Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
com
Abstract Introduction: Our primary objective was to ascertain which commonly used 12-to-Frank-lead
transformation yields spatial QRS-T angle values closest to those obtained from simultaneously
collected true Frank-lead recordings.
Materials and Methods: Simultaneous 12-lead and Frank XYZ-lead recordings were analyzed for 100
postmyocardial infarction patients and 50 controls. Relative agreement, with true Frank-lead results, of
12-to-Frank-lead–transformed results for the spatial QRS-T angle using Kors' regression versus inverse
Dower was assessed via analysis of variance, Lin's concordance, and Bland-Altman plots.
Results: Spatial QRS-T angles from the true Frank leads were not significantly different than
those derived from the Kors' regression-related transformation but were significantly smaller than
those derived from the inverse Dower-related transformation (P b .001). Independent of method,
spatial mean QRS-T angles were also always significantly larger than spatial “maximum”
(“peaks”) QRS-T angles.
Discussion: Spatial QRS-T angles are best approximated by regression-related transforms. Spatial
mean and spatial “peaks” QRS-T angles should not be used interchangeably.
Published by Elsevier Inc.
Keywords: Vectorcardiography; Spatial ventricular gradient; 3-Dimensional ECG; Lead reconstruction
Table 1
Results for the control group, post-MI group, and total sample for the true Frank leads and for the Kors' regression-related and inverse Dower-related
transformations
Parameter Control group (n = 50) Post-MI group (n = 100) Total sample (n = 150)
SM QRS-T angle (°)-true Frank leads 54.4 ± 25.8 84.3 ± 39.8 74.3 ± 38.3
SM QRS-T angle (°)-Kors' regression 53.4 ± 26.0 91.4 ± 39.4 78.8 ± 39.7
SM QRS-T angle (°)-inverse Dower 70.0 ± 28a,b 98.3 ± 40.9a 88.9 ± 39.4a,b
SP QRS-T angle (°)-true Frank leadsc 41.7 ± 27.1 68.9 ± 40.9 59.8 ± 38.9
SP QRS-T angle (°)-Kors' regressionc 40.7 ± 23.5 76.4 ± 42.1 64.5 ± 40.6
SP QRS-T angle (°)-inverse Dowerc 58.4 ± 27.8a,b 87.9 ± 45.7a 78.1 ± 42.9a,b
SVG magnitude (mV ⁎ ms)-true Frank leads 98.3 ± 44.7 46.7 ± 22.0 63.9 ± 39.7
SVG magnitude (mV ⁎ ms)-Kors' regression 101.9 ± 44.1 51.6 ± 25.2 68.4 ± 40.3
SVG magnitude (mV ⁎ ms)-inverse Dower 97.3 ± 44.3 54.0 ± 28.0 68.4 ± 39.8
SVG elevation (°)-true Frank leads 32.1 ± 11.0 28.3 ± 33.6 29.2 ± 28.2
SVG elevation (°)-Kors' regression 32.6 ± 9.9 24.7 ± 29.1 27.3 ± 24.7
SVG elevation (°)-inverse Dower 28.5 ± 11.4 20.1 ± 26.9 22.9 ± 23.2a
SVG azimuth (°)-true Frank leads 20.0 ± 21.7 −8.1 ± 62.3 1.3 ± 53.9
SVG azimuth (°)-Kors' regression 19.9 ± 17.6 −0.5 ± 64.7 6.3 ± 54.5
SVG azimuth (°)-inverse Dower 11.5 ± 20.0 −5.4 ± 70.9 0.3 ± 59.4
a
Derived VCG result significantly differs from true Frank-lead result by ANOVA.
b
Derived VCG results significantly differ from one another by ANOVA.
c
SP QRS-T angle results significantly differ from SM QRS-T angle results.
