IMBALANCES IN REGIONAL LUNG VENTILATION: A VALIDATION

STUDY ON ELECTRICAL IMPEDANCE TOMOGRAPHY

SUBJECT CATEGORY (DESCRIPTOR NUMBER): 2 (ALTERNATIVES: 145, 14, 9)WORD COUNT: 3622

AUTHORS:

VICTORINO, JOSUÉ A. 1

BORGES, JOÃO B. 1

OKAMOTO, VALDELIS N.1

MATOS, GUSTAVO F. J. 1

TUCCI, MAURO R. 1

CARAMEZ, MARIA P. R. 2

TANAKA, HARKI 1

SUAREZ SIPMANN, FERNANDO 3

SANTOS, DURVAL C. B. 4

BARBAS, CARMEN S. V. 1

CARVALHO, C. R. R. 1

AMATO, MARCELO B. P. 1

1. Respiratory ICU – Hospital das Clínicas – Pulmonary Department – University of São Paulo, Brazil

2. General ICU – Hospital das Clínicas – Emergency Clinics Division – University of São Paulo, Brazil

3. Department of Intensive Care. Fundación Jiménez Díaz, Madrid, Spain

4. Radiology Department – Hospital das Clínicas – University of São Paulo, Brazil

This article has an online data supplement, which is accessible from this issue's table of content online at www.atsjournals.org

FUNDED: FAPESP - Fundação de Amparo à Pesquisa do Estado de São Paulo (without any conflict of interest in the study outcome)

CORRESPONDENCE SHOULD BE ADDRESSED TO:

MARCELO AMATO, M.D.,Laboratório de Pneumologia - LIM09 Faculdade de Medicina da USPAv Dr Arnaldo, 455 sala 2206 (2nd floor)CEP: 01246-903 São Paulo, SP, BrazilEmail: amato@unisys.com.brPhone: 55-11-30667361 / FAX: 55-11-30612492

AJRCCM Articles in Press. Published on December 23, 2003 as doi:10.1164/rccm.200301-133OC

Copyright (C) 2003 by the American Thoracic Society.

1

ABSTRACT (WORD COUNT: 200)

Imbalances in regional lung ventilation, with gravity dependent collapse and

overdistention of nondependent zones, are likely associated to ventilator induced lung

injury. Electric impedance tomography is a new imaging technique potentially capable of

monitoring those imbalances. The aim of this study was to validate EIT measurements

of ventilation distribution, by comparison with dynamic computerized tomography in a

heterogeneous population of critically ill patients under mechanical ventilation. Multiple

scans with both devices were collected during slow-inflation breaths. Six repeated

breaths were monitored by impedance tomography, showing acceptable reproducibility.

We observed acceptable agreement between both technologies in detecting right-left

ventilation imbalances (bias = 0% and limits of agreement = -10 to 10%). Relative

distribution of ventilation into regions or layers representing one fourth of the thoracic

section could also be assessed with good precision. Depending on electrode

positioning, impedance tomography slightly overestimated ventilation imbalances along

gravitational axis. Ventilation was gravitationally dependent in all patients, with some

transient blockages in dependent regions synchronously detected by both scanning

techniques. Among variables derived from computerized tomography, changes in

absolute air-content best explained the integral of impedance changes inside regions of

interest (R2 ≥ 0.92). Conclusion: impedance tomography can reliably assess ventilation

distribution during mechanical ventilation.

KEY WORDS: Artificial Respiration; Physiologic Monitoring; Validation Studies; Adult

Respiratory Distress Syndrome; Respiratory Insufficiency.

2

Introduction (word count 446)

Patients under artificial ventilation often present heterogeneous lung aeration, with

inadequate distribution of tidal volume (1, 2). Prevalent conditions like increased lung

weight (3), lung compression by the heart (4, 5), abnormalities of chest wall (6, 7) and

impaired surfactant function (8) promote not only collapse of dependent lung zones, but

also hyperdistention of non-dependent zones (9-11). Such imbalances create zones of

stress concentration inside the parenchyma, with increased risks for ventilator induced

lung injury (12).

