198 IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERNG, VOL. 43, NO. 2, FEBRUARY 1996

for Analysis of Atrial Activlation ronic Atrial Fibrillation in

Magnus Holm,* Member, ZEEE, Rolf Johansson, Member, ZEEE, S. Bert2 Olsson, Johan Brandt, and Carsten Luhrs

Abstruct- To further clarify the mechanisms maintaining chronic atrial fibrillation (CAF), a method identifying preferable activation patterns of the atria during fibrillation, by time averaging of multiple discrete excitation vectors, vas developed. Repeated recordings, each of 56 atrial bipolar electrograms simultaneously acquired during 8 s, were made at multiple sites in the right atrial free wall and the left atrial appendage in 16 patients with CAF using a 2.17 x 3.54 cm electrode array. The local activation times (LAT's) in each recording were estimated as the median activation time at the respective measurement point. By calculating the time difference between the LAT's at adjacent measurement points in two spatial dimensions, a direction vector was created for each activation wave passing each set of measurement points, a total of 42 sets. By time averaging of the individual direction vectors (typically m = 55) at each set of measurement points, preferable activation patterns were determined. Three types of activation patterns were found 1) inconsistent activation (n = 9, 2) consistent activation with preferential propagation directions (n = 7) and 3) consistent activation with impulses originating from a localizable site within the recording area (n = 4). All activation patterns were reproducible and the two latter patterns were proven signiscant using statistical tests. We conclude that this new method is useful in further clarification of the mechanisms involved in the maintenance of atrial fibrillation.

I. INTRODUCTION ESPITE the fact that atrial fibrillation is one of the most common cardiac arrhythmias in man, the mechanisms

maintaining this arrhythmia have not yet been satisfactorily clarified. In a paper published in 1962, Moe presented his so-called multiple wavelet hypothesis to explain the charac- teristics of atrial fibrillation [ 11. According to this hypothesis, atrial fibrillation is maintained by the coexistence of a number of independent activation waves, called wavelets, that travel randomly through the atrial myocardium around multiple islets or strands of nonexcitable tissue. By endocardial activation mapping, Allessie and co-workers have confirmed Moes hy- pothesis as a basis of acetylcholine-induced atrial fibrillation in isolated canine hearts [2]. The same group has shown, by high

Manuscript received July 5, 1994; revised September 28, 1995. This work was supported by the Swedish Heart Lung Foundation, the Crafoord Foundation and the Torsten Westerstrom Foundation. This study has been approved by the Ethical Committee of the Medical Faculty, Lnnd University, Sweden, under approval LU 184-90. Asterisk indicates corresponding author.

*M. Holm is with the Department of Cardiology, University Hospital S-221 85 Lund, Sweden(e-mail: magnush@kard.usil.lu.se).

R. Johansson is with the Department of Automatic Control, Lund Institute of Technology, Sweden.

S. B. Olsson is with the Department of Cardiology, University Hospital, Lund, Sweden

J. Brandt and C. Luhrs are with the Department of Cardiothoracic Surgery, University Hospital, Lund, Sweden.

Publisher Item Identifier S 0018-9294(96)01054-3.

density epicardial activation mapping during cardiac surgery, that during electrically induced atrial fibrillation in patients with Wolff-Parkinson-White syndrome, the right atrium is activated by one or multiple activation waves propagating in different directions [3]. They identified three types of right atrial activation patterns, characterized by different numbers and dimensions of the intra-atrial reentrant circuits. Another study, performed by Cox and co-workers, included epicardial activation mapping of electrically induced atrial fibrillation both in dogs with created mitral regurgitation and in patients with Wolff-Parkinson-White syndrome 141. The whole epi- cardial surface of both the right and the left atria was mapped using 208 electrodes in the animal set-up and 160 electrodes in the clinical set-up. The study demonstrated that multiple wave fronts, nonuniform conduction, bidirectional block and large reentrant circuits occurred during induced atrial fibrillation. Small reentrant circuits, however, were not reported by the investigators, possibly due to the limited spatial resolution. The role of the multiple wavelet hypothesis in chronic atrial fibrillation in man, however, has, to our knowledge, only been addressed in one study, where the tendency for wave fronts to follow paths of previous excitation, so-called linking, was examined by analyzing atrial endocardial electrograms acquired using an orthogonal catheter [5]. In a majority of the patients, the investigators found groups of consecutive atrial beats where the direction of activation of each beat was within 30" of the previous beat and, furthermore, that this finding was consistent from the first minute of recording to the second. Their conclusion was that atrial activation during chronic atrial fibrillation in man is not entirely random.

To further investigate the degree of spatial and temporal organization of atrial activation during fibrillation, we have developed a new method of identifying preferable activation pattern by time averaging of multiple discrete excitation vectors calculated from simultaneously acquired atrial electrograms. Based on a hypothesis of dual atria-nodal conduction routes, including a possible reentry mechanism participating in the maintenance of the fibrillation [6]-[8], we have applied this method in patients with chronic atrial fibrillation and, for evaluation purposes, also in patients with stable sinus rhythm, undergoing open-heart surgery. The purpose of this paper is to describe the new method in detail and illustrate the potential of the method.

