Improving the diagnosis of bundle branch block by analysis of body surface potential maps

Improving the diagnosis of bundle branch block by analysis of body surface potential maps

Available online at www.sciencedirect.com Journal of Electrocardiology 42 (2009) 651 – 659 www.jecgonline.com Improving the diagnosis of bundle bran...

596KB Sizes 0 Downloads 27 Views

Available online at www.sciencedirect.com

Journal of Electrocardiology 42 (2009) 651 – 659 www.jecgonline.com

Improving the diagnosis of bundle branch block by analysis of body surface potential maps Victoria Donis, MS, a,⁎ M. Salud Guillem, MS, a Andreu M. Climent, MS, a Francisco Castells, PhD, a Francisco J. Chorro, MD, b Jose Millet, PhD a b

a ITACA, Universidad Politécnica de Valencia, Valencia, Spain Cardiology Service, Hospital Clínico Universitario de Valencia, Valencia, Spain Received 31 July 2008

Abstract

Bundle branch block (BBB) is a defect on the electrical conduction system of the heart diagnosed by analyzing electrocardiogram (ECG) morphology. Our study aims to determine whether mapping information, specifically QRS duration and observation data from maps obtained using body surface potential mapping (BSPM), can be helpful in BBB diagnosis. We studied 64-lead BSPM recordings of 18 BBB patients and 9 controls with normal ventricular conduction. QRS duration was measured from the BSPM information obtained. The BSPM maps were computed along the QRS complex for each individual and group, and maps for each group were compared with maps for each individual. QRS complexes of the 12 standard leads were computed for each individual and group, and complexes of each group were compared with the complexes of each individual. QRS duration measured for all available leads (64 unipolar leads and 12 standard leads) was 7.4 ± 3.9 milliseconds longer than QRS duration measured only in the standard 12-lead ECG for left BBB patients (LBBB) and 15.3 ± 10.8 milliseconds longer for right BBB with left anterior fascicular block patients (RBBB_LAFB). In case of comparisons based on the standard ECG, sensitivity was 76.9% for LBBB patients and 66.6% for control subjects. However, classification based on map comparisons showed a sensitivity of 93% for LBBB patients and 89% for controls. QRS duration measured from BSPM information does not differ significantly from 12-lead standard ECG measurement for LBBB. However, differences are higher for RBBB_LAFB patients. Representative BSPM maps permit an automatic classification of the subjects. © 2009 Elsevier Inc. All rights reserved.

Keywords:

Bundle branch block; Body surface potential mapping; Fascicular block; QRS duration; Automatic classification

Introduction Bundle branch block (BBB) is a defect on the electrical conduction system of the heart, which develops when one of the branches or the fascicles of the His bundle cannot transmit the electrical impulses that cause ventricular contraction. As a result, these impulses have to be transmitted by another route, and consequently, both ventricles fail to contract simultaneously. The diagnosis of BBB is based on the morphology of the electrocardiogram (ECG) signal and the duration of the QRS complex. If QRS duration exceeds 120 milliseconds, the blockade is considered to be complete.1,2 ⁎ Corresponding author. ITACA (BIO) Edificio 8G, 46022 Valencia, Spain. E-mail address: [email protected] 0022-0736/$ – see front matter © 2009 Elsevier Inc. All rights reserved. doi:10.1016/j.jelectrocard.2009.01.006

Invasive methods have been used to study the relationship between QRS duration in the presence of left BBB (LBBB) and block location, and QRS duration was found to be directly related to block severity.3 For this reason, we evaluated 2 different methods for measuring QRS duration and also investigated whether there are significant differences when using the standard ECG or a body surface potential mapping (BSPM) system. Although the 12-lead ECG is the most widely used technique in cardiology, several studies have been conducted to determine whether the use of more leads would provide more information.4,5 Basically, BSPM increases the amount of information recorded in the region of the thorax6-8 because the information obtained is limited when only 6 precordial leads are available. In BSPM systems, 30 electrodes are commonly considered to account for most of the diagnostic information obtained from an ECG.9,10

652

V. Donis et al. / Journal of Electrocardiology 42 (2009) 651–659

We can regard QRS duration measured in BSPM recordings as more realistic than QRS duration obtained using 12-lead ECG because it is based on more data from the thorax. In addition, spatial and temporal information concerning the propagation of the heart's electrical signal is available from body surface potential maps. Hence, it is possible to observe the surface activation pattern in a specific subject, and this can be useful in the diagnosis of heart diseases. Previous studies of other heart diseases such as myocardial infarction (MI) have proved that a BSPM system can help to improve diagnosis.11-17 Kornreich et al11 proved that it was possible to classify MI patients into 2 groups (anterior and inferior) by analyzing BSPM features. In addition, it is possible to identify areas where the most significant features of a disease are patent,13,15 as for example, areas of the torso where the most significant ST changes most frequently occur in acute MI13 can be identified, or to obtain optimized lead systems for the diagnosis of some diseases.18-21 Obtention of BSPM recordings requires more time than a standard ECG recording, and moreover, physicians do not usually have BSPM systems available, but they have 12-lead equipments. However, BSPM systems can be used to find out which leads provide a better reconstruction of the body surface potentials22 or if there are new positions for the leads that make possible to observe an indication of a specific disease.23 Previous studies have obtained body surface maps of BBB patients and quantitatively described potential distribution patterns.24-27 However, these studies did not have any clinical repercussions, given that they did not suggest any improvement in BBB diagnosis. The present study aims to determine whether, with the information provided by electrodes located on the entire body surface, measured QRS duration is greater than QRS duration recorded with the 12 standard leads. We hypothesized that, given that the standard 12-lead ECG mostly reflects the activity of the left ventricle, the inclusion of leads on the right side of the torso may reflect the activation of the right ventricle more accurately and thus provide a more realistic measurement of QRS duration. On the other hand, we aimed at determining whether BSPM maps of subjects with the same pathologic characteristic present similarities and, in case it occurred, to compute representative maps of the pathologic condition. And also to use this information for detecting the presence of BBB with an automatic method based on map similarity.