Although global indexes of lung function like blood gases (13, 14), lung mechanics (15,

16), and plethysmography (17) have been used to track those ventilatory imbalances,

they provide limited information. Imaging techniques like magnetic resonance (18) or

computerized tomography (CT) can provide better information about lung

heterogeneities (14, 19-21), but they lack the dynamic features and bedside monitoring

capabilities needed for intensive care.

Electrical Impedance Tomography (EIT) has emerged as a new imaging tool for bedside

use (22-25). It is a noninvasive and radiation free technique based on the measurement

of electric potentials at the chest wall surface. Within a particular cross-sectional plane,

harmless electrical currents are driven across the thorax in a rotating pattern, generating

a potential gradient at the surface, which is then transformed into a two-dimensional

image of the electric impedance distribution within the thorax.

Recent experimental studies have suggested that EIT images are very sensitive to

regional changes in lung aeration (26-32). The dynamic behavior and the qualitative

3

information extracted from EIT images look similar to that reported in dynamic CT

studies (2, 33, 34) or in ventilation scintigraphy (31, 35). Its potential use as an on-line

PEEP titration tool has also been proposed, since EIT apparently provides reliable

information about the recruitment/derecruitment of dependent lung regions (27, 28, 36, 37),

and thus about the associated risk of ventilator induced lung injury.

However, the poor spatial resolution of current EIT devices casts doubts on the

promises above. As EIT does not keep perfect anatomical correspondence with CT

images, we do not know yet whether we can translate the knowledge acquired from CT

studies to the EIT universe. Although a recent animal study (38) suggested a good linear

relationship between regional impedance changes and density changes (measured in

Hounsfield units), we do not know how to best use the quantitative pixel information

provided by EIT, nor how reliable it is in critically ill patients with acute lung injury.

We designed the present study to answer the questions above, and to test specifically if

EIT can consistently quantify ventilation imbalances caused by gravitational forces on

the injured lung. We also tested if some minimal anatomical /functional agreement with

dynamic CT images can be obtained in critically ill patients.

Part of this investigation has been previously reported in the form of abstracts (26, 39).

4

Methods: (word count = 924)

Ten adult patients under mechanical ventilation were recruited (table 1), after obtaining

informed consent from patients’ relatives.

Experimental protocol

Dynamic sequences of EIT and CT scans, repeatedly at the same thoracic plane, during

a slow-flow inflation maneuver were compared in supine patients. It was impossible to

obtain simultaneous EIT and CT images due to excessive electromagnetic interference.

Therefore, we performed three sets of slow-inflations in the ICU, monitored by EIT

(DAS-01P, Sheffield, UK), followed by one set monitored by CT (GE HighSpeed,

Milwalkee, USA). Back to ICU, three additional slow-inflations were again monitored by

EIT. By repeating EIT acquisition before and after patient transport to the CT room, we

fully tested EIT reproducibility.

In order to start lung inflations from same approximate resting volume, lung history was

homogenized before each one of the seven slow inflations, by applying CPAP of

40 cmH2O, lasting 20 seconds, followed by disconnection against atmosphere for 15

seconds.

The slow-inflation was initiated by directing a constant flow generator (1 L/min) towards

the endotracheal tube through a three-way stopcock, linked in series to a proximal

pressure/flow sensor. Data was sampled at 100 Hertz. Inflation stopped at 45 cmH2O,

enough to obtain approximately 100 EIT scans (0.8 image/second) or 45 CT scans (0.3

image/second).

5

We always started the slow-inflation 1.0 to 1.5 seconds before starting the first EIT or CT

scan. Hardware scanning-time was 1.0 second for both devices. Pressure/flow signals

were continuously stored (100 Hertz sampling) in a personal computer with its internal

clock previously synchronized with EIT and CT machine clocks.

