- Email: [email protected]
Identification of best electrocardiographic leads for diagnosing anterior and inferior myocardial infarction by statistical analysis of body surface potential maps
NOVEMBER 1, 1986
The
American
Journal
of CARDIOLOGY
®
VOLUME 58 NUMBER 10
CORONARY ARTERY DISEASE
Identification of Best Electrocardiographic Leads for Diagnosing Anterior and Inferior Myocardial Infarction by Statistical Analysis of Body Surface Potential Maps FRED KORNREICH, MD, PhD, TERRENCE J. MONTAGUE, MD, PENTTI M. RAUTAHARJU, MD, PhD, PIERRE BLOCK, MD, PhD, JAMES W. WARREN, MSc, and MILAN B. HORACEK, PhD
In view of the increasing interest in quantifying and modifying the size of myocardial infarction (MI), it is important to look for clinically practical subsets of electrocardiographic leads that allow the earliest and most accurate diagnosis of the presence and electrocardiographic type of MI. A practical approach is described, taking advantage of the increased information content of body surface potential maps over standard electrocardiographic techniques for facilitating clinical use of body surface potential maps for such a purpose. Multivariate analysis was performed on 120-lead electrocardiographic data, simultaneously recorded in 236 normal subjects, 114 patients with anterior MI and 144 patients with inferior MI, using as features instantaneous voltages on time-normalized QRS and ST-T waveforms. Leads and features for optimal separation of normal subjects from, respectively, anterior MI and inferior MI patients were selected. Features measured on leads originating from the upper left precordial area, lower midthoracic region and the
back correctly identified 97% of anterior MI patients, with a specificity of 95 %; in patients with inferior MI, features obtained from leads located in the lower left back, left log, right subclavicular area, upper dorsal region and lower right chest correctly classified 94 % of the group, with specificity kept at 95 %. Most features were measured in early and mid-QRS, although very potent discriminators were found in the late portion of the T wave. Repeatability of the results was investigated by separating the study population in training and testing sets; no deterioration was observed when the discriminant functions computed on the training sets were run on the testing sets. In comparison, at the same level of specificity and with the same number of features (n = 6), the standard 12-lead electrocardiogram correctly diagnosed 89 % of anterior MI and 85 % of inferior MI patients. Thus, diagnosis of anterior and inferior MI can be substantially improved by appropriate selection of electrocardiographic leads and features. (Am J Cardiol 1986;58:863-871)
863
864
OPTIMALLEADS FOR ANTERIOR AND INFERIOR MYOCARDIAL INFARCTION
T h e 12-lead electrocardiogram is one of the most reliable methods of diagnosing the presence and general location of myocardial infarction (MI}. Despite the undisputed merits of the standard electrocardiogram, there is little doubt that body surface potential maps contain diagnostic information not present in conventional lead systems2 -4 The increased information content of body surface potential maps over standard electrocardiographic techniques has shown promising results in assessment of patients with MI~-7; however, to facilitate clinical use of body surface potential maps, data reduction is required. 8-12 In a previous study 13we described a practical approach for the extraction of diagnostic information from body surface potential maps by performing multivariate analysis of 120-lead data simultaneously recorded from 361 persons; a subset of optimal leads and features was found for separation of 184 normal subjects from 177 patients with MI, irrespective of the type of infarction. The present study uses body surface potential maps from 494 subjects to identify optimal leads and features to distinguish 236 normal subjects from 114 patients with anterior MI and 144 with inferior MI. This approach, by identifying specific locations of electrocardiographic leads with improved diagnostic information, may lead to further progress in quantitative assessment of MI.
Methods Study population: Five hundred sixty-three subjects were studied. Group A consisted of 236 normal control subjects. None of these subjects had evidence of heart disease by history, physical examination, 12lead electrocardiogram and, when available, echocardiogram. They were 20 to 55 years old (mean 36). Group B included 327 patients with first MI; all had a typical history of prolonged, ischemic-type cardiac pain and characteristic changes in enzyme levels. In many cases the diagnosis was further substantiated by coronary angiography and ventriculography, echocardiography or nuclear imaging. All patients had at least one 12-lead electrocardiogram recorded in the acute phase of the infarction. Patients were excluded if they had electrocardiographic evidence of complete left or right bundle branch block or maj or nonspecific intraventricular conduction delay {QRS 120 ms or longer). The study population was further classified according to the appearance of the QRS complex. From the Unit for Cardiovascular Research and Engineering and Department of Cardiology, Free University of Brussels, Brussels, Belgium, and the Departments of Physiologyand Biophysics and Medicine, Dalhousie University, and the Victoria General Hospital, Halifax, Nova Scotia, Canada. This study was supported by the National Fund for Scientific Research {FGWO 3.0017.75 and FGWO 3.0095.86}, the Medical Research Council of Canada (PG-30}and the Nova Scotia Heart Foundation. Manuscript received May 5, 1986; revised manuscript received July 10, 1986, accepted July 11, 1986. Address for reprints: Fred Kornreich, MD, PhD, Unit for Cardiovascular Research and Engineering, Free University Brussels, Laarbeeklaan 103, B-1090 Brussels, Belgium.
Patients were classified as having anterior MI if they had Q waves at least 30 ms in duration in leads I, aVL or V~ to V6, or initial R waves of 0.2 mV or less in leads V1 and V2. Patients were classified as having inferior (posterior) MI if abnormal Q waves were present in leads II or aVF or if R waves in V~, V2 were exceedingly tall (R to S ratio 1 or more) or broad (at least 40 ms). Body surface mapping: We used previously described methods of recording, processing and displaying 120 body surface electrocardiographic signals. 14 Briefly, digitized signals were recorded simultaneously from 117 torso and 3 limb leads with Wilson's central terminal as reference potential, at 500 samples/s per channel. Tracing quality was monitored visually during the recording; later, the stored data were processed by performing selective averaging and again carefully inspected and edited. All leads judged invalid were deleted and replaced by interpolation of data from surrounding leads. We time-normalized separately the QRS waveform and the ST-T waveform and represented them by 70 and 180 points, respectively. Patients with MI were studied a mean of 10 months after MI (range I day to 24 months); the interval between the acute event and the recording of body surface potential maps was less than 1 week in 39 patients (15%), between 1 week and 1 month in 30 (12%) and more than 1 month in 189 (73%). Discrimtnant maps: To identify leads and features that were significant in discriminating between 2 groups of maps, a previously described technique was used23 Briefly, averaged 120-lead maps were computed separately for the normal group and each MI group by averaging sample-by-sample, time-normalized electrocardiograms from subjects in each group. Difference maps were then calculated by subtracting corresponding sample values of 120 averaged signals of normal and respectively, anterior MI and inferior MI groups. Discriminant maps were obtained for each 2-group comparison (normal vs anterior MI or normal vs inferior MI} by dividing each instantaneous difference with the corresponding composite standard deviation computed from the population under consideration. The values thus achieved, referred to as discriminant indexes, were strictly proportional to Student t test statistics and provided, in contrast to difference maps, information on the capability for each measurement in each lead, to separate each class of MI patients from this normal group. As noted earlier, 13 inspection of discriminant maps provided clues for identifying leads and features for bigix~p_ differentiation. Feature extraction and statistical analysis: We used instantaneous amplitude measurements obtained by sampling time-normalized QRS and ST-T waveforms at equal intervals; this resulted in 8 sam. ples for QRS and 7 samples for ST-T, i.e., 15 × 120 per subject. For each bigroup classification, programs from the Biomedical Program library were used (BMDPTM)2 ~ The statistical procedure involved 2 steps. First, the best classifiers were selected from the total number of available features by a stepwise selec-
November 1, 1986 THE AMERICAN JOURNAL OF CARDIOLOGY Volume 58
865
tribution of the sites with large discriminant ability depends on whether that ability occurs during depolarization or repolarization. The best leads for QRS cluster around lead 50 in the upper left precordial and subclavicular areas and show peak discriminant indexes in early depolarization {1/8 QRS and mainly 2/8 QRS}. A similar pattern can be observed in the back on the mid-dorsal line {e.g., lead 100} and probably represents, in part, a mirror image. A secondary area of discriminant capability is observed later in the depolarization in the lower thoracic region, under and to the left of the xiphoid appendix. The sites of high ST-T discriminant indexes are noted mainly in the left precordial area, extending to the lower anterior and posterior axillary regions, and in the upper right thoracic and subclavicular areas, extending under the right axilla into the upper dorsal region. The maximal discriminant capability appears at 6/8 ST-T, slightly after the peak of the T wave, as exemplified in leads 69 and aVR. Except for QRS in lead V2 and ST-T in leads V5 and V6, the precordial leads are on the margin of the areas with the largest discriminant indexes. Table I lists the results of the discriminant analysis. The first 6 variables are listed in their order of entry in the stepwise procedure, although practically no further improvement is achieved after 3 steps. Adjusting the specificity to 95% instead of 99%, the sensitivity
tion; then, these selected classifiers were combined into the properly weighted linear discriminant functions. To test the robustness of the discriminant functions, the "jackknife" procedure, available in the BMDP7M package, was also used. With this procedure, each case is eliminated in turn from the computation and subsequently assigned to 1 of the groups formed by the remaining cases. Because of the very large number of variables {1,800 instantaneous measurements per patient}, the program processed 80 input variables at a time and selected the 20 best from each batch, repeating the procedure until it acquired a final set of 20 optimal classifiers. Varying the composition of the batches and their order of entry did not affect the first 6 classifiers selected.
