1) The computer displayed the 12-lead ECG and the user selected the most frequent QRS-T waveforms for averaging. 2) The computer generated and displayed the averaged QRS-T waveform. 3) The user selected waveforms suspected to have PACs superimposed on the ST-T segments. 4) The computer generated and displayed the difference waveform by subtracting the average waveform from the suspect waveforms. 5) The user recognized and p-waves in the difference waveform and marked the onset and offset by positioning the cursor and clicking. 6) The computer then measured p-wave amplitudes, durations, and areas and displayed the recognized p-wave. The program was developed using digital data from 2 ECGs and tested on 39 ECGs with suspected PACs and 26 control ECGs. RESULTS: The software and user interaction recognized 79 PACs in the suspect group. Control ECGs were analyzed using 3 complexes with the same leads and onsets as the test group. Of the 79 PACs found in the suspect group, mean values included an area under the curve of 4.0 +/- 3.2 microV-s for the test group versus 0.4 +/- 0.4 microV-s for the control (P < 0.001) and peak-to-trough voltage amplitudes of 104 +/- 66 microV versus 15 +/- 7 microV for the control (P < 0.001). The average time of onset of the premature complexes was 282 +/- 120 msec, and their duration was 100 +/- 28 msec. CONCLUSION: Custom software combined the superior human pattern recognition with the digital signal processing of the computer. This enhanced recognition of ectopic atrial activity.

Use of the signal-averaged ECG to predict maximum ventricular tachycardia rate.

Rapid ventricular tachycardia is poorly tolerated, and estimation of maximum ventricular tachycardia rate prior to programmed ventricular stimulation is difficult. A method to estimate maximum ventricular tachycardia rate using late potential duration from the signal-averaged ECG and ventricular functional refractory period is described. Late potentials recorded in patients with ventricular tachycardia may represent delayed conduction through arrhythmogenic ventricular myocardium. This delay may be rate limiting in determining the minimum cycle length of reentrant ventricular tachycardia originating from these areas. Using the ratio of ventricular activation time (VAT), which equals QRS plus late potential duration, to unfiltered QRS duration (QRS) as estimate of this delay, the following relationship is proposed: Minimum ventricular tachycardia cycle length = (FRP 400 - 12.5 ms) (VAT/QRS). Twenty patients with late potentials who had sustained, monomorphic ventricular tachycardia at programmed stimulation were evaluated. Predicted cycle lengths ranged from 326 to 214 ms. Predicted and observed cycle lengths were significantly correlated (r = 0.91, SEE = 11.9 ms, p less than 0.0005), with predicted and observed cycle lengths differing by less than 3.5%. Predicted cycle lengths were more accurate than cycle lengths estimated using FRP alone (p less than 0.01). Accurate prediction of minimum ventricular tachycardia cycle length using this relationship suggests that late potential duration is proportional to the conduction delay occurring in arrhythmogenic ventricular myocardium.

Decision rules for the ECG diagnosis of inferior myocardial infarction.

ECG measurements from 341 patients with inferior myocardial infarction (IMI) and 327 normal subjects were used to develop and test decision rules for the ECG diagnosis of IMI. Recursive partitioning provided a simple decision rule with 75% sensitivity and 97% specificity, using Q amplitude and Q duration in a VF, Q duration in III, and T-wave axis in the frontal plane as decision variables. Dropping T-wave axis from the decision rule led to a 10% decrease in sensitivity. Multiple logistic regression provided sensitivities and specificities which were similar to those for recursive partitioning. Both methods outperformed traditional noncontour criteria for IMI.

Increased sensitivity for the diagnosis of healed myocardial infarction using vectorial information in the 12-lead ECG.

The aim of this study was to use the vectorial information in the conventional 12-lead electrocardiographic (ECG) recording and to investigate whether this information, in combination with well-known ECG criteria, could increase the diagnostic performance for healed anterior or inferior myocardial infarction. A total of 1,458 subjects were included in the study; 272 patients with anterior myocardial infarction, 356 patients with inferior myocardial infarction, and 830 subjects classified as normal. New 12-lead vectorcardiographic criteria for anterior and inferior myocardial infarction were developed and used in combination with well-known ECG criteria. The combined criteria showed a sensitivity of 80.0% and 72.4% for the diagnosis of anterior and inferior myocardial infarction, respectively. The corresponding sensitivities for the conventional ECG criteria were significantly lower. In conclusion, the addition of vectorial parameters into ECG interpretation programs could be of value.