Body Surface Two-Dimensional Spectral Map of Atrial Fibrillation

Using Vector-Projected 187 Channel Electrocardiography

Kenji Nakai,1 MD, Manabu Itoh,1, Hitoshi Okabayashi,2 MD, Junichi Tsuboi,2 MD,
Yoshino Mitsunaga,2 MD, Takashi Komatsu,3 MD, and Kunihiro Yoshioka,4 MD
Summary
Atrial cycle length during atrial fibrillation and flutter waves may be correlated with atrial refractoriness and organization. The nature of the frequency by spectral analysis may reflect a profile of atrial cycle length. In this study, we developed a novel body surface 2-dimensional spectral map during fibrillation using vector-projected 187 channel ECG
(187ch VP-ECG).
The study consisted of 28 patients (24 with atrial fibrillation (AFIB) and 4 atrial flutter (AFL) with valvular heart
disease). We performed spectral analysis by maximum entropy modeling (MEM) in 4 second nonaveraged 187ch electrical current waves by 187ch VP-ECG. Body surface spectral features were displayed according to the frequency and
power magnitude components. We verified the accuracy of the spectral features by a 64ch magnetocardiography (MCG).
The average dominant frequency in AFL by 187ch VP-ECG was lower than those in AFIB (4.6 0.9 Hz in AFL,
7.2 0.9 Hz in AFIB, P < 0.01). Comparison of average dominant frequency between 187ch VP-ECG and 64 ch MCG
demonstrated good agreement (y = 0.86x+0.84, r2 = 0.89, P < 0.0001). Body surface 2-dimensional spectral features
demonstrated homogenous spectrum patterns in AFL, and in-homogenous spectrum patterns in AFIB.
In conclusion, novel body surface spectral mapping using 187ch VP-ECG may represent a 2-dimensional spectral
feature that may be related to atrial refractoriness and organization. (Int Heart J 2012; 53: 5-10)
Key words: Atrial fibrillation, Body surface map, Vector projection, Spectral analysis, Magnetocardiography

trial fibrillation (AFIB) was reported to be an independent risk factor for stroke in the Framingham Study. In
patients with coronary heart disease or cardiac failure,
atrial fibrillation doubled the stroke risk in men and trebled the
risk in women.1,2) Left atrial myocardial extension on to pulmonary veins in humans by anatomic observation might be
relevant for atrial arrhythmias.3) Certain cases of AFIB may be
maintained by small re-entrant rotors that result in a hierarchical distribution of frequencies throughout the atria.4,5) Recent
advances in interventional catheter ablation and surgical pulmonary vein isolation, radical treatments have been applied in
patients with atrial flutter (AFL) and AFIB.6,7) Holma, et al reported noninvasive assessment of the atrial cycle length during
atrial fibrillation in humans.8) Sanders, et al reported spectral
analysis and frequency mapping identify localized sites of
high-frequency activity during AFIB in humans with different
distributions in paroxysmal and permanent AFIB.9,10) Determination of the features of cycle length of excitation of atrial fibrillation is important for a therapeutic strategy.
Magnetocardiography (MCG) has the potential to detect
3-dimensional (3D) electrical current density and electrical repolarization phenomena because the magnetic field is unaffected by the shape of the lungs and the torso.11,12) Previously,
we reported a 3D spectral map of the atrial fibrillation generated by 64ch cardiomagnetic fields B in the z direction using a

space filter.13) Recently, we developed a 187 channel vectorprojected ECG (187ch VP-ECG) that could evaluate 2-dimensional (2D) characteristics of recovery time dispersion and
high frequency late potentials.14-16)
We hypothesized that AF cycle length may be correlated
locally with atrial refractoriness and globally with the degree
of atrial organization. In the present study, we developed body
surface spectral maps of AFIB and AFL using 187ch VP-ECG,
and verified the significance of the accuracy by 64ch MCG.