Results Anterior infarction: In 17 of 114 patients {15% }with initial electrocardiographic evidence of isolated anterior MI, the standard 12-lead electrocardiogram no longer revealed depolarization abnormalities at the time of body surface potential map recording. The mean interval between the acute event and the recording of 120-lead data was 8 months {range 1 day to 20 months}. Figure 1 shows the averaged 120-lead maps for the normal and the anterior MI populations. Figure 2 shows the corresponding discriminant map. The dis-
RA
LA
LL
rrrr
.,
-
A_~.. A_~. ~_.._ . . ~ 69
V5
V4 (
!:
,
;
~
.:
."
..~_..~ ,~ .....
lOO
M6 ;
,,
FIGURE 1. Averaged 120-lead map of normal subjects and paUents with anterior myocardial infarction (MI). The top row Is at the level of the sternal notch and the botfom rowat the umbilical level; a wider vertical gap separates the anterior from the posterior portion of the thorax. At each electrode site, the superimposed tracings represent averaged, time-normalized electrocardiographic signals for the normal (dotted line) and the anterior MI (solid line) populations, respectively. The precordlal leads are labeled V1 to V e, the limb leads RA (right arm), LA (left arm) and LL (left leg). Leads 20, 32, 50, 69 and 100 indicate specific leads selected in the discriminant analysis for the separation of normal persons from patients with anterior MI (Table I).
866
OPTIMAL LEADS FOR ANTERIOR AND INFERIOR MYOCARDIAL INFARCTION
TABLE I DifferenUation of 236 Normal Subjects from 114 Patients with Anterior Myocardial Infarction Using Instantaneous Measurements on 120-Lead Data Stepwise Selection of Variable Type Lead Lead
Lead Lead Lead Lead
50 69 32 100 50 20
2/8 618 4/8 1/8 1/8 2/8
Specificity (% )
Sensitivity (%)
95 97 98 98 99 99
89 94 96 95 94 95
QRS ST-T QRS QRS QRS ST-T
with 6 variables reached 97%. The most potent discriminators are the amplitudes measured on leads 50, 69 and 32; these leads are located in areas described as showing high discriminant index values. Figure 3 shows the leads and features selected in the discriminant analysis and listed in Table I. Early QRS measurements are selected in leads 50 and 100; they correspond to large Q waves (lead 50) or absent Q waves (lead 100) observed in the respective regions from w h i c h these leads originate. The discriminant capability observed later in QRS also represents, as in lead 32, loss of anterior forces in the lower middle and left
TABLE II Differentiation of 236 Normal Subjects from 144 Patients with Inferior Myocardial InfarcUon Using Instantaneous Measurements on 120-Lead Data Stepwise Selection of Variable Type
Lead LL or Lead Lead Lead Lead
87 aVF 20 97 65 4
6/8 ST-T 218 QRS 4/8 QRS 2/8 QRS 1/8 ST-T 3/8 QRS
Specificity (%)
Sensitivity (%)
92 93 95 97 98 98
83 84 91 92 93 93
thoracic region. The maximal index for lead 69 peaks in late T and is located in a large are of near-0 or negative voltages on the left anterior chest, extending to the lower left axillary region while the selected variable on lead 20 is an early ST-T measurement, relating to ST-segment elevation in the anterior MI group. The diagnostic performance achieved with multivariate analysis on body surface potential maps was compared with that of the standard 12-lead electrocardiogram, recorded at the same time. Using 6 instantaneous measurements as variables and keeping the
DIll u RA
LA
LL
L,,.,L,,,., L,..,L L L L. L ~'0
L,,,,,L L_ L IL,,,,[.,,,, L 50
V1
V2
L 69
V4
V5
L1 O0
V6
L 32
FIGURE 2. Discrlminant 120-lead map for separation of normal subjects from patients with anterior myocardial Infarction. At each electrode site, Instantaneous dlscrlmlnant Indexes (see text), calculated at 8 equidistant Ume instants for QRS (bold vertical bars) and for ST-T (fine vertical bars) are shown; the amplitude of the dlscrlmlnant Index (DI) equals 1 U when the corresponding mean voltages at 1/8 QRS, 2/8 GRS, and so on, for normal subjects and patients with anterior myocardial Infarction are 1 (pooled) standard deviation apart. Leads are labeled as in Figure 1.
November 1, 1986 THE AMERICAN JOURNAL OF CARDIOLOGY Volume 58
specificity at 95%, 89% of the anterior MIs were correctly classified. Although most of the 12-lead variables selected in the stepwise procedure were early QRS measurements in leads V1, V2 and aVL, the best classifier was 6/8 ST:T in lead Vs. This finding probably accounts for the discrepancy between the diagnostic yield of Q-wave criteria {17 missed anterior MI cases} and variables selected with multivariate statistics {13 missed anterior MI cases}. Inferior infarction: Of the 144 patients with initial 12-lead electrocardiographic criteria of inferior MI, 29 {20%} failed to meet these criteria at the time body surface potential maps were recorded, a mean of 11 months later {range 1 day to 24 months}. Figure 4 shows the averaged 120-lead map for the normal and the inferior MI groups. Figure 5 shows the corresponding discriminant map. In contrast to the anterior MI group, the areas with the largest discriminant indexes are common to QRS and ST-T. These areas are located outside the precordial area. Except for the ST-T wave in leads V~ and Vs, the limb leads have higher discriminatory capability than the precordial leads. The sites where initial QRS measurements {2/8 QRS} are optimal for diagnosing inferior MI are located in the right upper back and the lower right side of the thorax. Later in depolarization {3/8 and 4/8 QRS}, additional sites are observed over the entire inferior border around the thorax as well as high in both subclavicular regions. The discriminant ability of the ST-T wave at 6/8 ST-T is als0 maximal in the same areas {except the lower right and upper left thorax}. High discriminant indexes were observed in the left precordial area {e~g., lead 65} during early ST segment, a pattern different from those in the other leads with high T indexes in late repolarization. Table II is a list of the results of the discriminant analysis. The best 6 classifiers are shown; maximal differentiation is achieved with 5 variables. In contrast to the results listed in Table I, the increase, with each Step in diagnostic yield, is slower and levels off later. Adjusting the specificity at step 6 at 95%, a sensitivity of 94% was observed. All selected variables are measured on leads originating from the sites previously described. Figure 6 shows these leads. The most potent single classifier is found in lead 87 and corresponds approximately to the peak of the T wave. In fact, the entire repolarization in the inferior MI group is characterized by flat or slightly negative T waves over the upper right anterior and dorsal areas and the lower left flank. The early QRS measurements in aVF and lead 4 and in lead 97 indicate the presence of Q waves in the former leads and a reciprocal R wave in the latter. Later QRS voltages reflect loss of electrical forces in the upper right chest and the lower left posterior flank {3/8 and 4/8 QRS in leads 20 and 87}. Here too, multivariate analysis was performed on 12 lead measurements at the time body surface potential maps were obtained. With 6 variables and specificity adjusted at 95%, a sensitivity of 85% was reached {22 missed inferior MI cases}. This compared favorably with the 29 missed inferior MI cases when Q-wave criteria were applied. Again, the best classifier was 6/8 ST-T in lead
867
,/ .',°
/
LEAD
50
LEAD 100
JJZLLLU LEAD
69
LEAD
,32
LEAD
20
FIGURE 3. Best measurements of the linear dlscriminant function computed for separation of the normal group from the anterl0r myocardial Infarction group. The leads are 20, 32, 50, 69 and 100 (same as Figure 1), from which optimal classifiers are extracted. The normal (dotted line) and anterior MI (solid line) waveforms are averaged, time-normalized and superimposed. The bottom curve in each box represents the discriminent index as it varies in time; the marks on the abscissa indicate the 8 equidistant time instants on QRS and ST-T; discriminant indexes corresponding to these time instants were shown In Figure 2 for all 120 leads of the map. Arrows point to the features selected in the stepwise discriminant procedure.