11. MATERIAL Three patients with stable sinus rhythm, one patient with

induced atrial fibrillation and sixteen patients with chronic

0018-9294/96$05.00 0 1996 IEEE

HOLM et al.: A NEW METHOD FOR ANALYSIS OF ATRIAL ACTIVATION DURING CHRONIC ATRIAL FIBRILLATION IN MAN 199

TABLE I CLINICAL CHARACTERISTICS AND THE NUMBER OF RECORDINGS AT EACH SITE IN EACH PATIENT

No. of recordings at each site

RAFW RAFW LAA Patient Sex Age Basal Basal heart LA size Duration of # M/F (years) Rhythm disease (mm) chronic AF site A site B

1 M 64 Stable SR IHD NA - 1 2 M 61 Stable SR IHD 46 1 3 M 62 Stable SR IHD NA - 1 4 F 54 CAF MI 67 10 months 1 5 F 64 CAF MShHD 53 10 years 1 6 M 59 IAF IHD 29 1 7 F 68 CAF MUMS 54 7 years 3 1 8 F 71 CAF MUMS 84 18 years 3 1 9 M 46 CAFPAF MUMV Prolaps 54 3 months (PAF 20 years) 3 2 10 F 74 CAF MI/MS+AI/AS 60 9years 3 3 1 1 M 66 CAF MI 54 19 years 3 3 12 M 69 CAF IHD NA 10 years 3 3 13 F 69 CAF MS 57 20 years 3 3 14 M 59 CAF AI 40 3 months 2 2 2 15 M 73 CAF MVMS+AI 63 5 years 3 3 3 16 F 59 CAF MI/MS+IHD 55 2 months 3 3 3 17 M 71 CAFPAF IHD 44 1.5 years (PAF 13 years) 3 3 2 18 M 73 CAF IHD 46 10 months 3 3 19 M 51 CAFPAF AS 40 2 months (PAF 9 years) 3 3 3 20 M 76 CAF AS 38 4 years 3 3 3

___- ______

~ -

SR=Shus Rhythm; CAF=Chronic Atrial Fibrillation; IAF=Induced Atrial Fibrillation (by the surgical procedure); PAF=Paroxysmal Atrial Fibrillation; MI=Mitral valve Insufliciency; MS=Mitral valve Stenosis; IHD=Ischemic Heart Disease; MV=Mitral Valve; AI=Aortic valve Insufficiency; AS=Aortic valve Stenosis; LA=Left atrium; RAFW site A/B=Right Atrium Free Wall site AD3 according to Figure 1; LAA=Left Atrial Appendage; NA=Not Available

atrial fibrillation were included in the study. The clinical characteristics of the patients are shown in Table I.

111. METHODS

A. Data Acquisition

Patients were anesthetized with fentanyl (10-30 pgkg), droperidol (2.5-5 mg), pancuronium bromide (4-10 mg) or alcuronium (5-15 mg), isoflurane (0.5 MAC) and nitrous oxide. After median sternotomy and exposure of the heart, epicardial recordings were made at predefined sites in the right atrial free wall and in the left atrial appendage, Fig. 1. The number of recordings made from each site in each patient is shown in Table I. The time between consecutive recordings from the same site was 1 min. Each recording consisted of 56 atrial bipolar electrograms, simultaneously acquired during 8 s. The electrode used to acquire data was a disposable intraoperative electrode array (4-cm Plaque Array, Bard Electrophysiology Division, C.R. Bard Inc., Haverhill, MA, USA). It consisted of 112 contact points (each 1 mm in diameter), forming 56 electrode pairs used as bipolar electrodes, placed in 7 columns with 8 bipolar electrodes in each column, Fig. 2. The distance between the electrodes in each pair varied between 2.8-3.2 mm with an average of 3 mm, the distance between each column between 5.6-6.2 mm with an average of 5.9 mm (= Ax), and the distance between each row between 2.9-3.3 mm with an average of 3.1 mm (= Ay). Each bipolar electrode was considered to receive signals from a measurement point spatially having its center

( 4 (b)

Fig. 1. Predefined sites from which we made repeated recordings of 56 simultaneously acquired bipolar electrograms each. Two adjacent sites, named A and B, at the right atrial free wall (a) and one site at the left atrial appendage (b). The electrode array was oriented with the cables parallel to the AV-groove (see Fig. 2).

in the middle between the electrode pair. The electrode array was connected to the 56-channel amplifier of a BARD Cardiac Mapping System, used to acquire and to temporarily store the signals obtained from the electrodes. The amplifiers limited the frequency response of the signals to 30-300 Hz. Signals were acquired with a sampling rate of 1 kHz and AD-converted with a resolution of 12 bits.

200 IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 43, NO 2, FEBRUARY 1996

t ( k ) = a g m i n (Id, - b(k)l - ( A ( k ) / 2 ) ) I t o t ( k ) ( k )

Fig. 2. A disposable electrode array, measuring 2.17 x 3.84 cm, used to acquire atrial bipolar electrograms. The electrode array (4 cm Plaque Array, Bard Electrophysiology Division, C.R. Bard Inc., Haverhill, MA, USA) consisted of 112 contact points (1 mm in diameter) forming 56 electrode pairs used as bipolar electrodes, placed in 7 columns with 8 bipolar electrodes in each column. The average distance between the electrodes in each pair was 3 mm (range 2.8-3.2 mm). The average distance between columns of electrode pairs was 5.9 mm (range 5.66.2 mm) and between rows of electrode pairs 3.1 mm (range 2.9-3.3 mm). Each bipolar electrode was considered to receive signals from a measurement point, spatially having its center in the middle between the electrode pair.

. (3)

Our choice of bipolar electrograms in this study was partly based on the fact that we experienced substantial difficulties acquiring noise-free unipolar electrograms in the operating room and, in addition, that no regard has to be taken to distant electrical phenomena in the analysis of bipolar electrograms.