Methods
Subjects and surgical procedure: The study subjects consisted

of 28 patients with organic heart disease (24 AFIB and 4 AFL

with valvular heart disease) who had chronic AFIB. Patients
had a routine examination that included a chest X-ray, standard
ECG, 187ch VP-ECG, and a 64ch MCG. The baseline characteristics and clinical courses of the study patients are listed in
the Table. The institutional ethical commission approved the
research protocol (H18-2) and all patients provided informed
consent. All procedures were performed in accordance with institutional guidelines.
187ch VP-ECG recording and spectral analysis: The prototype
187ch VP-ECG consists of an electrode lead system, an input

From the Departments of 1 Internal Medicine of Dentistry, 2 Cardiac Surgery, 3 Cardiology and 4 Radiology of Memorial Heart Center, Iwate Medical University, Iwate,
Japan.
This study was supported by a Grant-in-Aid for Scientific Research (C)(22590792) of The Ministry of Education, Culture, Sports, Science and Technology (MEXT).
Address for correspondence: Kenji Nakai, MD, Internal Medicine of Dentistry, Iwate Medical University, 19-1 Uchimaru, Morioka, Iwate 020-8505, Japan.
Received for publication July 25, 2011.
Revised and accepted November 17, 2011.
5

Int Heart J
January 2012

NAKAI, ET AL
Table. Baseline Characteristics of the Study Patients
No

ID

Name

Age

Sex

DIAG

Rhythm

HR

fQRS

MCG mF

VP-ECG mF

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28

20061002
20061115
20070207
20071105
20060522
20060526
20060529
20060621
20060719
20061013
20061025
20061025
20061129
20061127
20061115
20061108
20061122
20070611
20070516
20070528
20070608
20070612
20070702
20070928
20071003
20071022
20080121
20080128

KY
KN
YH
AS
IS
KY
TY
TY
OE
IC
MH
MY
MY
SS
ES
YK
YN
KM
KT
KI
IY
TS
TT
ST
YS
AS
KK
HY

70
68
71
58
74
51
73
73
73
55
64
69
69
61
54
80
60
60
83
65
21
75
55
63
73
58
65
65

M
F
M
M
F
M
M
M
M
M
M
M
M
M
F
M
F
M
M
M
M
F
F
F
M
M
M
F

MR,MVP
MS
AR
AR
MR
HHD
MR
MR
MR,TR
MR,AR
MR
MR
MR
MR
MR
MR
MS
Lone AFIB
HHD
HHD
MR
MR
MR
MR,AS
MR
AR
MR
MS

AFL
AFL
AFL
AFL
AFIB
AFIB
AFIB
AFIB
AFIB
AFIB
AFIB
AFIB
AFIB
AFIB
AFIB
AFIB
AFIB
AFIB
AFIB
AFIB
AFIB
AFIB
AFIB
AFIB
AFIB
AFIB
AFIB
AFIB

97
66
90
91
54
79
54
83
49
87
55
59
78
52
92
91
58
79
81
98
57
66
46
64
81
52
75
51

114
110
107
131
104
114
106
100
109
99
170
133
120
109
116
100
93
106
87
95
119
106
114
111
112
132
101
109

4.4
3.9
4.5
5.4
7.0
7.2
9.5
9.8
8.0
7.1
6.8
6.8
6.6
9.0
6.8
6.8
7.8
7.5
5.6
6.3
6.5
7.8
5.7
7.3
9.3
7.1
7.0
6.8

4.5
3.6
4.5
5.7
7.1
6.9
9.0
8.7
7.6
7.4
6.2
6.9
7.2
8.2
5.4
6.7
8.1
7.5
5.5
6.4
6.4
6.9
6.1
7.5
8.9
7.2
6.9
6.9

DIAG indicates diagnosis; HR, heart rate (beat/minutes); fQRS, filtered QRS (msec); MCG mF, mean frequency by MCG (Hz); VP-ECG mF, mean frequency by VP-ECG (Hz); MR, mitral regurgitation; MVP, mitral prolapse; AR, aortic regurgitation; HHD, hypertensive heart disease; TR, tricuspid regurgitation; MS, mitral stenosis; Lone AFIB, lone atrial fibrillation; AS, aortic stenosis; AFL, atrial flutter; and AFIB, atrial fibrillation.

box, a high precision amplifier (prototype; Fukuda Denshi Co.