868
OPTIMAL LEADS FOR ANTERIOR AND INFERIOR MYOCARDIAL INFARCTION
DII, followed by aVF {2/8 QRS and 4/8 QRS), aVL {7/8 QRS} and ST-T measurements in leads V2 {1/8 ST-T} and V6 (6/8 ST}. Repeatability of results: The discriminant programs {BMDP7M) were run without and with the jackknife option, the latter enabling each subject of the design set to be classified in turn with the discriminant function derived from all the remaining subjects. The classifications shown in Tables I and II were identical whether or not the jackknife option was used, suggesting that the 6-variable discriminant functions might be robust enough to yield replicable results on new samples. We nevertheless verified that hypothesis by separating the available population into training sets {150 normal subjects and 76 and 96 patients for the anterior MI and inferior MI groups} and testing sets {the other 86 normal subj ects, 38 anterior MI and 48 inferior MI patients}; we calculated discriminant functions on the training sets and ran them on the testing sets to classify the remaining patients. Results for specificity and sensitivity in the training sets were similar to those for the entire group {Tables I and II}; almost no deterioration {less than 1%} was noticed in the testing sets.
nonelectrocardiographic evidence, further exclusion and assignment of the remaining subjects to an anterior MI or an inferior MI group were performed according to 12-lead electrocardiographic criteria in the acute phase. As a result, highly selected groups of patients were subjected to subsequent analysis. This probably accounts for the high rates of correct classification observed. Although other clinical tests such as the echocardiogram or radionuclide imaging may be better than the electrocardiogram for determining the presence and location of infarcts, they also yield a number of false-positive and false-negative responses. ~6-18 Moreover, several reports correlating electrocardiographic and pathologic findings agree that the electrocardiogram is reliable in detecting necrosis and in indicating its general location19-21; in the present study, the terms anterior MI and inferior MI are used in their generally accepted electrocardiographic sense. Although they do not consistently and exactly describe the site and extent of necrosis, they were acceptable descriptors of underlying myocardial involvement. There is often overlap between these 2 categories, e.g., because of the apical involvementY It is also recognized that anterior MIs generally involve more extensive damage than inferior MIs, and thus the extent of the involvement may play a significant role in assignment of individual patients into these 2 classes. 22 This
Discussion Limitations of the study: From a general population of MI patients collected on the basis of available
RA
'
¢
:'
~
~
~
"
"
LA
' 97
LL
v4
v5
v6
I
4
87
FIGURE 4. A v e r a g e d 120-lead map of normal subjects and patients with inferior myocardial infarction (M|). At each electrode site, the superimposed tracings represent averaged, time-normalized electrocardiographic signals for the normal (doffed line) and the inferior MI ( solid line) populations, respecUvely. The precordial leads are labeled V1 to Vs, the limb leads RA (right arm), LA (left arm) and LL (left leg). Leads 4, 20, 65, 87, 97 and LL indicate specific leads selected in the discriminant analysis for the separation of normal subjects from
patients with inferior MI (Table II).
November 1, 1986
limitation also applies to the lead sets selected in this study from body surface potential maps for optimal classification of patients with either anterior or inferior MI. Relation to previous work: Results of the present study agree with those of other studies 5A22 showing that body surface potential maps contain diagnostic information not present in the standard 12-lead electrocardiogram. At the time body surface potential maps were recorded, 46 of 258 patients (18%] failed to show the abnormalities in the initial standard electrocardiographic recordings. The mean interval between the acute event and the 120-lead mapping was 10 months (range I day to 24 months). Most normalization occurred when the interval exceeded I month, but in 5 patients, the diagnostic Q waves disappeared within the first month. Vincent et al 6 studied 100 patients with old MI; they found that in a group of 28 patients with electrocardiographic evidence of previous isolated posterior MI, 8 patients (29%) had either a normal 12lead electrocardiogram or showed only nonspecific ST-T abnormalities at the time body surface potential maps were recorded. In a related study, Flowers et al 5 showed specific map abnormalities in 19 of 22 patients
THE AMERICAN JOURNAL OF CARDIOLOGY Volume 58
869
with acute posterior MI 2 to 4 weeks after the event, while only 8 standard electrocardiograms revealed persistent Q waves and 14 (64%] did not. In our study, 20% of the patients with previous electrocardiographic evidence of inferior MI returned tO nonspecific patterns. Since the patient selection and the diagnostic Q wave criteria are comparable in our study and in the above-cited studies, the discrepancy could be ascribed to the relatively small samples reported.
Optimal leads and electrocardiographic features: For both MI classes, specific leads selected from 120lead data significantly increase the diagnostic classification over the standard 12-lead electrocardiogram. The sites from which these leads originate are different for each group, although common areas can be identified, such as the right Subclavicular area, the upper dorsal region and the lower left anterior and posterior axillary areas. In a previous study 13 we showed that leads recorded in these common areas were optimal for the separation of normal subjects from patients with MI, irrespective of the type of the infarcts. These leads, however, were not found to be optimal in differentiating patients with anterior MI
DIIIU RA
LA
LL
L 97
L 2O
L
V1
£ L -,.___
•_.=_ ~_.__ L...~ ~ - . ~
V2
./.,~
~,~,_
L 65 V3
V4
V5
V6
L4. LB7 FIGURE 5. Olscriminant 120-1ead map for separation of normal subjects from patientswith Inferiormyocardlal infarction.At each electrode site, instantaneous discriminant indexes, calculated at 8 equidistant time instants for ORS (bold vertical bars) and for ST-T (fine vertical bars) are depicted; the amplitude of the discriminant Index (DI) equals 1 U when the corresponding mean voltages at ! / 8 QRS, 2 / 8 QRS, and so on, for normal subjects and inferior infarction patients are I (pooled) standard deviation apart. Corresponding leads are labeled as In Figure 4.