Electrode-positioning

For EIT measurements, 16 standard electrocardiograph electrodes were placed around

the thorax, at the transverse plane crossing the 5th intercostal space at midclavicular

line. To check potential interferences of positioning of electrodes on image

reconstruction (figure 1), two different electrode-positioning arrangements were tested,

exactly at the same transverse plane:

a) standard positioning – equally spaced - the distance between two adjacent

electrodes kept constant along thoracic perimeter. The first electrode was always

placed at sternum.

b) test positioning – electrodes 5 (left armpit) and 13 (right armpit – figure 1) were

displaced upwards (3 cm), closer to anterior axillary line. Inter-electrode distances

were evenly shortened on anterior thoracic surface, and evenly expanded on

posterior surface.

Electrode positioning for the first scan was randomly selected. The second scan was

performed under the alternative positioning. For the third, no electrode replacements

were made. The previous set of electrodes was completely removed whenever we

changed electrode positioning or before transport to CT.

EIT scans

6

The EIT device injected an alternating current (51 kHz, 2.1 RMS) between sequential

pairs of adjacent electrodes. During each injection pattern, voltage differences between

adjacent pairs of non-injecting electrodes were collected. The first scanning cycle

worked as reference voltage set, with all image pixels (pixel = minimal element for image

reconstruction) assigned to zero. Subsequently, new scanning cycles were collected

every 1.2-second, each one providing information to reconstruct one new relative image.

By using long scanning time (1-second), impedance changes were mostly related to

changes in lung aeration, with negligible effects of perfusion waves (40-45). Each image

represented the relative change in impedance distribution within the transverse section

of the chest, from the first scan (right after starting slow-inflation) to current scan. Images

were reconstructed through a mathematical algorithm called back-projection (46, 47), in

which pixel values were expressed as percent changes of local impedance, not

providing any information about absolute values of tissue impedance. In its formulation,

the algorithm assumes that voltages were collected from a nearly rounded section of the

body, projecting its estimates of impedance changes over a 32 x 32 circular matrix. A

customized software automatically extracted pixel information from regions of interest

(ROIs) correspondent to those assigned on CT images (figure 2).

CT scans

After a new homogenizing maneuver, sequential CT slices (every 3 seconds, scanning-

time = 1-second) were taken during slow-inflation, without interruption and repeatedly at

the same cross-sectional plane defined for EIT. The collimation was set at 10 mm.

From each image, we obtained frequency distributions of CT numbers corresponding to

manually determined regions of interest, according to the topography shown in figure 2.

7

A customized software converted regional CT histograms into 3 derived variables:

mean-density, gas/tissue ratio and air-content, according to published formulas (48, 49).

Tidal volume distribution and statistical analysis

Retrospectively, we looked at airway flow tracings, identifying the start of slow-inflation

(error of ± 0.02 s). Using synchronized time information, we referenced EIT or CT scans

relative to this time origin. Off-line we synchronized EIT and CT acquisitions, by linearly

interpolating EIT image data to the same points in time where we had CT scans, getting

30-45 synchronized images per inflation. Since we used constant-flow generator, lungs

were inflated up to equivalent volumes for all matched images.

The relationship between CT and EIT variables was addressed by multiple linear

regression. By taking only the first and the last matched images, we calculated the

relative distribution of tidal volume across the ROIs. For CT, the percent of tidal

ventilation directed towards a particular ROI was calculated as the increment in air-

content for that ROI, divided by air-content increment for the whole slice. For EIT, we

took the last image and calculated the integral of pixel value over that corresponding

ROI (50, 51), divided by integral of pixel value over the whole slice. Based on these

estimates - presented as dimensionless numbers or percentages – we tested EIT

reproducibility (comparing EIT estimates before vs. after CT scan), and EIT vs. CT

agreement, according to the principles proposed by Bland and Altman (52).

Additional details are provided in the online data supplement.

8

RESULTS

Stability of lung mechanics along the study

Cross correlations among the 7 pressure-time curves obtained for each patient were

calculated. Since patient #2 presented at least one correlation coefficient < 0.9,

interpreted as a signal of poor stability of lung mechanics across the measurements, he

was excluded from subsequent analysis. Although discarded, the dynamics of lung

inflation in this case was illustrative of the spatial resolution of EIT, being presented in

the animations #1 and #2 in the online data supplement).