vectors defined by the areas of these same loops (spatial analyses on data from: (1) PTB patients 001 through 101, all
“mean” QRS-T angle4-11,17-22,26-29—see Appendix 1 for of whom were being evaluated for a recent myocardial
our own understanding of the difference). Even more infarction (MI) and (2) the first 50 “healthy controls” in the
confusingly, healthy subjects have also typically had higher PTB database, beginning with PTB subject 0104, as
mean values for the spatial “mean” QRS-T angle than for the specified by the PTB demographic files that accompany
spatial “maximum” QRS-T angle (means for the latter the recorded raw data. In the recent MI group, the mean (SD)
ranging from 11 to 21 degrees for inverse Dower age was 58.8 (11.3) years, and 72% were males. In the
transformations1,3,23 and from 42 to 51 degrees for the true healthy control group, the mean (SD) age was 43.5 (14.7)
Frank leads).24,25 In the present study, we used Lin's years, and 76% were males.
concordance correlation coefficient to ascertain the relative
concordance, to true Frank-lead results, of 12-to-Frank- Data analyses
lead–transformed results derived from Kors' regression- The raw binary data files from the PTB database were
related versus inverse Dower-related reconstructions. Based processed using software developed by the authors at
on the comparatively smaller mean quadratic deviations NASA's Johnson Space Center.20,31 Initial analyses revealed
from the true XYZ leads that have been noted for the Kors' that all of the selected files except recent MI-patient file 079
regression method in the past,16 we hypothesized that this (which demonstrated a paced rhythm and was not analyz-
method would also better approximate important secondarily able) had, at a minimum, 40 QRS-T complexes that were
derived VCG parameters such as the spatial QRS-T angle. acceptable for signal averaging in all channels when using a
We also sought to better define any differences between minimum cross-correlation cutoff of 97% against the signal-
results for the SM versus spatial “maximum” (“peaks”) QRS- averaged QRS templates formed for each channel in each
T angles using both the true Frank leads and the 2 different file, as previously described.20,31 Thus, 40-complex signal
reconstruction methods. averages were ultimately constructed for each patient's file,
the principal purpose of signal averaging being to help
eliminate any transient or nonreproducible effects that would
Methods more likely influence single complexes than signal averages,
such as the precise location of a given complex within the
Data collection respiratory cycle.
The data were obtained from a publicly available source,
VCG parameters from the true and derived Frank leads
the Physikalisch-Technische Bundesanstalt (PTB) Diagnos-
tic ECG Database30 available at http://www.physionet.org/ The principal VCG parameters studied were the SM and
physiobank/database/ptbdb/. The PTB ECG data were spatial “peaks” (SP) QRS-T angles, the magnitude of the
collected in the 1990s by Dr Michael Oeff et al at the spatial ventricular gradient (SVG), and the elevation and
Department of Cardiology of University Clinic Benjamin azimuth angles of the SVG. Time integrals (areas) for the X,
Franklin in Berlin, Germany, using a noncommercial Y, and Z signals were first determined separately for the
prototype recorder that allowed the simultaneous acquisition QRS and T complexes by measuring the areas of the
of both 12-lead ECG and true Frank-lead VCG data stored at complexes above and below the baseline and subtracting
1000 samples per second per channel. We focused our own negative areas from positive areas. The time integrals of the
304 D.L. Cortez, T.T. Schlegel / Journal of Electrocardiology 43 (2010) 302–309
Fig. 1. Representative 3-dimensional vectorcardiographic loops with accompanying diagrammatic SP QRS-T angles for 1 healthy subject. R indicates right;
L, left, A, anterior. The arrows shown (heavily dashed for the QRS loop, lightly dashed for the T loop) express only the vector direction from the origin to the
peak of the given loop, not the vector magnitudes. The SP QRS-T angle from the true Frank leads (A) was similar to that obtained from the Kors' regression
method (B) but notably smaller than that obtained from the inverse Dower method (C). In this particular case, the larger SP QRS angle obtained from the inverse
Dower method may have been driven in part by a QRS loop that was more posteriorly directed than that obtained from either the true Frank leads or the Kors'
regression method.
D.L. Cortez, T.T. Schlegel / Journal of Electrocardiology 43 (2010) 302–309 305
Fig. 2. Line of concordance for the SM QRS-T angle as obtained from the true Frank leads versus the Kors' regression-related (A) and inverse Dower-related (B)
reconstructions, respectively, for the entire data set. The greater concord of the true Frank-lead results to the Kors' regression-related results than to the inverse
Dower-related results is readily apparent.