870
oPTIMALLEADS FOR ANTERIOR AND INFERIOR MYOCARDIAL INFARCTION
1
"
/
,,,,iilll l LEAD 87
LEAD
97
LEAD
65
I
i,,,lllllliJJ aVF
/
I]llllllll I Illll LEAD
20
LEAD
4
FIGURE 6. Best measurements of the linear discriminant function computed for the separation of the normal group from the Inferior myocardial infarction group. The depicted leads are 4, 20, 65, 87, 97 and LL (same as Figure 4), from which optimal classifiers are extracted. The normal (dotted line) and inferior infarction (solid line) waveforms are averaged, time-normalized and superimposed. The bottom curve In each box represents the discrimlnant index as it varies in time; marks on abscissa Indicate the 8 equidistant time Instants on QRS and ST-T; discriminant indexes corresponding to these time Instants were shown in Figure 5 for all 120 leads of the map. Arrows point to the features selected in the stepwlse discrimlnant procedure. r
from those with inferior MI, nor either of these groups from normal subjects. The most discriminating leads for the anterior MI group are located "centrally," in the vicinity of the precordial area {e.g., leads 50, 69 and 32} or symmetrically in the back {lead 100}; in contrast, the best leads for the inferior MI group are distributed "peripherally": high on the anterior thorax and in the back and about the lower anterior and posterior chest circumference. Such clustering of typical patterns in anterior MI and their dispersion in inferior MI was also reported by McPherson et al. 22 In both groups, features in early QRS are potent discriminators, corresponding to Q waves or reciprocal R-wave changes. They are generally observed in the superior half Of the left hemithorax and in the opposite lower half of the back in anterior MI patients. In inferior MI patients they are almost always located in the lower right chest, often extending inferiorly toward the lower back and in the upper dorsal area. Similar patterns were described by Vincent et al 6 in their patients with old inferior MI. Later QRS features, however, also show high discriminant indexes. They are measured in the anterior MI group in the lower anterior chest as either reduced R waves or Q waves. In most cases, disappearance of Q waves in the back leads is noticed. In the inferior MI group, mid-QRS measurements show reduced negative voltages in the right subclavicular area and prominent R waves in the left subclavicular region; oppositely directed, deep and large Q waves or small R waves are noted in the lower back. The large zone of positivity in the upper left anterior chest in mid-QRS was also observed by Flowers et al. 5 The high discriminant ability of repolarization features is noted in late T for both groups. The measurements yield low-voltage positive or negative T waves, peripherally located in inferior MI, whereas biphasic T waves with terminal negativity are distributed over the left hemithorax in anterior MI. A similar behavior of late T potentials in anterior MI was reported by Mirvis. z3 Early ST-segment measurements also exhibit discriminant capacity, although to a lesser degree; when present, they reveal ST elevation in anterior MI and ST depression in inferior MI and could, in part, result from presence of acute or recent MIs, Although ST-T changes are usually considered nonspecific, they accounted for much of the separation achieved between normal subjects and patients with either anterior MI or inferior MI. Part of this effectiveness is probably related to the relative young age of the normal reference group. The most important fact remains that in most patients in which the standard 12-lead electrocardiogram no longer met the criteria for the recognition of MI, typical body surface potential map patterns were still present. In most cases, the specific map abnormalities were either outside the location of standard electrode positions or stayed for too short a time at sites where diagnostic Q waves could be inscribed in corresponding standard leads,
November 1, 1986
Clinical applications and prospects: The leads identified in the present study for each MI class are practically suited for clinical applications using either a 1-step, 3-group classification procedure or a 2-step approach separating the normal subjects from all MI patients in the first step and the anterior MI from the inferior MI patients in the second. We compared both strategies by using as a lead set the ensemble of optimal leads selected for each class separately and found almost identical diagnostic performance. The present approach, identifying optimal lead sites from body surface potential maps for subclasses of MI patients in combination with anatomic and other nonelectrocardiographic data, may provide a basis for quantitative evaluation of MI. In view of the superior diagnostic information content of body surface maps, development of validated criteria for optimally selected leads could provide the clinical cardiologist with a practical, inexpensive, noninvasive and quantitative tool for the assessment of MI.
References 1. De Ambroggi L. Taccardi B. Macchi E. Body sarface maps of heart poten-
tials, tentative localization of pre-exited areas in forty-two Wolff-ParkinsonWhite patients. Circulation 1976:54:251-263 2. Abildskov ]A, Burgess MI, Urie PM. Lux RL. Wyatt RF. The unidentified information content of the electrocardiogram. Circ Res 1977;40:3-7. 3. Kornreich F, Rautaharju PM. The missing waveform and diagnostic infor-
mation in the standard 12-lead electrocardiogram. I Electrocardiol 1981; 14:341-350. 4. Stilli D, Musso E, Macchi E, Taccardi B, Rolli A, Aurier E, Favaro L, Botti G. Diagnostic value of body surface maps in left bundle-branch block. Adv Car-
dial 1981;28:36-40. 5. Flowers NC, Horan LG, Gurbachan SS, Hand RC, Johnson JC. New evidence far inferoposterior myocardial infarction on surface potential maps.
Am I Cardiol 1976;38:576-581. 6. Vincent GM, Abildskov ]A, Burgess MS, Millar K, Lux RL, Wyatt RF. Diagnosis of old inferior myocardial infarction by body surface isopotential
maps. Am I Cardiol 1977;39:510-515. 7. Kornreich F, Brismee D. The missing waveiorm information in the orthogonaJ electrocardiogram (Frank leads). II. Diagnosis of left ventricular hypertrophy and myocardial infarction from "total" surface waveform information. Circulation 1973;48:984-1004. 8. Barr RC, Spach MS, Herman-Giddens GS. Selection of the number and
THE AMERICAN JOURNAL OF CARDIOLOGY
Volume 58
871
positions of measuring locations far electrocardiography. IEEE Trans Biomed~ Eng 1971;18:125-138. 9. Lux RL, Burgess MJ, Wyatt RF, Evans AK, Vincent GM, Abildskov JA.
Clinically practical lead systems far improved electrocardiography: comparison with precordial grids and conventional lead systems. Circulation 1979; 2:356-363. 10. Uijen GIH, Heringa A, Van Oosterom A. Data redaction of body surface potential maps by means of orthogonal expansions. IEEE Trans Biomed Eng 1984;BME-31:706-714. 11. Lux RL, Evans AK, Burgess MJ, Wyatt RF, Abildskov JA. Redundancy reduction for improved display and analysis of body surface potential maps. I. Spatial compression. Circ Res 1981;49:186-196. 12. Evans AK, Lux RL, Burgess MJ, Wyatt RF, Abildskov JA. Redundancy reduction for improved display and analysis of body surface potential maps. II. Temporal compression. Circ Res 1981;49:197-203. 13. Kornreich F, Rautaharju PM, Warren ], Montague TJ, Horacek BM. Identification of best electrocardiographic leads for diagnosing myocardial infarction by statistical analysis of body surface potential maps. Am I Cardiol 1985;56:852-856. 14, Montague TJ, Smith ER, Cameron DA, Rautaharju PM, Klassen GA, Flemington CS, Horacek BM. Isointegral analysis of body surface maps: sur-
face distribution and temporal variability in normal subjects. Circulation 1981;63:1166-1172. 15. Dixon WJ, Brown MB, eds. BMDP-79. Biomedical Computer Programs. P. Series. Berkeley, CA: University of California Press, 1979. 16. Van Reet RE, Quinones MA, Poliner LR, Nelson ]G, Waggoner AD, Kanon D, Lubetkin SI, Pratt CM, Winters WL. Comparison of two-dimensional echocardiography with gated radienucleide ventriculography in the evaluation of global and regional ventricular function in acute myocardial infarction. IACC
1984;3:243-252. 17. Berman DS, Amsterdam EA, Hines HH, Salel AF, Bailey G], DeNard0 GL, Mason DT. New approach to interpretation of technetium-99m pyrophosphate scintigraphy in detection of acute myocardial infarction. Am [ Cardiol
1977;39:341-346. 18. Williams RA, Cohn PF, Vokonas PS, Young E, Herman MV, Gorlin R.
Electrocardiographic, arteriographic and ventriculographic correlations in transmural myocardial infarction. Am I Cardiol 1973;31:595~599. 19. Roberts WC, Gardin ]M. Location of myocardial infarcts: a confusion of terms and definitions. Am I Cardiol 1978;42:868-872. 20. Sullivan W, Vl0daver Z, Tuna N, Long L, Edwards JE. Correlation of electrocardiographic and pathologic findings in healed myocardial infarction. Am I Cardiol 1978;42:724-732. 21. Savage RM, Wagner GS, Ideker RE, Podolsky SA, Hackel DB. Correlation of postmortem anatomic findings with electrocardiographic changes in patients with myocardial infarction. Retrospective study of patients with typical anterior and posterior infarcts. Circulation 1977;55:279-285. 22. McPherson DD, Horacek BM, Johnstone DE, Lalonde LD, Spencer A, Montague T]. Q-wave infarction: pathophysiolagy of body surface potential map and ventriculographic patterns in anterior and inferior groups. Can 1
Cardiol 1986;suppl A:91A-98A. 23. Mirvis DM. Body surface distributions of repolarization forces during acute myocardial infarction. I. Isopotential and isoarea mapping. Circulation 1980;62:878-887.