Materials Recording system Our BSPM system was used to obtain the ECG recordings.21,28 It is a commercial 64-lead recording system for biopotential measurements (Active One; Biosemi, Amsterdam, the Netherlands). The sampling rate used was 2048 Hz, and the quantization rate was 1 microvolt per bit. Electrodes were distributed nonuniformly using a vest placed upon the chest, with 16 electrodes on the back and 48 on the front and higher electrode density at positions in the region over the heart (see Fig. 1).

Anterior

Posterior

Fig. 1. Arrangements of the electrodes used for BSPM registers. Electrode positions are represented by circles. From the 64 electrodes available, 48 are placed on the anterior part and 16 on the posterior part of the thorax.

Study population The database consisted of 1-minute ECG BSPM recordings from BBB patients and control subjects with normal ventricular conduction. A conventional 12-lead standard ECG was available for each subject. An echocardiogram was obtained to assess ventricular wall thickness. The BSPM recordings were obtained for 18 consecutive patients hospitalized in Hospital Clinico Universitario of Valencia (Valencia, Spain), aged 75.1 ± 8.6, including 11 men (see Table 1 for more details), with BBB as a primary or secondary diagnosis. Diagnosis was established based on the morphology of the QRS complex and its duration (LBBB, initial downward in V1 and no normal small Q wave in V6, terminally upward deflections in V5 and V6 leads; left anterior fascicular block [LAFB], positive deflection [initial R wave] in the inferior leads [II, III; aVF], small Q wave in leads I and aVL, R wave inscribed in leads I and aVL, S wave in the inferior leads, and left axis deviation of the QRS mean axis; right BBB [RBBB]_LAFB, broad terminal S wave in V5-V6, a double R wave in V1 to RSR' complex, and axis of the QRS deflection oriented superiorly with very small r/S ratio in lead II). Duration of QRS complex of BBB patients was higher than 120 milliseconds (except for the patient having LAFB). Left ventricular hypertrophy (VH) was determined as a wall thickness (N11 mm) in the echocardiography. The BSPM recordings of 9 control subjects, aged 25.5 ± 4.1, including 6 men, were acquired. Diagnosis was based on no history of cardiac disease, normal ECG morphology, no elevation of the ST segment, no axis deviation, and normal QRS duration. Subjects were divided into groups only depending on their BBB diagnosis. Characteristics of BBB patients in the study is listed in Table 1. Methods Signal processing Signals were processed using Matlab 7.0 (The Mathworks Inc, Natick, MA, USA). A high pass filter (cutoff frequency, 0.5 Hz) was used to eliminate baseline fluctuation caused by

V. Donis et al. / Journal of Electrocardiology 42 (2009) 651–659 Table 1 Characteristics of study subjects Control Age, y Male, n (%) BBB Age, y Male, n (%) Ejection fraction, % Syncope episodes, n

25.5 ± 4.1 6 (67) 75.1 ± 8.6 11 (61) 52.7 ± 10.9 2

detections and to avoid possible errors introduced by the algorithm in QRS duration measurements. To reduce the offset voltage, which is produced at the skin-electrode interface, the signal before P-wave onset and after T-wave offset, which is the signal within electrically inactive segments, was averaged for each averaged beat and subtracted from the total signal value for the averaged beat. QRS duration measurement

Risk factorsa Smoking habit, n Arterial hypertension, n Diabetes mellitus type 2, n Obesity, n Hypercholesterolemia, n

9 8 3 6 7

Heart diseasea Left VH, n Ischemic cardiomyopathy, n Atrial fibrillation, n Valvular insufficiency, n Complete LBBB, n Complete RBBB_LAFB, n Complete LBBB_VH, n LAFB, n

1 6 4 4 13 3 1 1

a

653

More than one possible.