Reproducibility

Reproducibility in EIT estimates of tidal volume distribution was assessed by calculating

the within-subject-standard-deviation-between-repeated-measures (SW) (53). For each

ROI, we calculated SW and bias observed between two consecutive measurements,

always under the same electrode positioning. Considering all ROIs and both electrode-

positioning arrangements together, we observed global Sw of 4.9%, when electrodes

were kept in place, and 7.4% when we replaced electrode array after CT (separately

considered: 7.0% for standard, and 7.7% for test positioning). This demonstrates that

replacement of electrodes increased random errors in our measurements. The bias was

less than 1% for all situations.

All these results were below our a priori reproducibility cutoff of 9%.

Agreement

9

Agreement in estimates of tidal volume distribution according to EIT vs. CT is presented

in figure 3. Even smaller ROIs presented acceptable agreement (i.e. sample limits of

agreement did not exceed the boundaries established a priori) for either electrode-

positioning. Agreement was better for right-left imbalances than for upper-lower

imbalances in ventilation. The worst agreement was observed in layer 1, with standard

positioning (bias = + 9.4% and sample limits of agreement = - 6.4% to 25%).

Translating this agreement into images, figures 4 and 5 exemplify typical EIT images -

contrasted with synchronized CT images.

Relative distribution of tidal volume according to EIT and CT

Figure 6 shows the distribution of ventilation according to the horizontal and vertical

axes in CT and EIT images. When considering potential imbalances between right/left

fields, EIT and CT exhibited comparable estimates for regional ventilation (bias = 0%

and limits of agreement = -10% to 10%, P = 0.31, figure 6-left). Pooled measurements

across patients suggested a rather homogeneous (≅ 1:1) distribution of ventilation

between right/left fields. However, there were some outliers, exemplified by patient #8

(figure 4), who had a solid mass entirely blocking the right lung, and who obtained an

estimate of ventilation towards the right field = 2% in CT analysis, versus –3% in EIT

analysis. CT and EIT similarly detected all outliers.

Likewise, both techniques detected equivalent imbalances when the upper and lower

parts of the thorax were considered (upper/lower ratio = 82% / 18% and 75% / 25%, for

EIT and CT, respectively), also with a good case-by-case match. The overall

10

inhomogeneity between the upper/lower fields was marginally larger with EIT

(considering the standard electrode-positioning) than with CT (P = 0.04).

Similarly to CT, EIT detected a large vertical gradient of regional ventilation across the 4

superimposed layers in all patients. The standard positioning of electrodes caused a

slight overestimation of regional ventilation to layer 1, underestimating the ventilation to

layer 3. The test positioning partially corrected this distortion (figure 7).

Multiple regression analysis (figure 8) further checked two potential errors in EIT

analysis: (a) image distortions, and (b) lack of linear relationship between electrical

properties vs. density (X-ray attenuation) of tissues. We assumed CT based variables as

“gold standard” (independent variables), intentionally plotting the whole data sequence

for all regions together, in the same X-Y plane. We reasoned that both potential errors

were expected to compromise the overall coefficient of determination 1.

Figure 8 shows that air-content in CT presented best coefficient of determination (R2 =

0.92, standard positioning; R2 = 0.93, test positioning, not shown) to predict regional

impedance changes. Linear plots for each region were consistently observed, with very

similar slopes across regions and patients. The same was not true for the relationships

with CT mean-density (R2 = 0.57) or with gas/tissue ratio (R2 = 0.56), where different

slopes for each region compromised the overall coefficient of determination.

A common phenomenon observed in our patients was illustrated in figure 9. Dependent

lung zones presented transient blockage of regional ventilation at the beginning of slow-

inflation, although there was no visible collapse on CT at the start of slow-inflation. After

varied periods of time, this blockage was overcome and the slope of impedance

1 See complete footnote at page 29.

11

changes along time line suddenly increased in dependent zones, synchronously with the

sudden increase in air-content in dependent zones of CT slices.