The
American
Journal
of CARDIOLOGY
®
VOLUME 58 NUMBER 10
CORONARY ARTERY DISEASE
Identification of Best Electrocardiographic Leads for Diagnosing Anterior and Inferior Myocardial Infarction by Statistical Analysis of Body Surface Potential Maps FRED KORNREICH, MD, PhD, TERRENCE J. MONTAGUE, MD, PENTTI M. RAUTAHARJU, MD, PhD, PIERRE BLOCK, MD, PhD, JAMES W. WARREN, MSc, and MILAN B. HORACEK, PhD
In view of the increasing interest in quantifying and modifying the size of myocardial infarction (MI), it is important to look for clinically practical subsets of electrocardiographic leads that allow the earliest and most accurate diagnosis of the presence and electrocardiographic type of MI. A practical approach is described, taking advantage of the increased information content of body surface potential maps over standard electrocardiographic techniques for facilitating clinical use of body surface potential maps for such a purpose. Multivariate analysis was performed on 120-lead electrocardiographic data, simultaneously recorded in 236 normal subjects, 114 patients with anterior MI and 144 patients with inferior MI, using as features instantaneous voltages on time-normalized QRS and ST-T waveforms. Leads and features for optimal separation of normal subjects from, respectively, anterior MI and inferior MI patients were selected. Features measured on leads originating from the upper left precordial area, lower midthoracic region and the
back correctly identified 97% of anterior MI patients, with a specificity of 95 %; in patients with inferior MI, features obtained from leads located in the lower left back, left log, right subclavicular area, upper dorsal region and lower right chest correctly classified 94 % of the group, with specificity kept at 95 %. Most features were measured in early and mid-QRS, although very potent discriminators were found in the late portion of the T wave. Repeatability of the results was investigated by separating the study population in training and testing sets; no deterioration was observed when the discriminant functions computed on the training sets were run on the testing sets. In comparison, at the same level of specificity and with the same number of features (n = 6), the standard 12-lead electrocardiogram correctly diagnosed 89 % of anterior MI and 85 % of inferior MI patients. Thus, diagnosis of anterior and inferior MI can be substantially improved by appropriate selection of electrocardiographic leads and features. (Am J Cardiol 1986;58:863-871)
863
864
OPTIMALLEADS FOR ANTERIOR AND INFERIOR MYOCARDIAL INFARCTION
T h e 12-lead electrocardiogram is one of the most reliable methods of diagnosing the presence and general location of myocardial infarction (MI}. Despite the undisputed merits of the standard electrocardiogram, there is little doubt that body surface potential maps contain diagnostic information not present in conventional lead systems2 -4 The increased information content of body surface potential maps over standard electrocardiographic techniques has shown promising results in assessment of patients with MI~-7; however, to facilitate clinical use of body surface potential maps, data reduction is required. 8-12 In a previous study 13we described a practical approach for the extraction of diagnostic information from body surface potential maps by performing multivariate analysis of 120-lead data simultaneously recorded from 361 persons; a subset of optimal leads and features was found for separation of 184 normal subjects from 177 patients with MI, irrespective of the type of infarction. The present study uses body surface potential maps from 494 subjects to identify optimal leads and features to distinguish 236 normal subjects from 114 patients with anterior MI and 144 with inferior MI. This approach, by identifying specific locations of electrocardiographic leads with improved diagnostic information, may lead to further progress in quantitative assessment of MI.
Methods Study population: Five hundred sixty-three subjects were studied. Group A consisted of 236 normal control subjects. None of these subjects had evidence of heart disease by history, physical examination, 12lead electrocardiogram and, when available, echocardiogram. They were 20 to 55 years old (mean 36). Group B included 327 patients with first MI; all had a typical history of prolonged, ischemic-type cardiac pain and characteristic changes in enzyme levels. In many cases the diagnosis was further substantiated by coronary angiography and ventriculography, echocardiography or nuclear imaging. All patients had at least one 12-lead electrocardiogram recorded in the acute phase of the infarction. Patients were excluded if they had electrocardiographic evidence of complete left or right bundle branch block or maj or nonspecific intraventricular conduction delay {QRS 120 ms or longer). The study population was further classified according to the appearance of the QRS complex. From the Unit for Cardiovascular Research and Engineering and Department of Cardiology, Free University of Brussels, Brussels, Belgium, and the Departments of Physiologyand Biophysics and Medicine, Dalhousie University, and the Victoria General Hospital, Halifax, Nova Scotia, Canada. This study was supported by the National Fund for Scientific Research {FGWO 3.0017.75 and FGWO 3.0095.86}, the Medical Research Council of Canada (PG-30}and the Nova Scotia Heart Foundation. Manuscript received May 5, 1986; revised manuscript received July 10, 1986, accepted July 11, 1986. Address for reprints: Fred Kornreich, MD, PhD, Unit for Cardiovascular Research and Engineering, Free University Brussels, Laarbeeklaan 103, B-1090 Brussels, Belgium.
Patients were classified as having anterior MI if they had Q waves at least 30 ms in duration in leads I, aVL or V~ to V6, or initial R waves of 0.2 mV or less in leads V1 and V2. Patients were classified as having inferior (posterior) MI if abnormal Q waves were present in leads II or aVF or if R waves in V~, V2 were exceedingly tall (R to S ratio 1 or more) or broad (at least 40 ms). Body surface mapping: We used previously described methods of recording, processing and displaying 120 body surface electrocardiographic signals. 14 Briefly, digitized signals were recorded simultaneously from 117 torso and 3 limb leads with Wilson's central terminal as reference potential, at 500 samples/s per channel. Tracing quality was monitored visually during the recording; later, the stored data were processed by performing selective averaging and again carefully inspected and edited. All leads judged invalid were deleted and replaced by interpolation of data from surrounding leads. We time-normalized separately the QRS waveform and the ST-T waveform and represented them by 70 and 180 points, respectively. Patients with MI were studied a mean of 10 months after MI (range I day to 24 months); the interval between the acute event and the recording of body surface potential maps was less than 1 week in 39 patients (15%), between 1 week and 1 month in 30 (12%) and more than 1 month in 189 (73%). Discrimtnant maps: To identify leads and features that were significant in discriminating between 2 groups of maps, a previously described technique was used23 Briefly, averaged 120-lead maps were computed separately for the normal group and each MI group by averaging sample-by-sample, time-normalized electrocardiograms from subjects in each group. Difference maps were then calculated by subtracting corresponding sample values of 120 averaged signals of normal and respectively, anterior MI and inferior MI groups. Discriminant maps were obtained for each 2-group comparison (normal vs anterior MI or normal vs inferior MI} by dividing each instantaneous difference with the corresponding composite standard deviation computed from the population under consideration. The values thus achieved, referred to as discriminant indexes, were strictly proportional to Student t test statistics and provided, in contrast to difference maps, information on the capability for each measurement in each lead, to separate each class of MI patients from this normal group. As noted earlier, 13 inspection of discriminant maps provided clues for identifying leads and features for bigix~p_ differentiation. Feature extraction and statistical analysis: We used instantaneous amplitude measurements obtained by sampling time-normalized QRS and ST-T waveforms at equal intervals; this resulted in 8 sam. ples for QRS and 7 samples for ST-T, i.e., 15 × 120 per subject. For each bigroup classification, programs from the Biomedical Program library were used (BMDPTM)2 ~ The statistical procedure involved 2 steps. First, the best classifiers were selected from the total number of available features by a stepwise selec-
November 1, 1986 THE AMERICAN JOURNAL OF CARDIOLOGY Volume 58
865
tribution of the sites with large discriminant ability depends on whether that ability occurs during depolarization or repolarization. The best leads for QRS cluster around lead 50 in the upper left precordial and subclavicular areas and show peak discriminant indexes in early depolarization {1/8 QRS and mainly 2/8 QRS}. A similar pattern can be observed in the back on the mid-dorsal line {e.g., lead 100} and probably represents, in part, a mirror image. A secondary area of discriminant capability is observed later in the depolarization in the lower thoracic region, under and to the left of the xiphoid appendix. The sites of high ST-T discriminant indexes are noted mainly in the left precordial area, extending to the lower anterior and posterior axillary regions, and in the upper right thoracic and subclavicular areas, extending under the right axilla into the upper dorsal region. The maximal discriminant capability appears at 6/8 ST-T, slightly after the peak of the T wave, as exemplified in leads 69 and aVR. Except for QRS in lead V2 and ST-T in leads V5 and V6, the precordial leads are on the margin of the areas with the largest discriminant indexes. Table I lists the results of the discriminant analysis. The first 6 variables are listed in their order of entry in the stepwise procedure, although practically no further improvement is achieved after 3 steps. Adjusting the specificity to 95% instead of 99%, the sensitivity
tion; then, these selected classifiers were combined into the properly weighted linear discriminant functions. To test the robustness of the discriminant functions, the "jackknife" procedure, available in the BMDP7M package, was also used. With this procedure, each case is eliminated in turn from the computation and subsequently assigned to 1 of the groups formed by the remaining cases. Because of the very large number of variables {1,800 instantaneous measurements per patient}, the program processed 80 input variables at a time and selected the 20 best from each batch, repeating the procedure until it acquired a final set of 20 optimal classifiers. Varying the composition of the batches and their order of entry did not affect the first 6 classifiers selected.