B. Data Analysis 1 ) Detecting Local Activations and Estimating the Local

Activation Times: We defined the start of a local activation as the point in the electrogram with a slope, calculated as the absolute value of the difference between two consecutive samples, exceeding a certain threshold, typically 0.2 VIS, manually chosen for each electrogram in order to obtain optimal agreement with manual demarcation of the LA’S. The end of the local activation was defined as the closest following point in the electrogram that fulfilled the following criteria: 1) the slope of the curve at that point was to be less than the threshold and 2) no point of the curve, up to 20 ms ahead, was to have a slope exceeding the threshold. This latter criterion was added to ensure that an activation wave propagating with a velocity of at least 0.15 m / s , causing a deflection in the electrogram with two peaks and a maximum time interval of 20 ms between the peaks, would be regarded as part of the same LA. The time interval between the start and endmarker was to be not less than 8 ms to be regarded as a LA and not more than 70 ms to be regarded as part of the same LA.

Cubic spline interpolation [9] was applied between the start and endmarker and the local activation time was estimated as that point of the local activation that divided the area demarcated horizontally by the curve and the baseline and vertically by the start and endmarkers into two equally large parts. The baseline was estimated as a moving average of the first 15 of the 20 latest consecutive samples. Between the start and endmarker, the baseline was estimated as the value of the

HOLM et al.: A NEW METHOD FOR ANALYSIS OF ATRIAL ACTIVATION DURING CHRONIC ATRIAL FIBRILLATION IN MAN

~

20 1

(b) Fig. 3. A typical local activation in a bipolar atrial electrogram (interpolar distance 3 mm) from a patient with chronic atrial fibrillation. Upper panel (a) shows the baseline (marked line), the start ( t o ) and the end ( t l ) of the local activation whereas the lower panel (b) shows the area (shaded) of the local activation, used to estimate the local activation time. Parameters are calculated according to the text.

wave-front passing the measurement points propagating with a velocity of less than 0.15 m/s, with the implication that only activation waves causing local activations at all four measurement points (within the time window) were included in further calculations.

The average time differences in the x and y direction for each set of LA’S at the four adjacent measurement points were calculated according to:

From these time differences, we created an individual direction vector R(1) for each activation wave passing the set of mea- surement points. Individual direction vectors were corrected for the difference of spacing in the x and y-direction and normalized, according to the following equations:

Rsc(l) = Atz(Z) . (&//Ax) . ((At&) . (AY/A4)2 + Aty(1)2)-0.5

%(Z) = At,(Z) . ((At,(I) . ( A ~ / A E ) ) ~ + At,(1)2)-0.5 R(0 = (Rsc(l), Ry(0)’ (6)

where Ax = 5.9 mm and Ay = 3.1 mm. 3) Calculating the Conduction Velocity of Individual Activation Waves: Using the data from above, we calculated the average conduction velocity of each activation wave when passing by the set of measurement points. The following notations were used (see Fig. 4):

a(Z) the average angle between the propagation direction of wavefront number 1 and the x-axis when passing by the set of measurement points.

w(Z) average conduction velocity activation wave number 1 when passing by the set of measurement points. (7)

The angle was calculated as follows:

a(Z) = arctan ((At,(Z) . Ax)/(Atz(Z) . Ay)) -7r/2 5 Q 5 7r/2 (8)

In order to avoid division by zero, we distinguished between two cases, 5 7r/4 and 7r/4< IQI 5 7r /2 . The conduction velocity was calculated according to

v(Z) = ((COSQ . AZ)/A~,(Z)(, (a( 2 7r/4 w(1) = [(sins. Ay)/At,(Z)I, 7r/4< (a1 5 n/2. (9)

4) Time Averaging the Direction Vectors: To determine the preferable propagation direction of the activation waves, we calculated the average of all the individual direction vectors corresponding to the four adjacent measurement points. As each individual direction vector represented the propagation of activation waves at a certain time segment, the average was actually a time average. The circumstance that the individual direction vectors were normalized meant that the direction of the time-averaged direction vector indicated the dominating propagation direction whereas the magnitude indicated how strong the dominance was.

The calculations in the three latter sections (Sections 11-IV) were based on data from four adjacent measurement points. These calculations were applied to each set of four adjacent measurement points, a total of 42 sets in each recording. As each measurement point, except for those at the edge of each recording area, were included in four sets of measurement points, the direction vectors and the conduction velocities obtained in this way were not statistically independent. 5) Calculating Intervals Between Local Activations: We de- termined the periodicity of atrial activation by calculating the time interval between detected local activations at each measurement point according to

L y ( k . ) = t zy(k . ) - tzy(k. - 1) (10)

202 IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 43, NO. 2, FEBRUARY 1996

where Izy(k) is the time interval between local activation number IC and k - 1 corresponding to measurement point (x, y) and k = 1 . . . nZy. The average of the individual time intervals and its standard deviation were also calculated at each measurement point.

obtain detailed information about temporal changes in the activation pattern, the method of constructing consecutive activation maps with isochrones was applied in the recordings, using the Bard Cardiac Mapping System referred to above. The activation maps were based on manually determined LAT’s.

Thus, analysis of each recording resulted in: 1) A field of 42 time-averaged direction vectors, each indicating the preferable propagation direction of activation waves at a set of measure- ment points, 2) the conduction velocity of individual activation waves when passing by each set of measurement points, 3) the time intervals between local activations at each measurement point, and 4) a series of consecutive activation maps. The analysis, besides the construction of activation maps, was implemented in software developed at the department and run on a Pentium-based IBM-compatible personal computer. 7) Validating the Resultant Field of Time-Averaged Direction Vectors: The validation was done in two ways. First, the vector field was compared to a uniformly distributed vector field using the Kolmogorov-Smirnov two-sample test [ lo], in order to determine if the resultant vector field could have been produced by chance alone. The null hypothesis, HO: the resultant vector field is uniformly distributed, was tested at a significance level of 99%. The angles between each individual vector and the x-axis were used as test variables.