Ltd., Tokyo), and a personal computer. The input box generates a modified X, Y, Z-lead ECG and vector-projected 187ch
synthesized ECGs from a Mason-Likar lead system. The input
signal ( 550 mV) was digitized at 2 kHz by an A/D converter
with a resolution of 0.076 V. Acquired ECG signals were
processed by custom software (ECG manager, ICS Co. Ltd.,
Morioka, Japan).
The 10 electrodes were attached to the right shoulder
(RS), left shoulder (LS), left lower abdomen (LLA), right lower abdomen (RA), or V1 through V6 (quaternary intercostal
spaces of the left sternal border) according to the Mason-Likar
lead system. We used silver-silver chloride electrodes (TE18M-5, Magnerode, Fukuda Denshi, Co., Ltd., Tokyo). An
electrode was connected to an input box at one end. We recorded 10 minutes of synthesized 187ch VP-ECG in the resting supine position in a shield room to eliminate electrostatic
and electromagnetic noise.
We originally determined transformation matrices for 187
positions on a torso surface based on a projection theory reported by Frank.17) Representative positions are shown in Figure 1 as previously reported.13,14) In brief, we proposed 11 parallel, equispaced transverse levels with 2-inch spacing between
the levels. Level 5 coincides with the second intercostal space
and level 6 with the heart center. Seventeen letter designations
(A-Q) indicate the intersection with each transverse level of radial lines separated by equal angles of 11.25 emanating from
the longitudinal anatomic axis of the torso. A transformation

matrix of each channel (i = 1~187) on the torso surface model

was determined initially. The electric potential of the central
terminal (CT) was determined by the equation: CT =
(RS+LS+LLA)/3. Next, the virtual electrical potentials (VEP)
of the 187ch channels were calculated using the following
equation; VEPi = 1~187 = Xi - CT for X projection, Yi - CT for Y
projection, Zi - CT for Z projection. In this time, we calculated
the synthesized 187ch ECG referred real signal from precordial V2 by matrix calculation.
The body surface spectral map of AFIB was generated as
follows (Figure 2). A 187ch VP-ECG was acquired for 5 minutes from each patient. After high-pass filtering (0.5 Hz), the
QRS waves were subtracted at the zone of 50 msec from the
peak R using a template-matched algorithm. In order to remove residual signals of q and S waves in cases with bundle
branch block, QRS waves were subtracted at the zone of 70
ms from the peak R. The fibrillatory baseline signal was then
tapered at the edges to a 0 value by the Hamming window. We
performed spectral analysis by maximum entropy modeling
(MEM)18) in 4 second nonaveraged 187ch electrical current
waves estimated by VP-ECG. Frequency features for every
minute and a 2D dominant frequency map based on the power
magnitude of hierarchic frequency were displayed. Color coding indicates hierarchic dominant frequency features: ie, blue;
0~5 Hz, yellow; 5~7 Hz, red; 7~10 Hz, respectively.
64ch MCG and generation of 3D spectral map: The MCG
studies were performed in the Biomagnetism Laboratory at the
Memorial Heart Center of Iwate Medical University (proto-

Vol 53
No 1

SPECTRAL MAP OF AFIB BY VECTOR-PROJECTED ECG

Figure 1. Vector projections for 187 fixed positions on a torso surface

model included 11 parallel, equispaced transverse levels with 2-inch spacing between levels. Level 5 coincides with the second intercostals space
and level 6 with the heart center. Seventeen letter designations indicate the
intersection of each transverse level with radial lines separated by equal
angles of 11.25 emanating from the longitudinal anatomic axis of the torso.