Results Anterior infarction: In 17 of 114 patients {15% }with initial electrocardiographic evidence of isolated anterior MI, the standard 12-lead electrocardiogram no longer revealed depolarization abnormalities at the time of body surface potential map recording. The mean interval between the acute event and the recording of 120-lead data was 8 months {range 1 day to 20 months}. Figure 1 shows the averaged 120-lead maps for the normal and the anterior MI populations. Figure 2 shows the corresponding discriminant map. The dis-
RA
LA
LL
rrrr
.,
-
A_~.. A_~. ~_.._ . . ~ 69
V5
V4 (
!:
,
;
~
.:
."
..~_..~ ,~ .....
lOO
M6 ;
,,
FIGURE 1. Averaged 120-lead map of normal subjects and paUents with anterior myocardial infarction (MI). The top row Is at the level of the sternal notch and the botfom rowat the umbilical level; a wider vertical gap separates the anterior from the posterior portion of the thorax. At each electrode site, the superimposed tracings represent averaged, time-normalized electrocardiographic signals for the normal (dotted line) and the anterior MI (solid line) populations, respectively. The precordlal leads are labeled V1 to V e, the limb leads RA (right arm), LA (left arm) and LL (left leg). Leads 20, 32, 50, 69 and 100 indicate specific leads selected in the discriminant analysis for the separation of normal persons from patients with anterior MI (Table I).
866
OPTIMAL LEADS FOR ANTERIOR AND INFERIOR MYOCARDIAL INFARCTION
TABLE I DifferenUation of 236 Normal Subjects from 114 Patients with Anterior Myocardial Infarction Using Instantaneous Measurements on 120-Lead Data Stepwise Selection of Variable Type Lead Lead
Lead Lead Lead Lead
50 69 32 100 50 20
2/8 618 4/8 1/8 1/8 2/8
Specificity (% )
Sensitivity (%)
95 97 98 98 99 99
89 94 96 95 94 95
QRS ST-T QRS QRS QRS ST-T
with 6 variables reached 97%. The most potent discriminators are the amplitudes measured on leads 50, 69 and 32; these leads are located in areas described as showing high discriminant index values. Figure 3 shows the leads and features selected in the discriminant analysis and listed in Table I. Early QRS measurements are selected in leads 50 and 100; they correspond to large Q waves (lead 50) or absent Q waves (lead 100) observed in the respective regions from w h i c h these leads originate. The discriminant capability observed later in QRS also represents, as in lead 32, loss of anterior forces in the lower middle and left
TABLE II Differentiation of 236 Normal Subjects from 144 Patients with Inferior Myocardial InfarcUon Using Instantaneous Measurements on 120-Lead Data Stepwise Selection of Variable Type
Lead LL or Lead Lead Lead Lead
87 aVF 20 97 65 4
6/8 ST-T 218 QRS 4/8 QRS 2/8 QRS 1/8 ST-T 3/8 QRS
Specificity (%)
Sensitivity (%)
92 93 95 97 98 98
83 84 91 92 93 93
thoracic region. The maximal index for lead 69 peaks in late T and is located in a large are of near-0 or negative voltages on the left anterior chest, extending to the lower left axillary region while the selected variable on lead 20 is an early ST-T measurement, relating to ST-segment elevation in the anterior MI group. The diagnostic performance achieved with multivariate analysis on body surface potential maps was compared with that of the standard 12-lead electrocardiogram, recorded at the same time. Using 6 instantaneous measurements as variables and keeping the
DIll u RA
LA
LL
L,,.,L,,,., L,..,L L L L. L ~'0
L,,,,,L L_ L IL,,,,[.,,,, L 50
V1
V2
L 69
V4
V5
L1 O0
V6
L 32
FIGURE 2. Discrlminant 120-lead map for separation of normal subjects from patients with anterior myocardial Infarction. At each electrode site, Instantaneous dlscrlmlnant Indexes (see text), calculated at 8 equidistant Ume instants for QRS (bold vertical bars) and for ST-T (fine vertical bars) are shown; the amplitude of the dlscrlmlnant Index (DI) equals 1 U when the corresponding mean voltages at 1/8 QRS, 2/8 GRS, and so on, for normal subjects and patients with anterior myocardial Infarction are 1 (pooled) standard deviation apart. Leads are labeled as in Figure 1.
November 1, 1986 THE AMERICAN JOURNAL OF CARDIOLOGY Volume 58
specificity at 95%, 89% of the anterior MIs were correctly classified. Although most of the 12-lead variables selected in the stepwise procedure were early QRS measurements in leads V1, V2 and aVL, the best classifier was 6/8 ST:T in lead Vs. This finding probably accounts for the discrepancy between the diagnostic yield of Q-wave criteria {17 missed anterior MI cases} and variables selected with multivariate statistics {13 missed anterior MI cases}. Inferior infarction: Of the 144 patients with initial 12-lead electrocardiographic criteria of inferior MI, 29 {20%} failed to meet these criteria at the time body surface potential maps were recorded, a mean of 11 months later {range 1 day to 24 months}. Figure 4 shows the averaged 120-lead map for the normal and the inferior MI groups. Figure 5 shows the corresponding discriminant map. In contrast to the anterior MI group, the areas with the largest discriminant indexes are common to QRS and ST-T. These areas are located outside the precordial area. Except for the ST-T wave in leads V~ and Vs, the limb leads have higher discriminatory capability than the precordial leads. The sites where initial QRS measurements {2/8 QRS} are optimal for diagnosing inferior MI are located in the right upper back and the lower right side of the thorax. Later in depolarization {3/8 and 4/8 QRS}, additional sites are observed over the entire inferior border around the thorax as well as high in both subclavicular regions. The discriminant ability of the ST-T wave at 6/8 ST-T is als0 maximal in the same areas {except the lower right and upper left thorax}. High discriminant indexes were observed in the left precordial area {e~g., lead 65} during early ST segment, a pattern different from those in the other leads with high T indexes in late repolarization. Table II is a list of the results of the discriminant analysis. The best 6 classifiers are shown; maximal differentiation is achieved with 5 variables. In contrast to the results listed in Table I, the increase, with each Step in diagnostic yield, is slower and levels off later. Adjusting the specificity at step 6 at 95%, a sensitivity of 94% was observed. All selected variables are measured on leads originating from the sites previously described. Figure 6 shows these leads. The most potent single classifier is found in lead 87 and corresponds approximately to the peak of the T wave. In fact, the entire repolarization in the inferior MI group is characterized by flat or slightly negative T waves over the upper right anterior and dorsal areas and the lower left flank. The early QRS measurements in aVF and lead 4 and in lead 97 indicate the presence of Q waves in the former leads and a reciprocal R wave in the latter. Later QRS voltages reflect loss of electrical forces in the upper right chest and the lower left posterior flank {3/8 and 4/8 QRS in leads 20 and 87}. Here too, multivariate analysis was performed on 12 lead measurements at the time body surface potential maps were obtained. With 6 variables and specificity adjusted at 95%, a sensitivity of 85% was reached {22 missed inferior MI cases}. This compared favorably with the 29 missed inferior MI cases when Q-wave criteria were applied. Again, the best classifier was 6/8 ST-T in lead
867
,/ .',°
/
LEAD
50
LEAD 100
JJZLLLU LEAD
69
LEAD
,32
LEAD
20
FIGURE 3. Best measurements of the linear dlscriminant function computed for separation of the normal group from the anterl0r myocardial Infarction group. The leads are 20, 32, 50, 69 and 100 (same as Figure 1), from which optimal classifiers are extracted. The normal (dotted line) and anterior MI (solid line) waveforms are averaged, time-normalized and superimposed. The bottom curve in each box represents the discriminent index as it varies in time; the marks on the abscissa indicate the 8 equidistant time instants on QRS and ST-T; discriminant indexes corresponding to these time instants were shown In Figure 2 for all 120 leads of the map. Arrows point to the features selected in the stepwise discriminant procedure.