Fig. 3. Bland-Altman plots for the SM QRS-T angle as obtained from the true Frank leads versus the Kors' regression-related (A) and inverse Dower-related (B)
reconstructions, respectively, for the entire data set. Along with smaller absolute differences from the true Frank-lead results (Y axis values, middle lines), the
Kors' regression-related results also demonstrated closer 95% Bland-Altman limits of agreement (dashed lines at the top and bottom) to the true Frank-lead
results than did the inverse Dower-related results.
306 D.L. Cortez, T.T. Schlegel / Journal of Electrocardiology 43 (2010) 302–309
Table 2
Lin's concordance coefficient (ρ) and Bland-Altman 95% limits of agreement versus the true Frank leads for, respectively, the Kors' regression and inverse
Dower-related transformations
Kors' regression versus true Frank leads Inverse Dower versus true Frank leads
ρ (95% CI) Bland-Altman 95% ρ (95% CI) Bland-Altman 95%
limits of agreement limits of agreement
Control group (n = 50)
SM QRS-T angle 0.88 (0.81-0.94) −26.2 to 24.2 0.71 (0.59-0.83) −15.3 to 46.4
SP QRS-T angle 0.77 (0.66-0.88) −34.7 to 32.7 0.62 (0.48-0.77) −21.8 to 55.4
SVG magnitude 0.99 (0.98-0.99) −10 to 17 0.98 (0.97-0.99) −19 to 17
SVG elevation 0.76 (0.64-0.88) −12.7 to 15.5 0.73 (0.60-0.86) −18.0 to 12.8
SVG azimuth 0.70 (0.57-0.84) −30.0 to 29.8 0.63 (0.48-0.79) −40.9 to 23.9
Post-MI group (n = 100)
SM QRS-T angle 0.89 (0.85-0.93) −26.8 to 41.1 0.77 (0.69-0.84) −34.2 to 62.3
SP QRS-T angle 0.83 (0.77-0.89) −38.3 to 53.3 0.59 (0.47-0.71) −52.5 to 90.4
SVG magnitude 0.91 (0.88-0.94) −12 to 22 0.78 (0.71-0.85) −23 to 37
SVG elevation 0.93 (0.91-0.95) −25.7 to 18.6 0.86 (0.81-0.90) −36.6 to 20.3
SVG azimuth 0.74 (0.65-0.83) −81.3 to 96.4 0.60 (0.48-0.73) −113.7 to 119.1
Total sample (n = 150)
SM QRS-T angle 0.91 (0.88-0.93) −27.7 to 36.6 0.78 (0.73-0.84) −28.6 to 57.7
SP QRS-T angle 0.84 (0.80-0.89) −38.1 to 47.4 0.64 (0.55-0.72) −44.1 to 80.6
SVG magnitude 0.97 (0.97-0.98) −11 to 20 0.93 (0.91-0.95) −23 to 32
SVG elevation 0.92 (0.90-0.94) −22.2 to 18.4 0.85 (0.81-0.89) −31.7 to 19.0
SVG azimuth 0.75 (0.68-0.82) −69.8 to 79.7 0.62 (0.52-0.72) −98.3 to 96.2
smaller than those derived from the inverse Dower method. tested against the true Frank leads. Of note are the
On the other hand, SVG magnitudes, elevations, and consistently larger (closer to 1) ρ values as well as tighter
azimuths derived from the 3 different methods were not ρ CIs and Bland-Altman 95% limits of agreement for
significantly different, except that for the entire data set (N = every single parameter measured in all 3 groups for the Kors'
150); the SVG elevation derived from the inverse Dower regression-related method than for the inverse Dower-related
reconstruction was significantly smaller than that obtained method when using the true Frank-lead results as the
from the true Frank leads. Finally, the SM QRS-T angles criterion standard.