type: SQUID sensor, manufactured by Hitachi High-Technology, Co., Ltd., Tokyo). We recorded the 64ch MCG in a magnetically shielded room. A brief examination was performed in
order to guide the probe placement. Magnetic coils were
placed at precordial leads in the V1, V2, V4, and xiphoid regions. Current signals from these magnetic coils were initially
recorded at approximately 2-second intervals, followed by a
10-minute recording for each subject in the same position. The
recordings were digitized at 500 Hz and filtered with a bandpass frequency range of DC to 200 Hz. Subject interference
was removed using a matched filter.
The acquired MCG and MR images were processed by
custom software (prototype: MCG manager). The pre-processing module acquired MCG signals by noise filtration, the elimination of bad signals, and the interpolation of missing signals.
The generation of a space filter for determining the current
density from Bz magnetic fields has been reported previously.
Theoretically, the electric current and the magnetic field have a
linear relationship (B = LJ). Next, we determined the current
density by the next equation, J = L-1B. The dumped least
squares method with the Tikhonov regulation method was applied to handling signals (B) equally from the deep and shallow parts in the next equation, J = LT[LLT + I]-1B.12)
To determine the 3D heart polygons, the 3D current density was calculated from the Bz magnetic fields by applying
the space filter. The 3D heart polygons for the atrium and ventricle were determined from the magnetic field distribution.
Details of this procedure have been described previously.11,12)
The 3D spectral map of AFIB was generated using a previously reported method.12) MCG was acquired for 5 minutes
from each patient. We used a space filter to compose the 3D
current density map from the 64ch MCG data in the z direction. After high-pass filtering, the QRS complexes were subtracted using a template-matched algorithm. The resulting fibrillatory baseline signal was tapered at the edges to a 0 value
by the Hamming window. Next, we applied a 2048-point FastFourier transform (spectral resolution, 0.48 Hz) to obtain the

Figure 2. The body surface spectral map of AFIB was generated as follows: 187ch VP-ECG was acquired for 5 minutes from each patient. After
high-pass filtering (0.5 Hz), the QRS complexes were subtracted at the
zone of 50 msec from the peak R using a template-matched algorithm.
The fibrillatory baseline signal was then tapered at the edges to a 0 value
by the Hamming window. We performed spectral analysis by MEM in 4
second nonaveraged 187ch electrical current waves estimated by VPECG. Right lower panel indicates frequency feature for each minute and
2D dominant frequency map based on the power magnitude of hierarchic
frequency. Color coding indicates hierarchic dominant frequency feature,
blue; 0~5 Hz, yellow; 5~7 Hz, red; 7~10 Hz, respectively.

power spectrum of the nonaveraged 3D 64ch MCG data. Power spectral analysis allowed automatic determination of the
gradient frequency for each point acquired. We created 3D
color-coded frequency spectral maps within the heart outline
generated by the MCG, displaying low frequencies in blue and
high frequencies in red (blue represented 0~5 Hz, purple 5~7
Hz, and red 7~10 Hz).
MR images and reconstruction of the spectral map: We obtained MR images by standard techniques using a Signa Horizon LX instrument [GE Yokokawa, Tokyo]) in selected subjects (1 normal volunteer, 1 patient with common AFL, and 1
patient with AFIB) after applying a NifedipineR capsule at the
position of the precordial leads of V1, V2, V4, and the xiphoid
regions. The capsule could provide verification of the heart
outline generated by the MCG. Digital Imaging and Communications in Medicine (DICOM)-formatted MR images were
transferred to a personal computer (IBM Japan, Tokyo). A representative heart polygon of the atrium and ventricle determined by an integrated current density of the MCG was reconstructed with the silhouette on the MR images with adjustment
to the markers on the precordial position of the V1, V2, V4, and
xiphoid regions. Furthermore, the 3D power spectral map of
the common AFL was reconstructed on the MR images.
Statistical analysis: The mean standard deviations (SD) and
the comparison of mean frequency between 187ch VP-ECG
and 64ch MCG correlation were calculated and statistical analyses were performed by an unpaired t test using GraphPad
Prism4 (San Diego, CA, USA). Samples were considered to
be statistically significant when the two-tailed P value was <
0.05.