868
OPTIMAL LEADS FOR ANTERIOR AND INFERIOR MYOCARDIAL INFARCTION
DII, followed by aVF {2/8 QRS and 4/8 QRS), aVL {7/8 QRS} and ST-T measurements in leads V2 {1/8 ST-T} and V6 (6/8 ST}. Repeatability of results: The discriminant programs {BMDP7M) were run without and with the jackknife option, the latter enabling each subject of the design set to be classified in turn with the discriminant function derived from all the remaining subjects. The classifications shown in Tables I and II were identical whether or not the jackknife option was used, suggesting that the 6-variable discriminant functions might be robust enough to yield replicable results on new samples. We nevertheless verified that hypothesis by separating the available population into training sets {150 normal subjects and 76 and 96 patients for the anterior MI and inferior MI groups} and testing sets {the other 86 normal subj ects, 38 anterior MI and 48 inferior MI patients}; we calculated discriminant functions on the training sets and ran them on the testing sets to classify the remaining patients. Results for specificity and sensitivity in the training sets were similar to those for the entire group {Tables I and II}; almost no deterioration {less than 1%} was noticed in the testing sets.
nonelectrocardiographic evidence, further exclusion and assignment of the remaining subjects to an anterior MI or an inferior MI group were performed according to 12-lead electrocardiographic criteria in the acute phase. As a result, highly selected groups of patients were subjected to subsequent analysis. This probably accounts for the high rates of correct classification observed. Although other clinical tests such as the echocardiogram or radionuclide imaging may be better than the electrocardiogram for determining the presence and location of infarcts, they also yield a number of false-positive and false-negative responses. ~6-18 Moreover, several reports correlating electrocardiographic and pathologic findings agree that the electrocardiogram is reliable in detecting necrosis and in indicating its general location19-21; in the present study, the terms anterior MI and inferior MI are used in their generally accepted electrocardiographic sense. Although they do not consistently and exactly describe the site and extent of necrosis, they were acceptable descriptors of underlying myocardial involvement. There is often overlap between these 2 categories, e.g., because of the apical involvementY It is also recognized that anterior MIs generally involve more extensive damage than inferior MIs, and thus the extent of the involvement may play a significant role in assignment of individual patients into these 2 classes. 22 This
Discussion Limitations of the study: From a general population of MI patients collected on the basis of available
RA
'
¢
:'
~
~
~
"
"
LA
' 97
LL
v4
v5
v6
I
4
87
FIGURE 4. A v e r a g e d 120-lead map of normal subjects and patients with inferior myocardial infarction (M|). At each electrode site, the superimposed tracings represent averaged, time-normalized electrocardiographic signals for the normal (doffed line) and the inferior MI ( solid line) populations, respecUvely. The precordial leads are labeled V1 to Vs, the limb leads RA (right arm), LA (left arm) and LL (left leg). Leads 4, 20, 65, 87, 97 and LL indicate specific leads selected in the discriminant analysis for the separation of normal subjects from
patients with inferior MI (Table II).
November 1, 1986
limitation also applies to the lead sets selected in this study from body surface potential maps for optimal classification of patients with either anterior or inferior MI. Relation to previous work: Results of the present study agree with those of other studies 5A22 showing that body surface potential maps contain diagnostic information not present in the standard 12-lead electrocardiogram. At the time body surface potential maps were recorded, 46 of 258 patients (18%] failed to show the abnormalities in the initial standard electrocardiographic recordings. The mean interval between the acute event and the 120-lead mapping was 10 months (range I day to 24 months). Most normalization occurred when the interval exceeded I month, but in 5 patients, the diagnostic Q waves disappeared within the first month. Vincent et al 6 studied 100 patients with old MI; they found that in a group of 28 patients with electrocardiographic evidence of previous isolated posterior MI, 8 patients (29%) had either a normal 12lead electrocardiogram or showed only nonspecific ST-T abnormalities at the time body surface potential maps were recorded. In a related study, Flowers et al 5 showed specific map abnormalities in 19 of 22 patients
THE AMERICAN JOURNAL OF CARDIOLOGY Volume 58
869
with acute posterior MI 2 to 4 weeks after the event, while only 8 standard electrocardiograms revealed persistent Q waves and 14 (64%] did not. In our study, 20% of the patients with previous electrocardiographic evidence of inferior MI returned tO nonspecific patterns. Since the patient selection and the diagnostic Q wave criteria are comparable in our study and in the above-cited studies, the discrepancy could be ascribed to the relatively small samples reported.
Optimal leads and electrocardiographic features: For both MI classes, specific leads selected from 120lead data significantly increase the diagnostic classification over the standard 12-lead electrocardiogram. The sites from which these leads originate are different for each group, although common areas can be identified, such as the right Subclavicular area, the upper dorsal region and the lower left anterior and posterior axillary areas. In a previous study 13 we showed that leads recorded in these common areas were optimal for the separation of normal subjects from patients with MI, irrespective of the type of the infarcts. These leads, however, were not found to be optimal in differentiating patients with anterior MI
DIIIU RA
LA
LL
L 97
L 2O
L
V1
£ L -,.___
•_.=_ ~_.__ L...~ ~ - . ~
V2
./.,~
~,~,_
L 65 V3
V4
V5
V6
L4. LB7 FIGURE 5. Olscriminant 120-1ead map for separation of normal subjects from patientswith Inferiormyocardlal infarction.At each electrode site, instantaneous discriminant indexes, calculated at 8 equidistant time instants for ORS (bold vertical bars) and for ST-T (fine vertical bars) are depicted; the amplitude of the discriminant Index (DI) equals 1 U when the corresponding mean voltages at ! / 8 QRS, 2 / 8 QRS, and so on, for normal subjects and inferior infarction patients are I (pooled) standard deviation apart. Corresponding leads are labeled as In Figure 4.