breathing. A low pass filter was also used (cutoff frequency, 80 Hz) to avoid the high frequency noise caused by electromyogram interferences. Leads presenting a 50-Hz component with a voltage higher than 1% of the total lead voltage were filtered using a Notch filter. Fig. 1 illustrates the placement of the 64 electrodes. Twelve standard leads were selected from the 64 leads. Specifically, the 12 standard leads were obtained from signals recorded on the 9 electrodes closest to the 12-lead ECG standard electrode position. Signals were visually inspected to check that the morphology of the 12-lead standard ECG was exactly the same than the 12 leads computed from the BSPM system. Averaged cardiac cycles were obtained for each lead in every patient. Firstly, QRS peaks were detected using a modified version of Tompkins algorithm,29 and the mean value of the RR interval was calculated. A window including a number of RR mean dependent samples was defined around each complex. The window included 42% of the mean RR before the peak and 50% after the peak. After that, the median beat was obtained for each lead by computing the median voltage at each instant in the beats of each lead. Beats presenting a correlation value with the median beat lower than 0.8 were discarded to reject ectopic beats as well as noise fragments. Nondiscarded beats were averaged to obtain the averaged beat for each lead. All the averaged beats were visually inspected to discard any noisy averaged beat. Fiducial points corresponding to the beginning (onset) and the end (offset) of the QRS interval, and to the onset of the P wave and the offset of the T wave, were detected in each averaged beat, using a semiautomatic method. An algorithm based on ployline splitting 30 was used to obtain the detections. Human involvement was necessary to verify the

Two different methods were used to measure QRS duration. Standard QRS duration (QRSD) was defined as the longest QRS duration measured for the different averaged beats available. QRS maximum duration (QRSMD) was defined as starting at the earliest QRS onset detected and ending at the latest QRS offset detected. Both methods depend on the lead system under consideration. QRSD and QRSMD are defined in Eqs. (1) and (2), where N is the number of leads. QRSD = maxðQRSoffseti  QRSonseti Þ; ia1; 2; :::N

ð1Þ

  QRSMD = maxðQRSoffseti Þ  min QRSonsetj ; ia1; 2; :::N ; j a1; 2; :::; N

ð2Þ

QRSD and QRSMD were computed for the 12-lead ECG and the 70 averaged beats corresponding to each subject. The 70-lead system included the 64 leads corresponding to our BSPM system and the standard limb and augmented leads. Results for both measurements (QRSD and QRSMD) were compared for the different lead systems mentioned above and measurement differences were computed for each subject. Mean values were calculated for each group, and SDs were calculated for each group with more than one subject. Obtaining QRS representative averaged beats for each group We computed the 64 and 12 representative averaged beats for each group of subjects consisting of more than 5 members (LBBB and control subjects). We used 64 representative averaged beats to obtain representative maps of each group and compare their respective maps with the maps of individual subjects. The 12 representative averaged beats of each group were used to compare them with the averaged beats of individual subjects. To compute the representative averaged beats of a group, the 64 averaged beats for all the patients in the group had to be of the same length, and to homogenize the length of the beats, a linear interpolation was applied. Once having equal length, the 64 averaged beats for all the subjects in the group were averaged. Fiducial points (QRS onset and offset) were calculated for each representative averaged beat as a reference for further processing of BSPM maps of each group. QRS 12-lead averaged beats were computed from the 64 representative averaged beats.

654

V. Donis et al. / Journal of Electrocardiology 42 (2009) 651–659

Body surface potential maps Averaged QRS onset and offset were computed for generating BSPM maps. They were defined as shown in Eqs. (3) and (4), where N is the total number of leads in the lead system. They were computed for the averaged beats of each patient and each group. QRSonset =

N 1X QRSonseti N i=1

ð3Þ

QRSoffset =

N 1X QRSoffseti N i=1

ð4Þ

To obtain the potential maps, the averaged QRS onset and offset were regarded as the beginning and end of the QRS complex. Fourteen body surface potential maps, equally spaced along the QRS complex, were computed. Maps were generated using the 64 averaged beats of each individual and the 64 representative averaged beats of each group, interpolating the potential on the points of the surface not sampled by our electrode grid by means of cubic splines.31 The BSPM maps obtained for each group are considered isopotential mean maps, which are obtained from averaged beats computed from the information of the different subjects belonging to the group. Comparison of body surface potential maps Body surface potential maps for each individual were compared both with the maps of the group to which the subject belonged (own group) and with the maps of other groups. As the number of patients is limited in this study, to counteract any patient bias in the comparison, we implemented a comparison algorithm similar to the leave-one-out technique—the maps of the own group were computed from the averaged beats generated by using the averaged beats of all the individuals in the group, except for the patient under study. The maps of other groups were computed from the averaged beats generated by using the averaged beats of all subjects belonging to a particular group. Map comparison was performed by 2-dimensional correlation as defined in Eq. (5) where r is the correlation coefficient, A and B are matrix variables containing potential values of the body surface for an specific instant of the QRS complex (A, potentials representative of a group; B, potentials of an individual subject), A and B are the mean value of A and B, respectively, and m ∈ 1…M and n ∈ 1…N, where M is the number of rows (91) and N the number of columns (141) of the matrices. For each group and individual, the first 2 maps computed at the beginning of the QRS complex and the last 2 computed at the end were not considered for comparison purposes because of their low signal-to-noise ratio, owing to the low amplitude of the signal.   P P Amn  A Bmn  B m n r = rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi P P  2 P P  2  Amn  A Bmn  B m