Secondly, a discretized divergence operator [ 111 was applied to the resultant vector field, in order to localize sites within the recording area that acted as a source or sink of activation waves. The following equation was used:

6) Constructing Consecutive Activation Maps: To

where D(x,y) is the divergence value and R , (z ,y ) and R,(x, y) are the LG and y-component, respectively, of the time- averaged direction vector from the four adjacent measurement points where point (LG, y) provides the left lower point.

Applying the discretized divergence operator to the vector field resulted in 5 x 6 values, each value representing the divergence of four adjacent vectors. A positive value indicated a source and a negative value a sink of activation waves. Significance limits for the divergence of the vector fields were determined by: 1) Creating 500 new vector fields, where each individual vector in a field was the average of 32 randomly chosen vectors; 2) applying the discretized divergence operator to these new vector fields; and 3) calculating a 99% confidence interval for the divergence values of the vector fields, using the circumstance that averaged values can be assumed to be normally distributed according to the central limit theorem of statistics [lo].

A I G I m o o o o o

0 0 0 0 0 0 0 0 0 0

0 0 0 0 0 0 0 0 0 0 0 0

I

AY ‘T Fig. 4. Four adjacent measurement points placed in an z and y coordinate system, where Ax and Ay are the average distances between columns and rows, respectively, of measurement points, cy is the angle between the propagation direction of an activation wave and the z-axis and ti is the average conduction velocity of the activation wave when passing by the four measurement points.

Beside the validation above, the reproducibility of the individual resultant fields of time-averaged direction vectors was evaluated by analysis of consecutive recordings from each recording site.

IV. RESULTS

A. Sinus Rhythm The time-averaged direction vector field and one activa-

tion map from a typical recording in a patient with stable sinus rhythm (patient #2) are shown in Fig. 5. Note that the time-averaged direction vector field (left panel) summarizes eight seconds of activation at the recording area whereas the activation map (right panel) only represents one beat, or 50 ms of activation. The right panel shows that the activation at that particular beat enters the recording area at the upper right comer, anatomically close to the sinus node, and is spread more or less radially with what appears to be a constant velocity. The time-averaged direction vectors in the left panel show the same activation pattern, and also provide the information that the activation pattern changes very little during the recording, as all the time-averaged direction vectors have a magnitude close to 1.

B. Atrial Fibrillation Recordings from seventeen patients, sixteen with chronic

atrial fibrillation and one with induced atrial fibrillation, have been successfully analyzed with the method. Another patient had to be excluded from the material as it was impossible

HOLM et al.: A NEW METHOD FOR ANALYSIS OF ATRIAL ACTIVATION DURING CHRONIC ATRIAL FIBRILLATION IN MAN

~

203

J J J ! 5 J

f < / J

J

J 5 5 J 1

J J J

f

“I 20 15 105 0

I-+ = magnitude of a normalized vector

Fig. 5. The resultant field of time-averaged direction vectors (left panel), and an activation map with isochrones (right pannel), where the field of time-averaged direction vectors represents atrial activity during 8 seconds and the activation map represents one atrial beat within the same 8 seconds, calculated from a recording in the right atrial free wall (site B according to Fig. 1) in a patient (#2) with stable sinus rhythm.

to analyze the recordings due to poor signal quality, i.e., low signaUnoise ratio. Based on the the pattems of the time- averaged vector fields, the patients were divided into three categories. Category A consists of patients where we did not find consistent activation in any of the analyzed areas of the atria, category B consists of patients where we found consistent activation in some area(s) of the atria and category C consists of patients where we in some area(s) of the atria found a focal activation pattern. Activation was considered to be consistent if the time-averaged vector field differed significantly ( P < 0.01) from a uniformly distibuted field (using the Kolmogorov-Smimov two-sample test described above). Furthermore, an activation pattern was considered to be focal if the discretized divergence operator (described above) found (a) significant ( P < 0.01) source@) of activation waves, i.e., (a) certain area(s) with positive divergence values exceeding the significance limit. Table 11 shows the category affiliation of the individual patients and the activation pattern of individual recordings in each patient. Note that the patients in category B did not necessarily have consistent activation in all the analyzed areas and, similarly, that the patients in category C only had a focal activation pattern in one of the analyzed areas. Consecutive recordings from the same record-

ing site consistently belonged to the same category, though the dominance of a certain consistent activation pattern (category B and C) changed, i.e., the magnitude of individual time- averaged direction vectors varied in consecutive recordings. By applying the method of constructing consecutive activation maps in the recordings belonging to category B and C, we found that the activation, either consistent with a preferential propagation direction or focal, occurred in recurrent episodes lasting from a few atrial activations (300-350 ms) up to 20 activations (3.1 s). The activation between the episodes of consistent activation was inconsistent with multiple present activation waves constantly activating nonrefractory areas of the atria, whereas the activation between the episodes of focal activation was consistent with one or two simultaneously present activation waves.