870
oPTIMALLEADS FOR ANTERIOR AND INFERIOR MYOCARDIAL INFARCTION
1
"
/
,,,,iilll l LEAD 87
LEAD
97
LEAD
65
I
i,,,lllllliJJ aVF
/
I]llllllll I Illll LEAD
20
LEAD
4
FIGURE 6. Best measurements of the linear discriminant function computed for the separation of the normal group from the Inferior myocardial infarction group. The depicted leads are 4, 20, 65, 87, 97 and LL (same as Figure 4), from which optimal classifiers are extracted. The normal (dotted line) and inferior infarction (solid line) waveforms are averaged, time-normalized and superimposed. The bottom curve In each box represents the discrimlnant index as it varies in time; marks on abscissa Indicate the 8 equidistant time Instants on QRS and ST-T; discriminant indexes corresponding to these time Instants were shown in Figure 5 for all 120 leads of the map. Arrows point to the features selected in the stepwlse discrimlnant procedure. r
from those with inferior MI, nor either of these groups from normal subjects. The most discriminating leads for the anterior MI group are located "centrally," in the vicinity of the precordial area {e.g., leads 50, 69 and 32} or symmetrically in the back {lead 100}; in contrast, the best leads for the inferior MI group are distributed "peripherally": high on the anterior thorax and in the back and about the lower anterior and posterior chest circumference. Such clustering of typical patterns in anterior MI and their dispersion in inferior MI was also reported by McPherson et al. 22 In both groups, features in early QRS are potent discriminators, corresponding to Q waves or reciprocal R-wave changes. They are generally observed in the superior half Of the left hemithorax and in the opposite lower half of the back in anterior MI patients. In inferior MI patients they are almost always located in the lower right chest, often extending inferiorly toward the lower back and in the upper dorsal area. Similar patterns were described by Vincent et al 6 in their patients with old inferior MI. Later QRS features, however, also show high discriminant indexes. They are measured in the anterior MI group in the lower anterior chest as either reduced R waves or Q waves. In most cases, disappearance of Q waves in the back leads is noticed. In the inferior MI group, mid-QRS measurements show reduced negative voltages in the right subclavicular area and prominent R waves in the left subclavicular region; oppositely directed, deep and large Q waves or small R waves are noted in the lower back. The large zone of positivity in the upper left anterior chest in mid-QRS was also observed by Flowers et al. 5 The high discriminant ability of repolarization features is noted in late T for both groups. The measurements yield low-voltage positive or negative T waves, peripherally located in inferior MI, whereas biphasic T waves with terminal negativity are distributed over the left hemithorax in anterior MI. A similar behavior of late T potentials in anterior MI was reported by Mirvis. z3 Early ST-segment measurements also exhibit discriminant capacity, although to a lesser degree; when present, they reveal ST elevation in anterior MI and ST depression in inferior MI and could, in part, result from presence of acute or recent MIs, Although ST-T changes are usually considered nonspecific, they accounted for much of the separation achieved between normal subjects and patients with either anterior MI or inferior MI. Part of this effectiveness is probably related to the relative young age of the normal reference group. The most important fact remains that in most patients in which the standard 12-lead electrocardiogram no longer met the criteria for the recognition of MI, typical body surface potential map patterns were still present. In most cases, the specific map abnormalities were either outside the location of standard electrode positions or stayed for too short a time at sites where diagnostic Q waves could be inscribed in corresponding standard leads,
November 1, 1986
Clinical applications and prospects: The leads identified in the present study for each MI class are practically suited for clinical applications using either a 1-step, 3-group classification procedure or a 2-step approach separating the normal subjects from all MI patients in the first step and the anterior MI from the inferior MI patients in the second. We compared both strategies by using as a lead set the ensemble of optimal leads selected for each class separately and found almost identical diagnostic performance. The present approach, identifying optimal lead sites from body surface potential maps for subclasses of MI patients in combination with anatomic and other nonelectrocardiographic data, may provide a basis for quantitative evaluation of MI. In view of the superior diagnostic information content of body surface maps, development of validated criteria for optimally selected leads could provide the clinical cardiologist with a practical, inexpensive, noninvasive and quantitative tool for the assessment of MI.
References 1. De Ambroggi L. Taccardi B. Macchi E. Body sarface maps of heart poten-
tials, tentative localization of pre-exited areas in forty-two Wolff-ParkinsonWhite patients. Circulation 1976:54:251-263 2. Abildskov ]A, Burgess MI, Urie PM. Lux RL. Wyatt RF. The unidentified information content of the electrocardiogram. Circ Res 1977;40:3-7. 3. Kornreich F, Rautaharju PM. The missing waveform and diagnostic infor-
mation in the standard 12-lead electrocardiogram. I Electrocardiol 1981; 14:341-350. 4. Stilli D, Musso E, Macchi E, Taccardi B, Rolli A, Aurier E, Favaro L, Botti G. Diagnostic value of body surface maps in left bundle-branch block. Adv Car-
dial 1981;28:36-40. 5. Flowers NC, Horan LG, Gurbachan SS, Hand RC, Johnson JC. New evidence far inferoposterior myocardial infarction on surface potential maps.
Am I Cardiol 1976;38:576-581. 6. Vincent GM, Abildskov ]A, Burgess MS, Millar K, Lux RL, Wyatt RF. Diagnosis of old inferior myocardial infarction by body surface isopotential
maps. Am I Cardiol 1977;39:510-515. 7. Kornreich F, Brismee D. The missing waveiorm information in the orthogonaJ electrocardiogram (Frank leads). II. Diagnosis of left ventricular hypertrophy and myocardial infarction from "total" surface waveform information. Circulation 1973;48:984-1004. 8. Barr RC, Spach MS, Herman-Giddens GS. Selection of the number and
THE AMERICAN JOURNAL OF CARDIOLOGY
Volume 58
871
positions of measuring locations far electrocardiography. IEEE Trans Biomed~ Eng 1971;18:125-138. 9. Lux RL, Burgess MJ, Wyatt RF, Evans AK, Vincent GM, Abildskov JA.
Clinically practical lead systems far improved electrocardiography: comparison with precordial grids and conventional lead systems. Circulation 1979; 2:356-363. 10. Uijen GIH, Heringa A, Van Oosterom A. Data redaction of body surface potential maps by means of orthogonal expansions. IEEE Trans Biomed Eng 1984;BME-31:706-714. 11. Lux RL, Evans AK, Burgess MJ, Wyatt RF, Abildskov JA. Redundancy reduction for improved display and analysis of body surface potential maps. I. Spatial compression. Circ Res 1981;49:186-196. 12. Evans AK, Lux RL, Burgess MJ, Wyatt RF, Abildskov JA. Redundancy reduction for improved display and analysis of body surface potential maps. II. Temporal compression. Circ Res 1981;49:197-203. 13. Kornreich F, Rautaharju PM, Warren ], Montague TJ, Horacek BM. Identification of best electrocardiographic leads for diagnosing myocardial infarction by statistical analysis of body surface potential maps. Am I Cardiol 1985;56:852-856. 14, Montague TJ, Smith ER, Cameron DA, Rautaharju PM, Klassen GA, Flemington CS, Horacek BM. Isointegral analysis of body surface maps: sur-
face distribution and temporal variability in normal subjects. Circulation 1981;63:1166-1172. 15. Dixon WJ, Brown MB, eds. BMDP-79. Biomedical Computer Programs. P. Series. Berkeley, CA: University of California Press, 1979. 16. Van Reet RE, Quinones MA, Poliner LR, Nelson ]G, Waggoner AD, Kanon D, Lubetkin SI, Pratt CM, Winters WL. Comparison of two-dimensional echocardiography with gated radienucleide ventriculography in the evaluation of global and regional ventricular function in acute myocardial infarction. IACC
1984;3:243-252. 17. Berman DS, Amsterdam EA, Hines HH, Salel AF, Bailey G], DeNard0 GL, Mason DT. New approach to interpretation of technetium-99m pyrophosphate scintigraphy in detection of acute myocardial infarction. Am [ Cardiol
1977;39:341-346. 18. Williams RA, Cohn PF, Vokonas PS, Young E, Herman MV, Gorlin R.
Electrocardiographic, arteriographic and ventriculographic correlations in transmural myocardial infarction. Am I Cardiol 1973;31:595~599. 19. Roberts WC, Gardin ]M. Location of myocardial infarcts: a confusion of terms and definitions. Am I Cardiol 1978;42:868-872. 20. Sullivan W, Vl0daver Z, Tuna N, Long L, Edwards JE. Correlation of electrocardiographic and pathologic findings in healed myocardial infarction. Am I Cardiol 1978;42:724-732. 21. Savage RM, Wagner GS, Ideker RE, Podolsky SA, Hackel DB. Correlation of postmortem anatomic findings with electrocardiographic changes in patients with myocardial infarction. Retrospective study of patients with typical anterior and posterior infarcts. Circulation 1977;55:279-285. 22. McPherson DD, Horacek BM, Johnstone DE, Lalonde LD, Spencer A, Montague T]. Q-wave infarction: pathophysiolagy of body surface potential map and ventriculographic patterns in anterior and inferior groups. Can 1
Cardiol 1986;suppl A:91A-98A. 23. Mirvis DM. Body surface distributions of repolarization forces during acute myocardial infarction. I. Isopotential and isoarea mapping. Circulation 1980;62:878-887.