n

m

n

ð5Þ

Correlation indexes obtained from comparisons made between the 10 maps of each subject and the 10 maps of each group were averaged, providing a single correlation value for each subject. A subject was considered to belong to a specific group if the averaged correlation index obtained when comparing his maps with the maps of the group was higher than 0.7 (0.7 was established as the optimal threshold for correct classification empirically). When the group maps were compared with maps of a specific group of subjects, the mean and SD of the averaged correlation indexes for the subjects in the group was computed. Comparison of 12-lead ECG averaged beats As QRS 12-lead representative averaged beats were obtained from 64 representative averaged beats, the comparison of 12-lead averaged beats was also made by using the leave-one-out technique. Comparison was performed by a correlation as defined in Eq. (5), where A and B are vectors (m = 1) that contain the potential values of 1 of the 12 averaged beats from a group (A) or a subject (B). Averaged beats of each group were compared with the corresponding averaged beats of each subject. A subject was considered to belong to a group when the correlation coefficient obtained averaging the correlation coefficients of the 12 comparisons was superior to 0.7. Statistical analysis Statistical analysis for comparison of QRS duration measurements was performed by using SPSS (SPSS, Chicago, IL). Results are expressed as mean and SD. A paired Student t test was used for comparisons between QRSMD and QRSD for each lead system and for each QRS duration criteria. An independent sample t test was used to compare measurements between different groups of patients. Two significance levels were used for this 2-tailed test, P b 0.05 and P b 0.01. A P value higher than 0.05 was regarded as nonsignificant (ns). To assess the classification based on map comparisons and 12 averaged beats comparisons, accuracy, sensitivity, and specificity were computed for LBBB patients and control subjects. Table 2 Mean values ± SDs of QRS durations (in milliseconds) measured, according to 2 different definitions, in controls and in various subgroups of BBBs Ventricular conduction category

12-lead ECG QRSD

QRSMD

Control (n = 9) All BBBa (n = 18) LBBB (n = 13) RBBB_LAFB (n = 3)

104.8 ± 7.3 172.3 ± 29

108.7 ± 7.5 112.0 ± 9 118.1 ± 10.2 177.1 ± 28.4 175.5 ± 25.4 182.3 ± 25.3

a

70-lead ECG QRSD

QRSMD

184.1 ± 21.7 188.3 ± 21.4 185.1 ± 21 191.4 ± 20.4 155.0 ± 14.4 163.1 ± 9.8 160.5 ± 12.8 170.3 ± 12.9

Includes one ECG with LBBB and left VH and one with LAFB.

V. Donis et al. / Journal of Electrocardiology 42 (2009) 651–659

655

Results QRS duration measurement Results of QRS duration measurements are shown in Table 2. This table shows that results obtained measuring QRSMD are not very different from the results obtained measuring QRSD. The greatest difference appears for RBBB_LAFB patients in the 70-lead system—170.3 vs 160.5 milliseconds (P = ns). QRS duration measured with the 70 or the 12-lead system is not significantly different for LBBB patients. However, this does not occur with controls and RBBB_LAFB patients. Differences for controls are 112.0 vs 104.8 milliseconds for QRSD (P b 0.01) and 118.1 vs 108.7 milliseconds for QRSMD (P b 0.01) and for RBBB_LAFB are as follows: 160.5 vs 155.0 milliseconds for QRSD (P = ns) and 170.3 vs 163.1 milliseconds for QRSMD (P = ns). The largest differences are observed when comparing QRSMD measured with the 70-lead ECG and QRSD measured with the 12-lead ECG for controls and RBBB_LAFB subjects as follows: 118.1 vs 104.8 milliseconds for controls (P b 0.01) and 170.3 vs 155.0 milliseconds for RBBB_LAFB patients (P = ns). Fig. 2 shows the variation for each group of subjects when measuring the QRS duration of their patients with the QRSD method and the 12-lead system vs the QRSMD and the

Fig. 3. Potential averaged maps obtained for five different time instants of the QRS complex are shown: A, control body surface and B, LBBB body surface representative maps. Lead V1 is depicted in the lowermost panel. Time instants selected for representation of BSPM maps are indicated by numbers on lead V1. From the ten maps available for each case, maps represented are: second (1), fourth (2), sixth (3), eighth (4), and tenth map (5). In each map, the left section represents the anterior thorax while the right section represents the posterior thorax. Each dot represents an electrode position. The dotted line represents zero potential. The solid line represents positve potentials. The dashed line represents negative potentials. Maximum potential is indicated by the "+" symbol and minimum potential by the "-" symbol.

70-lead system. Differences between QRS durations are not homogeneous among different patients of a same group. Map comparisons

Fig. 2. Variations between QRSD (written in versales format) computed for the 12-lead ECG and QRSMD computed for the 70 leads. A, control subjects; B, LBBB patients; and C, RBBB_LAFB patients. The parameter c is the correlation coefficient; (*p b 0.01).