were also significantly larger than the SP QRS-T angles for
both groups and for the data set as a whole. Fig. 1 shows
Discussion
example 3-dimensional VCG loops and accompanying
diagrammatic SP QRS-T angles from a healthy individual. The most important finding from this study is that spatial
Consistent with data from both the healthy and post-MI QRS-T angle results derived from the Kors' regression-
groups, this individual's SP QRS-T angle obtained from the related transformation more closely approximate those from
true Frank leads (Fig. 1A) was very close to that obtained the true Frank leads than do corresponding results from the
from the Kors' regression method (Fig. 1B) but was notably inverse Dower-related transformation. Specifically, com-
smaller than that obtained from the inverse Dower method pared with the use of the Kors' regression transformation, the
(Fig. 1C). use of the inverse Dower transformation yields lower Lin's
Fig. 2 shows Lin's concordance line plots for the SM concordance correlation coefficients and wider associated
QRS-T angle as obtained from the true Frank leads versus CIs against the true Frank-lead criterion standard. The
the Kors' regression (Fig. 2A) and inverse Dower-related inverse Dower transformation, but not the Kors' regression
(Fig. 2B) reconstructions, respectively, for the entire data set. transformation, also significantly overestimates the absolute
The dashed line represents the line of true concordance, value for the SM and SP QRS-T angles both in healthy
whereas the solid line represents a best-fit line to the true subjects and post-MI patients. Therefore, results from
Frank-lead data using the respective reconstruction tech- inverse Dower-related transformations for spatial QRS-T
nique. Fig. 3 shows the corresponding Bland-Altman plots angle should not be considered tantamount to results from
for the same comparisons. In Fig. 3, the centrally located the true Frank leads, as has already been at least indirectly
dashed lines represent the best-fit line obtained by the true discovered.17,18 In addition, absolute values for spatial
Frank leads, whereas the solid lines represent the lines QRS-T angles and SVG that are “abnormal” when using
obtained by using the given reconstruction technique. Figs. 2 Kors' regression transformations may be falsely construed as
and 3 both visually demonstrate the greater concord of the “normal” if referencing absolute values from inverse Dower
Kors' regression-related results than the inverse Dower- transformations. Thus, assignments of “normal versus
related results to the true Frank-lead results. abnormal” should only be made on a transform-specific
Table 2 shows Lin's concordance coefficient (ρ) and its basis. Table 3 provides, for rapid reference purposes, a
accompanying confidence interval (CI) as well as the Bland- summary of prior studies according to the apparent type of
Altman 95% limits of agreement for each parameter in each spatial QRS-T angle measured and to the specific transfor-
group when the particular reconstruction technique was mation method used.
D.L. Cortez, T.T. Schlegel / Journal of Electrocardiology 43 (2010) 302–309 307
Acknowledgments
The other finding of interest from this study is that when
using the true Frank leads or either reconstruction, mean The authors especially thank Oeff et al and physionet.org
values for the SM QRS-T angle are significantly larger than for providing access to the PTB Diagnostic ECG Database.
those for the SP QRS-T angle, and thus, these 2 measures They also thank Drs W.B. Kulecz, J.L. DePalma, and E.C.
should not be used interchangeably. We prefer to refer to Greco for programming and technical assistance and Dr R.
the angular difference between the “maximum” QRS and T Ploutz-Snyder for statistical advice. Additional support was
vectors1,3,23-25 as the spatial “peaks” rather than spatial provided by the National Space Biomedical Research
“maximum” QRS-T angle because the use of the word Institute and NASA/Wyle Laboratories medical student
“maximum” results in confusing nomenclature, inasmuch as internship programs (DLC) and NASA Technology Invest-
in most individuals, the angle between the peaks of the ment Funds (TTS).
spatial QRS and T loops is actually smaller than the angle
between the QRS and T position vectors defined by the areas
of the same loops. The mathematical derivations for both Appendix 1: Derivation of VCG parameters
types of angle are provided in Appendix 1, along with that
for the SVG. A usual first step in combining the signals from the X, Y,
and Z leads from the true or derived Frank leads is to form a
Limitations vector magnitude V, which is
The SM QRS-T angle is in turn calculated by the dot of the ECG-derived QRS-T angle. Circ Arrhythm Electrophysiol
product in units of degrees as 2009;2:548.
5. de Torbal A, Kors JA, van Herpen G, et al. The electrical T-axis and the
spatial QRS-T angle are independent predictors of long-term mortality
SM QRS−T angle in patients admitted with acute ischemic chest pain. Cardiology
QRSxTx þ QRSyTy þ QRSzTz − 1
2004;101:199.
¼ cos : 6. Kardys I, Kors JA, van der Meer IM, Hofman A, van der Kuip DA,
j QRSarea j jTarea j Witteman JC. Spatial QRS-T angle predicts cardiac death in a general
ð5Þ population. Eur Heart J 2003;24:1357.