NAKAI, ET AL

Results
The baseline characteristics and clinical courses of the
patients are shown in the Table. A 3-dimensional spectral map
on 64ch MCG in representative AFL demonstrated homogenous distribution with a frequency of 4.4 Hz around the right
atrium verified by MRI (Figure 3A). A 3-dimensional spectral
map on 64ch MCG in representative AFIB demonstrated nonhomogenous distribution with a frequency of 6.8 Hz around
the atrium verified by MRI (Figure 3B).
The average mean frequency in AFL was lower than
those in AFIB (4.6 0.9 Hz in AFL, 7.2 0.9 Hz in AFIB, P
< 0.01). Comparison of mean frequency between 187ch VPECG and 64ch MCG indicated good agreement (y =

Int Heart J
January 2012

0.86x+0.84, r2 = 0.89, P < 0.0001) (Figure 4).

In a representative case with combined rheumatic valve
disease (mitral and aortic valves), MCG revealed chronic AFL.
The 3D spectral mapping of AFL showed the frequency (average, 4.4 Hz). In addition, the 3D dominant power spectrum
seemed to be distributed broadly within the right atrium. A
187ch VP-ECG demonstrated homogenous distribution with a
mean frequency (average, 4.5 Hz) (Figure 5).
In a representative case with combined rheumatic valve
disease (mitral valves), MCG revealed chronic AFIB. The 3D
spectral mapping of AFIB showed a frequency (average, 6.8
Hz). In addition, the 3D power spectrum seemed to be distributed broadly within the atria. A 187ch VP-ECG demonstrated
broad distribution with a mean frequency (average, 6.9 Hz)
(Figure 6).

Figure 3. Spectral map reconstructed with MR images in a representative

case with AFL and AFIB. A1: Nonaveraged 3D 64channel MCG data.
A2: A 3D spectral MCG map by FFT images exhibited homogenous distribution with a frequency of 4.4 Hz around the right atrium verified by
MR image (A2; horizontal view, A3; vertical view). B1: Nonaveraged 3D
64channel MCG data. B2: A 3D spectral MCG map by FFT images exhibited nonhomogenous distribution with a frequency of 6.8 Hz around
the atria verified by MR image (B2; horizontal view, B3; vertical view).

Figure 4. Relationship of frequency between 64ch MCG and 187ch VPECG. The average mean frequency in AFL was lower than that in AFIB
(4.6 0.9 Hz in AFL, 7.2 0.9 Hz in AFIB). Comparison of mean frequency between 187ch VP-ECG and 64ch MEG demonstrated good
agreement (y = 0.86x+0.84, r2 = 0.89, P < 0.0001).

Figure 5. In a representative case with combined rheumatic valve disease

(mitral and aortic valves), MCG revealed chronic AFL. The 3D spectral
mapping of AFL showed the frequency (average, 4.4 Hz). In addition, the
3D dominant power spectrum seemed to be distributed within the right
atria. A 187ch VP-ECG demonstrated homogenous distribution with the
mean frequency (average, 4.5Hz).

Figure 6. In a representative case with combined rheumatic valve disease

(mitral valves), MCG revealed chronic AFIB. The 3D spectral mapping of
AFIB showed the frequency (average, 6.8 Hz). In addition, the 3D power
spectrum seemed to be distributed broadly within the atria. A 187ch VPECG demonstrated broad distribution with the mean frequency (average,
6.9 Hz).