As shown in Fig. 3, there are visible differences related to surface activation sequences between an LBBB and a heart with normal ventricular conduction. In control subjects, it can be seen that in the first fifth of the QRS complex (see Fig. 3A panel 1), depolarization starts on the left (anterior) side of the chest, as it could be expected because normal depolarization starts on the left side of the septum. After that, maximum potential moves to a lower position but

656

V. Donis et al. / Journal of Electrocardiology 42 (2009) 651–659

remains on the left side. It is known that in a healthy heart, the apex and also the left ventricular wall depolarize after the septum. In the remaining instants, depolarization seems to move toward the back of the heart. In a healthy subject, the last part of the heart to be depolarized is the base and the back of the ventricles. However, in LBBB representative maps, the depolarization process is different. In the first fifth of the QRS complex (see Fig. 3B panel 1), we can see that maximum potential is recorded on the left axillary line at the level of the fifth intercostal space, which can be explained by the first initial depolarization of the lower part of the right septum and subsequent propagation from the right septum to the left ventricle. Finally, slow activation of the left ventricle keeps maximum potential at the back of the thorax but moves it slightly to the right.

Fig. 4. Body surface potential maps obtained for different time instants of the QRS complex are shown: panel A, RBBB_LAFB maps and panel B, LAFB maps. See Figure 3 caption.

Table 3 Results obtained from map comparisons Representative maps

Comparison group

LBBB (n = 13) Control (n = 9) RBBB_LAFB (n = 3) LBBB_VH LAFB

LBBB averaged

Control averaged

0.85 ± 0.06 0.12 ± 0.10 −0.32 ± 0.18

0.24 ± 0.09 0.82 ± 0.10 −0.24 ± 0.10

0.62 0.66

0.45 0.26

N is the number of patients in each group with more than one patient. The table includes mean correlation values of LBBB_VH and LAFB patients, and the average mean correlation index for the patients in each group (n N 1) and its SD.

As can be seen in Fig. 4, different kinds of block have different BSPM patterns. The RBBB_LAFB maps show that depolarization starts at the back and left lower side of the thorax (see Fig. 4A panel 1), which is coherent with activation via the left posterior fascicle—the only fascicle that is not blocked. This fascicle is posteroinferiorly located. Therefore, the first activation can be observed on the left posteroinferior side of the chest. The following maps show the activation process of the remaining myocardium progressing toward the anterosuperior region of the left ventricle and moving to the right side because the final depolarization of the left ventricle coincides with the delayed onset of right ventricle depolarization. Finally, we can see that initial depolarization of the ventricles in the LAFB patient (Fig. 4B panel 1), as seen on the body surface, is similar to the initial depolarization of the ventricles in RBBB_LAFB patients, but after this initial similarity, the rest of the depolarization process is different, given that the right ventricle in LAFB patients depolarizes at the beginning and the opposite is the case in the RBBB_LAFB patient. Results for different automatic map comparisons are shown in Table 3. When LBBB representative maps are compared to the maps of individuals having this pathologic

Fig. 5. Correlation indexes resulting from comparison between representative or averaged maps (Av) for the two groups (LBBB and control subjects) and individuals in each group (Sb).

V. Donis et al. / Journal of Electrocardiology 42 (2009) 651–659 Table 4 Results obtained from 12-lead averaged beats comparisons Representative averaged beats

Comparison group

LBBB (n = 13) Control (n = 9) RBBB_LAFB (n = 3) LBBB_VH LAFB

LBBB averaged

Control averaged

0.74 ± 0.14 0.21 ± 0.13 −0.27 ± 0.25

0.25 ± 0.09 0.71 ± 0.12 −0.41 ± 0.13

0.6013 0.6846

0.6768 0.2181

N is the number of patients in each group with more than one patient. The table includes mean correlation values of LBBB_VH and LAFB patients, and the average mean correlation index for the patients in each group (n N 1) and its SD.

condition, the averaged correlation index is 0.85 ± 0.06, but when they are compared with the maps of subjects belonging to other groups, the correlation index is lower—0.12 ± 0.10 for control subjects and −0.32 ± 0.18 for RBBB_LAFB subjects. And when LBBB representative maps are compared with the maps of groups consisting of only one subject, the mean correlation index is 0.62 for the patient having LBBB_VH and 0.66 for the LAFB patient. Comparisons made using control representative maps show that a higher correlation index (0.82 ± 0.10) is obtained when comparisons are made with control subjects. Correlation indexes are lower when comparisons are made with LBBB patients (0.24 ± 0.09) or RBBB_LAFB subjects (−0.24 ± 0.10). When these representative maps are compared with LBBB_VH maps, the correlation index is 0.45, and with LAFB maps, it is 0.26. As shown in Fig. 5, when LBBB representative maps (calculated without including the LBBB patient under evaluation) are compared with maps of LBBB patient correlation index ranges from 0.77 to 0.95, and when the representative maps obtained from the 13 LBBB patients are compared with maps of control subjects, it ranges from −0.03 to 0.24. On the other hand, when representative maps of control subjects (again obtained without including the subject under evaluation) are compared with those of control subjects, the correlation index ranges from 0.74 to 0.89 in most instances. Finally, when representative maps obtained using the information from the 9 control subjects are compared with the maps of LBBB patients, the correlation index ranges from 0.12 to 0.37. The results of sensitivity and specificity tests are shown in Table 5. It can be seen that for LBBB subjects sensitivity is 93% and specificity is 100%, whereas sensitivity for controls is 89% and specificity is 100%. Accuracy for both LBBB and control groups is 96.3%. Twelve-lead ECG averaged beats comparisons Averaged beats of the 12-lead ECG representative of each group when compared with each individual subject provided the results summarized in Table 4. It can be observed from Table 4 that, in general, only when comparing the representative averaged beats of one group with the subjects belonging to the group the correlation index is high.