7. Kors JA, Kardys I, van der Meer IM, et al. Spatial QRS-T angle as a risk
indicator of cardiac death in an elderly population. J Electrocardiol
On the other hand, in the simplest possible terms, the SP 2003;36(Suppl):113.
QRS-T angle is calculated in units of degrees as 8. Rautaharju PM, Ge S, Nelson JC, et al. Comparison of mortality risk for
−1 electrocardiographic abnormalities in men and women with and without
AB coronary heart disease (from the Cardiovascular Health Study). Am J
SP QRS−T angle ¼ cos ð6Þ
j A j jB j Cardiol 2006;97:309.
9. Rautaharju PM, Kooperberg C, Larson JC, LaCroix A. Electrocardio-
where A and B are the vectors of maximum magnitude of the graphic predictors of incident congestive heart failure and all-cause
mortality in postmenopausal women: the Women's Health Initiative.
QRS and T waves. Circulation 2006;113:481.
To define the vectors of maximum magnitude, first define 10. Rautaharju PM, Kooperberg C, Larson JC, LaCroix A. Electrocardio-
the R-wave position vector RP by the values of X, Y, and Z graphic abnormalities that predict coronary heart disease events and
at the location of the 3-dimensional R-wave (QRS loop) mortality in postmenopausal women: the Women's Health Initiative.
peak, where the peak is defined by the maximum value of V Circulation 2006;113:473.
11. Rautaharju PM, Prineas RJ, Wood J, Zhang ZM, Crow R, Heiss G.
in Eq. (1) over the QRS interval: Electrocardiographic predictors of new-onset heart failure in men and in
women free of coronary heart disease (from the Atherosclerosis in
RP ¼ RPx + RPy + RPz ð7Þ Communities [ARIC] Study). Am J Cardiol 2007;100:1437.
12. Yamazaki T, Froelicher VF, Myers J, Chun S, Wang P. Spatial QRS-T
Similarly, define the T-wave position vector TP by the angle predicts cardiac death in a clinical population. Heart Rhythm
values of X, Y, and Z at the location of the 3-dimensional T- 2005;2:73.
wave (T-loop) peak, where the peak is defined by the 13. Edenbrandt L, Pahlm O. Vectorcardiogram synthesized from a 12-lead
maximum value of V in Eq. (1) over the T interval: ECG: superiority of the inverse Dower matrix. J Electrocardiol
1988;21:361.
14. Kors JA, van Herpen G, Sittig AC, van Bemmel JH. Reconstruction of
TP ¼ TPx + TPy + TPz ð8Þ
the Frank vectorcardiogram from standard electrocardiographic leads:
diagnostic comparison of different methods. Eur Heart J 1990;11:1083.
The SP QRS-T angle is then the angle in degrees between 15. Levkov CL. Orthogonal electrocardiogram derived from the limb and
position vectors RP and TP as again calculated by the dot- chest electrodes of the conventional 12-lead system. Med Biol Eng
product: Comput 1987;25:155.
16. Rubel P, Benhadid I, Fayn J. Quantitative assessment of eight different
RPxTPx + RPyTPy + RPzTPz − 1 methods for synthesizing Frank VCGs from simultaneously recorded
SP QRS−T angle = cos ; standard ECG leads. J Electrocardiol 1992;24(Suppl):197.
jRPj j TPj 17. Scherptong RW, Man SC, Cessie S, et al. The Spatial QRS-T Angle and
ð9Þ the Spatial Ventricular Gradient: Normal Limits for Young Adults.
Comp Cardiol 2007;34:717.
where 18. Scherptong RW, Henkens IR, Man SC, et al. Normal limits of the spatial
QRS-T angle and ventricular gradient in 12-lead electrocardiograms of
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi young adults: dependence on sex and heart rate. J Electrocardiol
jRP j¼ ðRPxÞ2 + ðRPyÞ2 + ðRPzÞ2 ð10Þ 2008;41:648.
19. Geelen A, Brouwer IA, Zock PL, et al. (N-3) fatty acids do not affect
and electrocardiographic characteristics of healthy men and women. J Nutr
2002;132:3051.