In Fig. 6(a)-(c) the results from typical recordings in pa- tients belonging to category A, B and C , respectively, are shown. The left panels show the time-averaged direction vector field of the recordings and the distribution of the individual di- rection vectors at each set of measurement points, respectively, while the right panels show strips of raw data from selected measurement points and an activation map with isochrones representing a short time window, respectively. In Fig. 7

204 EEE TRANSACTIONS ON BIOMEDICAL ENGINEEFUNG, VOL. 43, NO. 2, FEBRUARY 1996

TABLE II CATEGORY !LF’FILIATION OF EACH PATENT AND THE &FERABLE ACTIVATION PAITERN FOUM) IN EACH RECORDING

Preferable activation panem

Inconsistent activation Consistent activation Focal activation

# A B C site A site B siteA siteB siteA siteB

1 X 1 2 X 1 3 X 1 4 X 1 5 X 1

6 X 1 7 X 3 1 8 X 3 1 9 X 3 2 10 X 3 3

Patient Category RAFW RAFW LAA RAFW RAFW LAA RAFW RAFW LAA

i i X 3 3 I2 X 13 X 3 3 14 X

15 X 3

3 3

2 2 2 3 3

16 X 3 3 3 17 X 3 3 2 18 X 3 3 19 X 3 3 3 20 X 3 3 3

RAFW site AIS = Right Atrial Free Wall site AIB according to Figure 1; LAA = Left Atrial Appendage; The ‘x’ shows the category affilliation of each patient and the numbas (122) show m how may d g s we found each preferable activation pattern in each patient. Category A consists of patients where no cons- consists of patients where consistent activation with a preferable pPopagation direction was found in one or both the analyzed mas and ca&gory C consists of patients where consistent activation ariginating &om a localizable site and spread in a radial fashion, forming a focal activation pattem. was found in me of #e anatyzed areas.

was found in the analyzed areas, category €3

histograms of the intervals between consecutive activations at measurement point E4 (left panels) and the conduction velocities of the individual activation waves passing the set of measurement points where E4 provides the left lower point (right panels) are shown. Panel a-c corresponds to the recordings in Fig. 6(a)-(c). In Fig. 8 the divergence values of the time-averaged direction vector field in Fig. 6(c) are shown. The two planes parallel to the I(: and y-axis represent the significance limits ( P < 0.01) of the divergence values. The positive divergence values exceeding the positive significance limit correspond to the area with a focal activation pattem seen in Fig. 6(c).

V. DISCUSSION

A. Accuracy of the Results

The relative errors in the calculated parameters due to the variability in Ax and Ay were estimated using a standard consideration of relative errors. The variability in Ax was 5 0 . 3 mm or 415.1% and the variability in Ay was 50 .2 mm or &6.5%. The maximum relative error of the angle a (see Fig. 4) of individual direction vectors was 12.2% (when a was small) and, thus, the maximum relative error of the time- averaged direction vectors was also 12.2%, as the true values

of Ax and Ay at each set of measurement points were fixed throughout each recording. The maximum relative error of the conduction velocity of individual direction vectors was 12.8% for LY close to &n/4.

3. Noise Tolerance In rare cases two types of noise were present in the electro-

grams. Regular noise, predominantly 50 Hz noise, slightly af- fected the demarcation of the LA’s, which partly was avoided by choosing a higher threshold, but did not affect the baseline at all. As the areas at the start and end of a EA (see Fig. 3) only represent a minor portion of the total area, demarcation errors only had a limited effect on the estimate of the LAT’s (k2 ms). Irregular noise, or artifacts, affected both detection and demarcation of the LA’s and the baseline at those points in the electrogram where they appeared, which had a major affect on the estimate of the corresponding LAT’s. As the method includes time averaging of the direction vectors derived from the LAT’s, a limited number of artifacts did not affect the time-averaged direction vector field very much. However, due to artifacts, the conduction velocity of individual activation waves and individual time intervals between LA’s might have been over or underestimated.

HOLM et al.: A NEW METHOD FOR ANALYSIS OF ATRIAL ACTIVATION DURING CHRONIC ATRIAL FIBRILLATION IN MAN 205

03

04

09

EO

E4

E5

F3

F4

FS

d

4

2

4

5

0

-2

2

3

2

A 1 . E l

. Q . % . % . $ . 6 . @ .

. & . L 6 . V $9.

.%.a.&.%. % . B . * + * C % * ~ * * , ~ . d g . , % * & , % . 9 , 4 3 8 * % * @ * % * % * * * % *

A 0 . V . % . ~ . ~ . * . m . 08

4msbetweentheisochroIles

(a)

Fig. 6. Results from typical recordings in patients with chronic atrial fibrillation belonging to (a) category A (patient #6), (b) category B (patient #19), and (c) category C (patient #16), respectively. See text for definition of the categories. The upper panel in each subfigure shows the recording site. The middle left panel shows the time-averaged direction vector field of the recording. The magnitude of a normalized vecor is also shown in order to determine the dominance of individual propagation directions. The lower left panel shows a field of angle histograms, where each histogram in the field shows the distribution of the angle a (see Fig. 4) of individual direction vectors at four adjacent measurement points in the recording. The middle right panel shows short strips (230 ms) of electrograms from nine adjacent measurement points (indicated in the middle left panel) in the recording. The local activation times (plus signs) were manually set. The vertical caliper indicates the local activation times of electrode E4. The numbers to the left of each electrogram are the LAT’s relative to the LAT of electrode E4. The lower right panel shows the activation map corresponding to the time window in the middle right panel. Black arrow(s) indicate the conduction route(s) of activation waves.