657

From Table 5, it can be observed that sensitivity for comparisons with 12-lead averaged beats for the LBBB group is 76.9% and for controls is 66.6%. Specificity is 100% for both groups and accuracy is 89%. Discussion QRS duration Higher QRS duration values have been proven to be associated with higher cardiovascular death risk.32 Specifically, it has been shown that cardiac death risk in the presence of RBBB correlates with an increase in QRS duration. An increase in the duration of the QRS complex of 10 milliseconds in RBBB patients increases the risk of cardiac death by 26.6%,33 suggesting that subtle differences in QRS duration are not negligible in the diagnosis of these patients. Other studies have established that LBBB and prolonged QRS durations predict mortality in patients with heart failure.34,35 Because QRS duration is an important parameter in BBB diagnosis and also for determining the severity of the blockade, this study has partly focused on measuring QRS duration using different methods to assess whether with either of them the depolarization duration measured provides more accurate results. We evaluated the differences in measured QRS duration using 2 different methods and different numbers of leads. Both when using different methods or different number of leads, there are little differences on QRS duration for LBBB patients. However, differences are higher for RBBB_LAFB and control subjects when using different lead systems. The group that shows the biggest difference when using different methods is RBBB_LAFB group. The 12-lead ECG provides quite realistic values for QRS duration which approximate values obtained with the 70lead system. The 12-lead ECG is particularly appropriate for LBBB patients, an observation that can be explained by the positioning of the precordial leads on the left side of the chest. However, 70-lead ECG provides more information than 12-lead ECG for controls or RBBB_LAFB patients. This may be caused by the observation in this subjects of more detail of the information coming from the right ventricle, given that the 12-lead ECG does not provide so much information from this part of the thorax. Consequently, the number of leads and the method used for measuring QRS duration may have considerable impact on the diagnosis of a patient having BBB because differences can exceed 10 milliseconds.

Table 5 Specificity, sensitivity, and accuracy values for the LBBB and control groups 12-lead ECG

Sensitivity Specificity Accuracy

Maps comparison

LBBB

Control

LBBB

Control

76.9% 100% 89%

66.6% 100% 89%

93% 100% 96.3%

89% 100% 96.3%

658

V. Donis et al. / Journal of Electrocardiology 42 (2009) 651–659

Analysis of BSPM maps The BSPM maps have been processed using different techniques such as isopotential,11 isointegral,36,37 departure,36 discriminant,11 laplacian,38 or scattering maps.11 Our study has focused on the analysis of isopotential maps. Although isopotential mean maps have been already described,11 they have never been used to directly classify subjects. The classification of patients in previous studies has mostly been done obtaining several descriptors of BSPM maps.11-13 Our study proposes a new method for classifying patients, specifically BBB patients, without the need to compare parameters extracted from maps but directly comparing isopotential mean maps for each group with sequential potential maps for each subject. Although methods for beat classification have been proposed previously, as for instance, methods based on neuronal networks39 that can provide a sensitivity of 95% for LBBB patients, no method proposed classification of BBB patients using BSPM information. In this study, BBB BSPM maps have been obtained that are consistent with previous studies.24,25 We have demonstrated that body surface potential maps of subjects having BBB are, to some extent, repeatable and that these patients can be diagnosed by automatic comparison of their BSPM maps. As mentioned above, when representative maps of LBBB patients are compared with maps of individuals affected by this pathologic condition, the mean correlation index is 0.85, significantly higher than correlation indexes resulting from comparisons with control subjects and RBBB_LAFB patients, which are close to zero. As it might be expected, the correlation index is high when comparing LBBB maps with maps of a patient with both LBBB and VH. Moreover, when LBBB maps are compared with the maps of the LAFB patient, the correlation index is 0.66. As for representative maps of control subjects, the averaged correlation index is high only when comparisons are made with maps from control subjects (0.82) and lower than 0.5 otherwise. Although other BSPM studies24-27 focused on BBB only used BSPM information to describe the different patterns for different kinds of block or some of their characteristics, the results of the aforementioned comparisons could be useful in improving the diagnosis of BBB. Sensitivity and specificity for LBBB patients are 93% and 100%, respectively, making this method a feasible approach for automatically classifying these patients into LBBB group only when they are having this pathologic condition. Although the 12-lead ECG may provide enough information to classify these patients, diagnosis depends on the expert cardiologist. This study proposes an automatic and quantitative method to diagnose BBB by BSPM information. Comparison of 12-lead averaged beats provides not as good results as maps comparison. The BSPM maps comparison improves sensitivity of LBBB patients in 16.1% and control subjects in 22.4%. The main limitation for using BSPM systems is the time that it takes to obtain the recordings (about 10 minutes) and that nowadays there are not BSPM systems available in