Vol 53
No 1

SPECTRAL MAP OF AFIB BY VECTOR-PROJECTED ECG

Discussion
This study demonstrated that novel body surface spectral
mapping using 187ch VP-ECG represents a 2D spectral feature. Body surface spectral mapping demonstrated a homogenous frequency pattern in AFL, and various frequency patterns
in AFIB. Thus, novel body surface dominant frequency mapping in humans could represent a 2D hierarchical gradient of
activity of spectral features that may be related to a profile of
atrial refractoriness and organization.
Frequency mapping in AFIB - Etiology of AFIB: Schuessler, et
al19) observed that activation patterns characterized by multiple
re-entrant circuits converted to a single high-frequency re-entrant circuit that resulted in fibrillation. Certain cases of chronic AFIB are maintained by small re-entrant sources that result
in a hierarchical distribution of frequencies throughout the
atria. The rotor frequency has been found to correlate well with
the dominant frequency that is determined by spectral analysis
using optical mapping.20) Recently, Konings reported that there
were significant differences among the 3 types of pacing-induced AFIB in median intervals (174, 150, 136 milliseconds)
using optical mapping.21)
Holma, et al hypothesized noninvasive assessment of the
atrial cycle length during atrial fibrillation in man.22) They proposed a new method named Frequencyanalysis of fibrillatory
ECG (FAF-ECG) originally developed by Slocum, et al.23) In
brief, (1) an average of the time-aligned QRST complexes was
created and subtracted from the original ECG signal, and (2) a
frequency spectrum of the residual ECG signal was estimated
using FFT. Of the 12 surface ECG leads, the precordial lead V1
was chosenfor the frequency analysis. This method has, however, some important limitations as pointed out by Holma.22) A
reliable and accurate identification of the atrial spectral component is dependent on suppression of the QRST complexes.
Our algorithm is different in several ways; 1) QRS complexes were subtracted at the zone of 50 msec from the peak
R (offset) in cases with a normal QRS interval, and 70 msec
from the peak R in cases with bundle branch block, 2) frequency spectrum was estimated by 187ch VP-ECG, 3) MEM
was used for frequency analysis, and 4) body surface spectrum
mapping was displayed. A MEM approach provides a more
accurate and noise-immune mean frequency estimation than
the FFT method.24) Most importantly, body surface maps enable us to display 2D frequency distribution.
Previous studies have shown that the atrial cycle length
during AFIB was from 300 to 600 msec, which may be correlated with the dominant frequency.25) In this study, the average
dominant frequency in AFL was lower than that in AFIB (4.2
0.39 Hz in AFL, 7.1 0.2 Hz in AFIB). Interestingly, this
study revealed that body surface spectral mapping demonstrated a homogenous frequency pattern in AFL, and various frequency patterns in AFIB.
Recent studies have revealed that the global cycle length
reflects the organization of the fibrillation and, further, that the
local cycle length correlates to atrial refractoriness.5) First, it
demonstrates the feasibility of performing spectral analysis
and endocardial dominant frequency (DF) mapping in humans
to recognize relatively small DF sites with spatiotemporal stability in a 3D representation with a hierarchical gradient of activity. Second, it identifies that, in patients with paroxysmal
AF, the DF sources of activity are often localized to the PVs.