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 20. Solaimanzadeh I, Schlegel TT, Feiveson AH, et al. Advanced
jTP j¼ ðTPxÞ2 + ðTPyÞ2 + ðTPzÞ2 : ð11Þ electrocardiographic predictors of mortality in familial dysautonomia.
Auton Neurosci 2008;144:76.
21. Ishizawa K. Mean QRS, ventricular gradient and left ventricular mass in
References patients with eccentric left ventricular hypertrophy. J Electrocardiol
1975;8:227.
1. Dilaveris P, Gialafos E, Pantazis A, Synetos A, Triposkiadis F, Gialafos 22. Libretti A, Zanchetti A. Spatial patterns of ventricular repolarization in
J. The spatial QRS-T angle as a marker of ventricular repolarisation in arterial hypertension. Am Heart J 1960;59:40.
hypertension. J Hum Hypertens 2001;15:63. 23. Dilaveris P, Pantazis A, Gialafos E, Triposkiadis F, Gialafos J. The
2. Triola B, Olson MB, Reis SE, et al. Electrocardiographic predictors of effects of cigarette smoking on the heterogeneity of ventricular
cardiovascular outcome in women: the National Heart, Lung, and Blood repolarization. Am Heart J 2001;142:833.
Institute-sponsored Women's Ischemia Syndrome Evaluation (WISE) 24. Rubulis A, Jensen J, Lundahl G, Tapanainen J, Bergfeldt L. Ischemia
study. J Am Coll Cardiol 2005;46:51. induces aggravation of baseline repolarization abnormalities in left
3. Voulgari C, Tentolouris N, Moyssakis I, et al. Spatial QRS-T angle: ventricular hypertrophy: a deleterious interaction. J Appl Physiol
association with diabetes and left ventricular performance. Eur J Clin 2006;101:102.
Invest 2006;36:608. 25. Rubulis A, Jensen J, Lundahl G, Tapanainen J, Wecke L, Bergfeldt L. T
4. Borleffs CJ, Scherptong RW, Man SC, et al. Predicting ventricular vector and loop characteristics in coronary artery disease and during
arrhythmias in patients with ischemic heart disease: clinical application acute ischemia. Heart Rhythm 2004;1:317.
D.L. Cortez, T.T. Schlegel / Journal of Electrocardiology 43 (2010) 302–309 309
26. Ball MF, Pipberger HV. The normal spatial QRS-T angle of the 31. Schlegel TT, Kulecz WB, DePalma JL, et al. Real-time 12-lead high
orthogonal vectorcardiogram. Am Heart J 1958;56:611. frequency QRS electrocardiography for enhanced detection of
27. Draper HW, Peffer CJ, Stallmann FW, Littmann D, Pipberger HV. The myocardial ischemia and coronary artery disease. Mayo Clin Proc
corrected orthogonal electrocardiogram and vectorcardiogram in 510 2004;79:339.
normal men (Frank lead system). Circulation 1964;30:853. 32. Lin LI. A concordance correlation coefficient to evaluate reproducibility.
28. Ishizawa K, Motomura M, Konishi T, Wakabayashi A. High reliability Biometrics 1989;45:255.
rates of spatial pattern analysis by vectorcardiogram in assessing the 33. Bland JM, Altman DG. Statistical methods for assessing agreement
severity of eccentric left ventricular hypertrophy. Am Heart J 1976;91:50. between two methods of clinical measurement. Lancet 1986;1:307.
29. Rautaharju PM, Prineas RJ, Zhang ZM. A simple procedure for 34. Ishizawa K, Murai K, Motomura M, Matsuoka J, Chen CH.
estimation of the spatial QRS/T angle from the standard 12-lead Vectorcardiographic findings in concentric and eccentric left ventricular
electrocardiogram. J Electrocardiol 2007;40:300. hypertrophy as determined by angiocardiograms. 1. Preliminary report.
30. Bousseljot R, Kreiseler D, Schnabel A. Nutzung der EKG- Jpn Circ J 1975;39:247.
Signaldatenbank CARDIODAT der PTB über das Internet. Biome- 35. Jeffreys H, Jeffreys BS. Numerical methods. Methods of mathematical
dizinische Technik 1995;40:S317. physics, 3rd ed. Cambridge: Cambridge University Press; 1988. p. 286.