every hospital or health care center. However, BSPM systems help to understand the ECG in presence of BBB and to consider how appropriate the 12-lead ECG is. In addition, it provides information about ventricular propagation reflected in the body surface of patients with the same kind of block, and this permits to assess their similarities and to obtain an automatic classification of the patients. Limitations and future work The number of BBB patients included in this study was 18, and it differs for each kind of block. It would be interesting to confirm the findings of this study by including a greater number of patients, especially in the case of the RBBB_LAFB group, which included 3 patients, and of RBBB patients. The low number of RBBB_LAFB patients included in this study is a consequence of the lowest incidence of hospitalized patients having this disease at the time of the recruitment of patients. Also, age of subjects under study is not homogeneous, existing significant differences between the age of controls and the age of BBB patients. In addition, our low number of patients might also have repercussions on map comparisons and the selection of the correlation index threshold, which was established to ensure high sensitivity and specificity values for the subjects under study. With regard to our digitalized 12-lead ECG obtained from BSPM recordings, maybe it is not totally accurate because of the electrode positioning. However, the morphology was consistent with their printed ECGs obtained with standard ECG recorders, and the slight variations introduced in precordial lead placement may be comparable to aberrant errors produced by the human factor. In the future, we would like to compare direct cardiac activation with BSPM information to determine the real accuracy of BSPM classifying method. Also, we are studying to use BSPM to measure the lack of synchrony in patients with heart failure.40 Although it may be impractical to obtain maps to diagnose BBB, it may be possible, and in future studies, it could be checked, to use maps obtained from a fewer number of leads (standard or not) to obtain an automatic diagnosis. In future works, it could be also assessed if this technique could be helpful in the diagnosis of other heart pathologic conditions. Conclusions QRS duration is not greatly increased by changing either the method used to measure duration or the number of exploring electrodes used. However, RBBB_LAFB and normal conduction activation patterns appear to be more accurately monitored when using a BSPM system and QRSMD method than with the 12-lead ECG and QRSD method, permitting the durations observed to be increased by more than 10 milliseconds. Obtaining averaged maps for a specific type of BBB or for subjects with normal ventricular conduction makes it

V. Donis et al. / Journal of Electrocardiology 42 (2009) 651–659

possible to classify a given subject in the appropriate group automatically. Consequently, this represents an automatic method to diagnose BBB patients.

19.

Acknowledgments This study received funding from the Spanish Ministry of Education and Science, as part of the TEC2005-08401 grant scheme, and the Universidad Politécnica de Valencia (UPV) through its research initiative program. The R + D + i Linguistic Support Service at the UPV offered support in the correction of the manuscript. References 1. Lilly Leonard S. Pathophysiology of heart disease. Baltimore, MD, USA: Lippincott Williams & Wilkins; 2003. 2. Castellano C, Pérez de Juan MA, Espinosa JS. Electrocardiografía clínica. Madrid, Spain: Mosby-Doyma Libros; 1996. 3. Auricchio A, Fantoni C, Regoli F, et al. Characterization of left ventricular activation in patients with heart failure and left bundlebranch block. Circulation 2004;109:1133. 4. Lux RL. Electrocardiographic body surface potential mapping. Crit Rev Biomed Eng 1982;8:253. 5. Taccardi B, Punske BB, Lux RL, et al. Useful lessons from body surface mapping. J Cardiovasc Electrophysiol 1998;9:773. 6. Maynard SJ, Menown IBA, Manoharan G, Allen J, Anderson JMcC, Adgey AAJ. Body surface mapping improves early diagnosis of acute myocardial infarction in patients with chest pain and left bundle branch block. Heart 2003;89:998. 7. Bruns H, Eckardt L, Vahlhaus C, et al. Body surface potential mapping in patients with Brugada syndrome: right precordial ST segment variations and reverse changes in left precordial leads. Cardiovascular Research 2002;54:58. 8. Hisamatsu K, Kusano KF, Morita H, et al. Usefulness of body surface mapping to differentiate patients with Brugada syndrome from patients with asymptomatic Brugada syndrome. Acta Med Okayama 2004;58: 29. 9. Kors JA, van Herpen G. How many electrodes and where? A “poldermodel” for electrocardiography. J Electrocardiol 2002;35:7. 10. Hoekema R, Uijen G, van Oosterom A. The number of independent signals in body surface maps. Methods Inf Med 1999;38:119. 11. Kornreich F, Montague TJ, Kavadias M, et al. Qualitative and quantitative analysis of characteristic body surface potential map features in anterior and inferior myocardial infarction. Am J Cardiol 1987;60:1230. 12. Kornreich F, Montague TJ, Rautaharju PM. Identification of first acute Q wave and non-Q wave myocardial infarction by multivariate analysis of body surface potential maps. Circulation 1991;84:2442. 13. Kornreich F, Montague TJ, Rautaharju PM. Body surface potential mapping of ST segment changes in acute myocardial infarction. Circulation 1993;87:773. 14. Medvegy M, Préda I, Savard P, et al. New body surface isopotential map evaluation method to detect minor potential losses in non-Q-wave myocardial infarction. Circulation 2000;101:1115. 15. Hänninen H, Takala P, Mäkijärvi M, et al. ST-segment level and slope in exercise-induced myocardial ischemia evaluated with body surface potential mapping. Am J Cardiol 2001;88:1152. 16. Boudik F, Anger Z, Aschermann M, Vojácek J, Tomecková M. Dipyridamole body surface potential mapping: noninvasive differentiation of syndrome X from coronary artery disease. J Electrocardiol 2002;35:181. 17. Guillem MS, Millet J, Bodi V, Chorro FJ. Q wave myocardial infarction analysed by body surface potential mapping. Comput Cardiol 2004;31:725. 18. Kornreich F, Montague TJ, Rautaharju PM, Block P, Warren JW, Horacek MB. Identification of best electrocardiographic leads for