In contrast, patients with permanent AF demonstrate DF sites

that are more often localized to the atria, including RA sites.
Sanders, et al reported spectral analysis and frequency mapping identify localized sites of high-frequency activity during
AFIB in humans with different distributions in paroxysmal and
permanent AFIB.9)
Thus, access to the cycle length may be of importance for
an understanding of the basic electrophysiological properties
of AFIB and for evaluating therapeutic interventions.
Clinical application by spectral analysis: Recently, Hassaguerre, et al reported that the spontaneous initiation of AFIB
by ectopic beats originates in the PV.6,7) The firing in the PV
region would be responsible for the initiation of paroxysmal
AFIB. As a result of advances in interventional catheter ablation, radical treatments have been performed in patients with
AFL and AFIB.26)
Studies performing mapping during AF have recognized
the presence of temporally and spatially periodic activity emanating from the PV region with regularity, suggesting that
these structures may have a role in maintaining AF through localized short CL reentry and/or focal high-frequency activity.6,20) The ability of spectral analysis and frequency mapping
to identify sources of activity maintaining AF raises the possibility of its use to identify the substrate for AF during ablation
procedures.
In a previous study, we found preoperative average frequency by spectral analysis using 64ch MCG was 6.1 Hz for
restored sinus rhythm and 7.2 Hz for persistent AFIB. In general, these average frequencies indicated 164 milliseconds and
139 milliseconds in median intervals of AFIB, respectively.
Our hypothesis therefore is higher frequency spectra (> 7.2
Hz) would reflect small and/or fast reentering wavelets, and
lower frequency spectra (< 6.1 Hz) would reflect large and/or
slow reentering wavelets. Furthermore, patients who had homogenous spectra with a higher frequency (> 7.2 Hz) around
the right atrium would indicate the presence of localized small
multiple reentering wavelets. Surgical PV isolation might have
a lesser effectiveness for restoration rhythm in these patients.
Three-dimensional spectral map using 64ch MCG may represent a meaningful noninvasive strategy for patients with AFIB
who receive an interventional procedure.
The ability of spectral analysis and frequency mapping to
identify sources of activity maintaining AF raises the possibility of its use to identify the substrate for AF during ablation
procedures.
Study limitations: Recent advances in interventional and surgical therapeutic procedures have been applied to patients with
AFL and AFIB. Sites of early activation and areas with late activation have been estimated with reasonable accuracy by using the invasive electric-anatomical mapping (CARTO) system
or by using a noncontact catheter.27,28)
This study showed that novel 2D body surface spectral
analysis and dominant frequency mapping using the 187ch
VP-ECG can be used to evaluate the features of AFIB.

Disclosure
Dr. Nakai has a license agreement concerning software use (ECG
manager) with Fukuda Denshi Co., Ltd., Japan.

10

NAKAI, ET AL

References

1. Kannel WB, Abbott RD, Savage DD, McNamara PM. Epidemiologic features of chronic atrial fibrillation: the Framingham study.
N Engl J Med 1982; 306: 1018-22.
2. Wolf PA, Abbott RD, Kannel WB. Atrial fibrillation as an independent risk factor for stroke: the Framingham Study. Stroke
1991; 22: 983-8.
3. Saito T, Waki K, Becker AE. Left atrial myocardial extension onto
pulmonary veins in humans: anatomic observations relevant for
atrial arrhythmias. J Cardiovasc Elctrophysiol 2000; 11: 888-94.
4. Allessie MA, Lammers WJEP, Bonke FIM, et al. Experimental
evaluation of Moes multiple wavelet hypothesis of atrial fibrillation. p. 265. In: Zipes DP, Jalife J (ed): Cardiac Arrhythmias.
Grune & Stratton, New York, 1985.
5. Skanes AC, Mandapati R, Berenfeld O, Davidenko JM, Jalife J.
Spatiotemporal periodicity during atrial fibrillation in the isolated
sheep heart. Circulation 1998; 98: 1236-48.
6. Hassaguerre M, Jas P, Shah DC, et al. Spontaneous initiation of
atrial fibrillation by ectopic beats originating in the pulmonary
veins. N Engl J Med 1998; 339: 659-66.
7. Hassaguerre M, Sanders P, Hocini M, et al. Changes in atrial fibrillation cycle length and inducibility during catheter ablation and
their relation to outcome. Circulation 2004; 109: 3007-13.
8. Holm M, Johansson R, Brandt J, Lhrs C, Olsson SB. Epicardial
right atrial free wall mapping in chronic atrial fibrillation. Documentation of repetitive activation with a focal spread -- a hitherto
unrecognised phenomenon in man. Eur Heart J 1997; 18: 290-310.
9. Sanders P, Berenfeld O, Hocini M, et al. Spectral analysis identifies sites of high-frequency activity maintaining atrial fibrillation
in humans. Circulation 2005; 112: 789-97.
10. Cohen D, Edelsack EA, Zimmerman JE. Magnetocardiograms
taken inside a shielded room with a superconducting point-contact
magnetometer. Appl Phys Lett 1970; 16: 278-80.
11. Nakai K, Izumoto H, Kawazoe K, et al. Three-dimensional recovery time dispersion map by 64-channel magnetocardiography may
demonstrate the location of a myocardial injury and heterogeneity
of repolarization. Int J Cardiovasc Imaging 2006; 22: 573-80.
12. Nakai K, Oka T, Okabayashi H, et al. Three-dimensional spectral
map of atrial fibrillation by a 64 channel magnetocardiogram. J
Electrocardiol 2008; 41: 123-30.
13. Fenici R, Brisinda D, Nenonen J, Fenici P. Phantom validation of
multichannel magnetocardiography source localization. Pacing
Clin Electrophysiol 2003; 26: 426-30.
14. Nakai K, Tsuboi J, Okabayashi H, et al. Development of a signalaveraged vector-projected 187-channel high-resolution electrocardiogram for the evaluation of the spatial location of high-frequency potentials and abnormal ventricular repolarization. Int Heart J
2007; 48: 701-13.