C. Evaluating our Estimate of the U T ’ S estimates of LAT’s in bipolar electrograms, i.e., 1) the ‘max- imum peak’-estimate used in previous studies [12], [13], and

mapping during arrhythmia surgery. We randomly selected six recordings from the patients with chronic atrial fibrillation (from our material) and both our estimate and the other two

As there is no well established way to estimate the local

rillation, we decided to develop our own estimate, which we based on the morphology of the local activation. w e have evaluated our estimate by comparing it with previously used

activation times in bipolar electrograms acquired during fib- 2) the ‘maximum slope’-estimate, widely used in activation

206 IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL 43, NO 2, FEBRUARY 1996

lA AB ' - - magnitude o f a normalized vector.

I I

AB -

Q'

P b.

B ' 0' 8 '

GB

04

ffi

E3

E4

€5

t

F3

F4

F5

5 m between the isochrones

@)

Fig. 6. (Continued).

estimates were applied to the recordings, which contained a total of almost 17000 local activations. The differences between our estimate and the other two estimates were 0.22 k 6.1 ms (mean k S.D.) and 0.27 rt 7.1 ms, respectively, see Fig. 9 for histograms. We manually examined all electrograms with the LAT's according to the three estimates marked to find categories of LA's with systematic differences. We found that in LA's consisting of multiple peaks with different amplitude and polarity, representing approximately 30% of the LA'S in our material, there was a substantial difference between our estimate and the other two estimates, see examples in Fig. 10.

In these cases with fragmented LA's, it made more sense to use an estimate that considered the shape of the whole LA, not just one peak or one slope, as all myocardial cells activated at a measurement point contribute to the shape of the corresponding LA. This decision was supported by the results of a previous study where several automatic algorithms for estimating the local activation times, both in unipolar and in bipolar electrograms, were compared with manual detection by electrophysiologists [ 141. The investigators found that a morphological algorithm (that divided the LA's into phases and determined the LAT on the number and position of the

HOLM et al.: A NEW METHOD FOR ANALYSIS OF ATRIAL ACTIVATION DURING CHRONIC ATRIAL FIBRILLATION IN MAN 207

3 DO

D4

DS

E3

E4

€5

F3

F4

FS

__t

Fig. 6. (Continued).

phases) was superior to algorithms based on slope or amplitude criteria. Though, as clearly pointed out in the study, under certain conditions, e.g., very slow conduction or block, bipolar recordings with wide interpolar distance are not an adequate tool to delineate the spread of activation.

D. Exclusive Analysis of Activation Waves Causing LA’S at All Four Adjacent Measurement Points

As the atrial myocardium during fibrillation is activated by multiple simultaneously present activation waves, as docu-

mented by several investigators [2]-[4], it is in some cases very difficult to determine which LAT at a measurement point is caused by a certain activation wave. A criterion excluding sets of LA’s, corresponding to activation waves propagating with a velocity below 0.15 m/s, was included in our method. This cut-off point was chosen as a trade-off between including LA’s from activation waves with very low velocity and avoiding including sets of LA’s not caused by the same activation wave, but either by an activation waves being blocked due to local refractoriness of the myocardial cells or by two activation waves colliding or fusing in the

208 IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 43, NO. 2, FEBRUARY 1996

0 100 200 300 0 100 200 300 Interval [ms] Conduction velocity [cmls]

(a)

+ 8 $ 8 L 5 I 5

6 6

b 4

r c 6

z 4 E E

L

n Q

2 2 1 2

0 0 100 200 300 0 100 200 300

Interval [ms] ConductJon velocrty [cmls]

(b)

101 I 101 I

5 3

Columns 6 1 Rows

Fig. 8. The discretized divergence operator applied to the resultant field of time-averaged direction vectors (same as the left middle panel of Fig. 6(c) from a recording in a patient (#16) belonging to category C. Positive values indicate sources and negative values sinks of activation waves. The two planes parallel to the I and y-axis represent the significance limits ( P < O . O l ) calculated according to the text. The divergence values exceeding the positive significance limit correspond to the area with a focal activation pattern, shown in Fig. 6(c).

functional block from analysis might have masked this type of phenomenon. During random reentry, also described in [3], the site of block continuously changes, which means that the activation waves possibly excluded due to the above criterion were more or less randomly spread over the recording area and therefore did not affect the results.

0 100 200 300 0 100 200 300 Interval [ms] Conduction velocity [cmls]

Fig. 7. Histograms showing the distribution of the intervals between con- secutive activations at measurement point E4 (left panels) and the distribution of the conduction velocities of the individual activation waves passing the set of measurement points where E4 provides the left lower point (right panels), from the same recordings as shown in Fig. 6. The upper panel (a) corresponds to the recording in Fig. 6 a (patient #6), the middle panel (b) to Fig. 6(b) (patient #19) and the lower panel (c) to Fig. 6(c) (patient #16). In the right panel, the outliers representing very high conduction velocities (>1.5 d s ) are caused by artifacts in the electrograms distorting the estimate of individual local activation times.

zone between the four measurement points. The percentage of LA’S excluded from analysis due to this criterion varied between 0 and 30% in different recordings and patients. This criterion might have affected the results, i.e., the time-averaged direction vectors, if the excluded activation waves at a set of measurement points were consistently propagating in almost the same direction. This situation might have occurred during so-called leading circle reentry, where the impulse circulates around an arc of functional conduction block, and at least one measurement point in a set was situated in the area of block. Konings et al., reported that during induced AF in humans, the average persistence of a leading circuit reentry in the right atrial free wall varied between 2.4 and 5.4 beats depending on the complexity of the activation and, furthermore, that fixed reentrant circuits were never seen [3]. Assuming that the phenomenon of leading circuit reentry also occurred in our patients with chronic AF, the circumstance that the method quite likely excluded activation waves close to an arc of

E. Time Averaging

Due to the nature of fibrillation, with constantly changing conduction properties of the myocardium, the time window used to capture the characteristics of this arrhythmia has to be much longer than in the case of a monomorph arrhythmia, where a time window of one heart cycle is adequate. In the proposed method eight seconds of data were included in the time average. The results show that the pattern, i.e., the class@cation, was consistent in consecutive recordings from the same area, which suggests that eight seconds is enough to capture the characteristic activation patterns of the fibrillatory atria in patients with chronic atrial fibrillation. However, the fact that there were some differences between individual time- averaged direction vectors in consecutive recordings indicates that the fibrillatory process also includes more slowly varying phenomena that were not captured within the present time window.