20.

21.

22.

23.

24.

25.

26.

27.

28.

29. 30. 31.

32.

33.

34. 35.

36.

37.

38.

39.

40.

659

diagnosing anterior and inferior myocardial infarction by statistical analysis of body surface potential maps. Am J Cardiol 1986;58: 863. Kornreich F, Montague TJ, Rautaharju PM, Kavadias M, Horacek MB. Identification of best electrocardiographic leads for diagnosing left ventricular hypertrophy by statistical analysis of body surface potential maps. Am J Cardiol 1988;62:1285. Horácek BM, Warren JW, Penney CJ, et al. Optimal electrocardiographic leads for detecting acute myocardial ischemia. J Electrocardiol 2001;34(Suppl):97. Guillem MS, Castells F, Climent AM, Bodí V, Chorro FJ, Millet J. Evaluation of lead selection methods for optimal reconstruction of body surface potential. J Electrocardiol 2008;41:26. Lux RL, Smith CR, Wyatt RF, Abildskov JA. Limited lead set selection for estimation of body surface potential maps in electrocardiography. IEEE Trans Biomed Eng 1978;25:270. Nademanee K, Priori SG, Towbin JA, Brugada P, et al. Proposed diagnostic criteria for the Brugada syndrome: consensus report. Circulation 2002;106:2514. Sohi GS, Flowers NC. Body surface map patterns of altered depolarization and repolarization in right bundle branch block. Circulation 1980;61:634. Sohi GS, Flowers NC, Horan LG, Sridharan MR, Johnson JC. Comparison of total body surface map depolarization patterns of left bundle branch block and normal axis with left bundle branch block and left-axis deviation. Circulation 1983;67:660. Liebman J, Rudy Y, Diaz P, Thomas CW, Plonsey R. The spectrum of right bundle branch block as manifested in electrocardiographic body surface potential maps. J Electrocardiol 1984;17:329. Pastore CA, Tobias N, Samesima N, et al. Body surface potential mapping investigating the ventricular activation patterns in the cardiac resynchronization of patients with left bundle-branch block and heart failure. J Electrocardiol 2006;39:93. Bodí V, Sanchis J, Guillem MS, et al. Analysis of the extension of Q-waves after infarction with body surface map: relationship with infarct size. Int J Cardiol 2006;111:399. Pan J, Tompkins WJ. A real-time QRS detection algorithm. IEEE Trans Biomed Eng 1985;32:230. Jain R, Kasturi R, Schunck BG. Machine vision. New York, NY, USA: Mc Graw Hill; 1995. p. 196. Schijvenaars BJ, Kors JA, van Herpen G, Kornreich F, van Bemmel JH. Interpolation of Body Surface Potential Maps. J Electrocardiol 1995;28:104. Desai AD, Yaw TS, Yamazaki T, Kaykha A, Chun S, Froelicher VF. Prognostic significance of quantitative QRS duration. Am J Med 2006;119:600. Adesanya CO, Yousuf KA, Co C. Is wider worse? QRS duration predicts cardiac mortality in patients with right bundle branch block. Ann Noninvasive Electrocardiol 2008;13:165. Padeletti L, Giaccardi M, Turreni F. Influence of QRS prolongation on the natural history of CHF. E Heart J Suppl 2004;6:79. Iuliano S, Fisher SG, Karasik PE, Fletcher RD, Singh SN. QRS duration and mortality in patients with congestive heart failure. Am Heart J 2002;143:1085. Tonooka I, Kubota I, Watanabe Y, Tsuiki K, Yasui S. Isointegral analysis of body surface maps for the assessment of location and size of myocardial infarction. Am J Cardiol 1983;52:1174. Montague TJ, Witkowski FX. The clinical utility of body surface potential mapping in coronary artery disease. Am J Cardiol 1989;64:378. Tsai H, Wu DS, Ceccoli H, et al. Localization of chronic myocardial infarction using body surface Laplacian maps. Comput Cardiol 1998;25:513. Dokur Z, Ölmez T. ECG beat classification by a novel hybrid neural network. Istanbul Comput Methods Programs Biomed 2001; 66:167. Guillem MS, Brugada R, Thibault B, Climent AM, Millet J. Analysis of body surface potential maps in cardiac resynchronization therapy. Comput Cardiol 2008;24.