Int Heart J
January 2012

15. Nakai K, Miyake F, Kasanuki H, et al. Newly developed signalaveraged vector-projected 187-channel electrocardiogram can
evaluate the spatial distribution of repolarization heterogeneity. Int
Heart J 2008; 49: 153-64.
16. Suzuki A, Shiga T, Nakai K, et al. Interlead difference between
T-peak to T-end intervals in resynchronization patients with an
implantable cardioverter-defibrillator. J Electrocardiol 2010; 43:
706-12.
17. Frank E. General theory of heat-vector projection. Circ Res 1954;
2: 258-70.
18. Jaynes ET. Information theory and statistical mechanics. Physical
Review 1957; 108: 171-90.
19. Schuessler RB, Grayson TM, Bromberg BI, Cox JL, Boineau JP.
Cholinergically mediated tachyarrhythmias induced by a single
extrastimulus in the isolated canine right atrium. Circ Res 1992;
71: 1254-67.
20. Jalife J. Rotors and spiral waves in atrial fibrillation. J Cardiovasc
Electrophysiol 2003; 14: 776-80. (Review)
21. Konings KT, Kirchhof CJ, Smeets JR, Wellens HJ, Penn OC, Allessie MA. High-density mapping of electrically induced atrial fibrillation in humans. Circulation 1994; 89: 1665-80.
22. Holm M, Pehrson S, Ingemansson M, et al. Non-invasive assessment of the atrial cycle length during atrial fibrillation in man: introducing, validating and illustrating a new ECG method. Cardiovasc Res 1998; 38: 69-81.
23. Slocum Je, Ropella Km. Correspondence between the frequency
domain characteristics of simultaneous surface and intra-atrial recordings of atrial fibrillation. IEEE Comput Cardiol 1994; 21:
781-4.
24. Kim SI, Lee KM. Mean frequency estimation of myoelectric signal using 2nd order maximum entropy method. Engineering in
Medicine and Biology Society, Proceedings of the Twelfth Annual
International Conference of the IEEE. 1990; 1-4: 2208-9.
25. Roithinger FX, Karch MR, Steiner PR, Sippens Groenewegen A,
Lesh MD. The spatial dispersion of atrial refractoriness and atrial
fibrillation vulnerability. J Interv Card Electrophysiol 1999; 3:
311-9.
26. Jas P, Shah DC, Hassaguerre M, et al. Efficacy and safety of septal and left-atrial linear ablation for atrial fibrillation. Am J Cardiol
1999; 84: 139R-146R.
27. Willems S, Weiss C, Ventura R, et al. Catheter ablation of atrial
flutter guided by electroanatomic mapping (CARTO): a randomized comparison to the conventional approach. J Cardiovasc
Electrophysiol 2000; 11: 1223-30.
28. Schilling RJ, Peters NS, Davies DW. Simultaneous endocardial
mapping in the human left ventricle using a noncontact catheter:
comparison of contact and reconstructed electrograms during sinus rhythm. Circulation 1998; 98: 887-98.