F. Consistent Activation and Focal Activation As pointed out by Gerstenfeld et al. [5] , given the constant

anatomy, the myocardial fiber orientation, cellular uncoupling and fibrosis, the finding of nonrandom activation patterns in atrial fibrillation is not unexpected. Their finding that nine of 15 patients with chronic atrial fibrillation demonstrated evidence of linking is in agreement with our finding in the present study, where 11 of 17 patients with chronic AF (category B and C) demonstrated atrial areas with nonrandom activation. The finding that in four of these 12 patients the right atrial free wall at site B (see Fig. 1) exhibited a

HOLM et al.: A NEW METHOD FOR ANALYSIS OF ATRIAL ACTIVATION DURING CHRONIC ATRIAL FIBRILLATION IN MAN

+ 2500 4

5

L

2000 0

z 1500-

209

-

-

40001 35001

30001

500u -30 0 -20 -10

M e a n d i = O 2 2 m s -

S.D = 6.1 ms

1

0 10 20 30 Difference in IAT [ms] between our estimate and the “max peak“ estimate

(a)

40000 3000 35001 Mean diff.= 0.27 ms

S.D.= 7.1 ms

1

-30 -20 -1 0 0 10 20 Difference in LAT [ms] between our estlmate and the “max slope” estimate

(b)

Fig. 9. The difference in LAT between our estimate and the ‘maximum peak’ estimate (upper panel) and our estimate and the ‘maximum slope’ estimate (lower panel) in atrial electrograms from three patients with chronic atrial fibrillation. A total of almost 17 000 LA’S were included in the comparison.

focal activation pattern is more unexpected. At this stage we have to emphasize that, by using the terms ‘focal activation pattern’ and ‘focal activation’, we do not imply that the mechanism of this phenomenon is a focus, i.e., automatic or triggered activity. Previous studies of induced AF in both laboratory animals [2] and patients with WPW-syndrome [3] have reported the occurrence of new impulses originating from specific sites of the atrial myocardium, but just as solitary events. The investigators of these studies thought that the most likely explanation for this phenomenon was epicardial breakthrough of an impulse propagating in a free-running endocardial trabecula. Simultaneous epicardial and endocardial activation mapping in isolated canine right atria has shown that free-running trabecula, forming a three-dimensional pathway, may lead to intramural reentry, seen as a repetitive focal activation pattern both on the endocardial and epicardial surfaces [ 151.

t-------l 100 ms

(c)

Fig. 10. Identical strips of a bipolar atrial electrogram from a patient with chronic atrial fibrillation, with the LAT’s (plus signs) calculated according to (a) the ‘maximum peak’ estimate, (b) the ‘maximum slope’ estimate, and (c) our estimate.

VI. CONCLUSION

Analysis of atrial activation during fibrillation has until now been performed using the method of constructing consecu- tive activation maps with isochrones and, more recently, by temporal correlation of excitation directions at single atrial sites. To extract both spatial and temporal information on the fibrillatory process, we have developed a new method, using a temporal averaging technique, that identifies characteristic activation patterns of the atria, using data from multiple simultaneously acquired atrial electrograms. The method has proven to be a rapid and reliable tool for identifying atrial areas with consistent activation during fibrillation. By applying the method in patients with chronic atrial fibrillation, three types of preferable activation patterns were identified: 1) Inconsistent activation, 2) consistent (nonrandom) activation with a preferable propagation direction, and 3) consistent activation with impulses originating from a specific site within the recording area.

Using this new method, we have demonstrated a new finding in chronic atrial fibrillation in man-focal atrial activation occurring as a repetitive phenomenon. Further development of the method and the data acquisition system to global atrial analysis is essential in order to adequately characterize this arrhythmia.

210 IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 43, NO. 2, FEBRUARY 1996

ACKNOWLEDGMENT The authors wish to thank M. Magnusson for secretarial

assistance and B. Smideberg for help with illustrations.

REFERENCES

[ l ] G. K. Moe, “On the multiple wavelet hypothesis of atrial fibrillation,” Arch. Int. Pharmacodyn. Ther., vol. 140, pp. 183-188, 1962.

[2] M. A. Allessie, W. J. E. P. Lammers, F. I. M. Bonke, and J. Hollen, “Experimental evaluation of Moe’s multiple wavelet hypothesis of atrial fibrillation,” in Cardiac Arrhythmias, D. P. Zipes, J. Jalife, Eds. New York Grune and Stratton, 1985, pp. 265-276.

[3] K. T. S. Konings, C. J. H. J. Kirchhof, J. R. L. M. Smeets, H. J. J. Wellens, 0. C. Penn, and M. A. Allessie, “High-density mapping of electrically induced atrial fibrillation in humans,” Circ., vol. 89, no 4, pp. 1665-1680, 1994.

[4] J. L. Cox, T. E. Canavan, R. B. Schuessler, M. E. Cain, B. D. Lindsay, C. Stone, P. K. Smith, P. B. Corr, and J. P. Boineau, “The surgical treatment of atrial fibrillation 11. Intraoperative electrophysiologic mapping and description of the electrophysiologic basis of atrial flutter and atrial fibrillation,” J. Thorac. Cardiovasc. Surg., vol 101, pp. 406426, 1991.

[5] E. P. Gerstenfeld, A. V. Sahakian, and S. Swiryn, “Evidence for transient linking of atrial excitation during atrial fibrillation in humans,” Circ., vol 86, pp. 375-382, 1992. N. Cai, C. M. Dohnal, and S. B. Olsson, “Methodological asDects of

-

the use of heart rate stratified RR interval’histograms in-the analysis of atrioventricular conduction during atrial fibrillation,” Cardiovasc. Res., vol. XXI, no. 6, pp. 455462, 1987. S. B. Olsson, N. Cai, C. M. Dohnal, and G. William-Olson, “Effect of crista terminalis myotomy on the AV nodal function in atrial fibrillation,” in The Atrium in Health and Disease, P. Attuel, P. Coumel, M. J. Janse, Eds. S. B. Olsson, N. Cai, C. M. Dohnal, and K. K. Talwar, ‘Noninvasive support for and characterization of multiple intranodal pathways in patients with mitral valve disease and atrial fibrillation,” Eur. Heart

Mount Kisco, NY: Futura, 1989, pp. 249-261.

j . , vol. 7, pp. 320-333, 1986. New York Springer-Verlag,

1978. [lo] R. R. Sokal and F. J. Rohlf, Introduction to Biosraristics. New York:

W. H. Freeman, 1987. [11] M. R. Spiegel, Theory and Problems of Vector Analysis and an Intro-

duction to Tensor Analysis. New York: McGraw-Hill, 1974. [12] T. Sano, and A. M. Scher, “Multiple recording during electrically

induced atrial fibrillation,” Circ. Res., vol. XIV, pp. 117-125, 1964. [13] J. Wang, G. W. Bourne, Z. Wang, C. Villemaire, M. Talajic, and

S. Nattel, “Comparative mechanisms of antiarrhythmic drug action in experimental atrial fibrillation,” Circ., vol 88, no 3, pp. 103&1044, 1993.

[14] C. F. Pieper, R. Blue, and A. Pacifico, “Activation time detection algorithms used in computerized intraoperative cardiac mapping: A comparison with manually determined activation times,’’ J. Curdiovasc. Electrophsyio., vol. 2. pp. 388-397, 1991.

[15] R. B. Schuessler, T. Kawamoto, D. E. Hand, M. Mitsuno, B. I. Bromberg, J. L. Cox, and J. P. Boineau, “Simultaneous epicardial and endocardial activation sequence mapping in the isolated canine right atrium,” Circ., vol. 88, pp. 250-263, 1993.

[9] C. de Boor, A Practical Guide to Splines.

Magnus Holm (S’93-M’95) received the M.Sc degree in applied physics and electrical engineering from Linkoping Institute of Technology, Sweden in 1987. He is currently studying towards the Ph.D. degree at the Department of Cardiology, University Hospital and Department of Automatic Control, Lund Institute of Technology, Lund, Sweden.

His research mainly concerns the electrophysio- logical mechanisms of atrial fibrillation.

Rolf Johansson (S’81-M’82) received the M.S. degree in technical physics in 1977, the B.M. degree in 1980, the Ph.D. degree in control theory 1983. He was appointed Docent in 1985, and received the M.D. degree in 1986, all from Lund University, Lund, Scandinavia.

In 1985 he spent six months with Laboratoire d’Automatique de Grenoble, Grenoble, France Since 1986, he has been an Associate Professor of control theory at the Lund Institute of Technology. In his scientific work, he has been involved

in research in adaptive system theory, mathematical modeling, system identification, robotics, signal processing theory and hiomedxal research. Currently he is Coordinating Director of Robotics Research with participants from several departments of the Lund Institute of Technology. Another focus of interest is biomedical research at the Lund University Hospital, where he also has been an Advisor to students presenting their Ph D. dissertations.

In 1993 he published the book System Modeling and Ident$cation, Englewood Cliffs, NJ. Prentice-Hall.

S. Bertil Olsson is Professor and Head of Cardiol- ogy at the University Hospital, Lund, Sweden.

His research has mainly concerned mechanisms of and nonpharmacological treatment of different types of cardiac arrhythmias. Currently, he is lead- ing several Ph.D.-projects involving medical as well as technological aspects of atrial fibrillation.

Johan Brandt received the M D degree in 1984 and the Ph D. degree in 1992 from Lund University, Lund, Sweden

He became a Specialist in Cardio-Thoracic Surgery in 1992 and worked as a Senior Resident at the Department of Cardiothoracic Surgery between 1992-1994 at the University Hospital in Lund, Sweden. Since 1994 he is a Consultant Car&othoracic Surgeon at the University Hospital in Malmo, Sweden.

Carsten L a r s received the M D degree in 1977 He became a Specialist in Cardio-Thoracic

Surgery in 1984 and has worked as a Senior Resident at the Department of Cardio-Thoracic Surgery between 1984-1989 at the University Hospital in Lund, Sweden Since 1989 he is a Consultant Cardio-Thoracic Surgeon with special interest in arrhythma surgery and he is also involved in the research concerning the mechanisms of atnal fibnllation