DISCUSSION

The major findings in this study can be summarized as follows: a) EIT images from

patients under controlled mechanical ventilation were reproducible and presented good

agreement to dynamic CT scanning; b) Electrode array replacement slightly deteriorated

the reproducibility of EIT measurements, and the inter-electrode spacing within the array

affected the agreement with CT; c) Although EIT estimates of right/left imbalances in

regional lung ventilation were more precise (and less dependent on inter-electrode

spacing), gravity related imbalances of regional lung ventilation could be reliably

assessed, even for layers corresponding to one fourth of anteroposterior thoracic

distance, and d) Regional impedance changes in the EIT slice were best explained by

the corresponding changes in air-content detected in the CT slice (explaining 92-93% of

its variance). Other CT derived variables, like regional X-ray mean-density or regional

gas-tissue ratio, did not parallel regional changes in impedance as consistently.

An important methodological aspect of this study is linked to the results above: we used

the integral of pixel values over each ROI - instead of simple pixel average – to

represent the regional changes in impedance. There are several advantages with this

approach. First, some bench tests using back-projection reconstruction have

demonstrated the superior consistency of this parameter to quantify impedance

12

perturbations all over the image slice – independently of its radial position (50, 51).

Secondly, it allows the estimation of the percentage of tidal volume directed towards a

particular ROI by simply calculating a normalized ratio (i.e. the integral over the ROI

divided by the integral over the whole slice). This approach obviously decreases the

between-patient variability. And finally, there was a strong rationale supporting this

approach, particularly for our study, as explained bellow. Because clear anatomical

marks were absent in EIT images, we adopted a reproducible procedure for ROI

delineation, independently of investigator or individual anatomy: we embraced structures

suffering aeration together with structures that were not (e.g. the chest wall - figure 2).

Thus, the amount of non-expandable tissue (with fixed localized impedance) necessarily

attenuated the mean impedance change inside each ROI – in the same manner that

they attenuated mean-density changes on CT. However, the same attenuation is not

expected to occur in the integral of pixel values. Varied amounts of compact tissue do

not affect calculations for air-content in CT analysis (since their calculations are not

based on average values, but ultimately on the sum of pixel values), and similar results

must be expected for the integral of EIT pixel values.

When estimating air-content for each pixel in CT, we calculate the absolute amount of

air contained in the voxel (voxel = minimum volume element to construct the image).

The idea that the sum of these estimates produces a reliable number expressing air-

content inside the whole slice is intuitive. However, the understanding of how pixel

values in EIT - expressed as percent changes in impedance - can be summed up to

estimate global changes in air-content is not trivial. Recently, Nopp et al. (45) provided a

theoretical framework supporting this convenient relationship, which was explored in this

13

study as well as in a recent publication (54). Using an appropriate mathematical model

for the alveolar structure and boundary conditions - like an almost invariant interstitial

space along inspiration (i.e. constant tissue volume within the slice), projected over the

same image pixels, and suffering moderate impedance changes (< 100%) - the author

demonstrated that each percent change in pixel impedance should parallel absolute

increments in air-content for that corresponding parenchymal region. No matter the initial

value for absolute resistivity in that region.

It follows that the integral of pixel value in EIT should parallel changes in air-content, as

calculated in CT slices. However, the same rationale does not stand for gas/tissue ratio

(%), or CT mean-densities (Hounsfield units), as suggested above: different amounts of

compact tissue across different ROIs are expected to cause a poor correlation between

EIT and these two latter CT variables. Figure 8 corroborates this hypothesis.

This stronger association with CT air-content was a key finding in our study. Recently,

Frerichs et al (38) reported acceptable correlations between local impedance changes

versus local changes in CT mean-densities (in Hounsfield units). However, even using a

less noisy EIT device in a controlled environment (they used normal pigs with

convenient rounded thoracic geometry, instead of patients with diseased lungs and

trapezoid thoracic shapes) they reported lower coefficients of determination (ranging

from 0.56 to 0.86). Methodological differences like their subjective ROI demarcations

and the use of pooled regression, instead of a more appropriate within-subject

regression (55), make any comparison difficult. However, altogether, those findings

suggest that the choice for better parameters quantifying aeration in CT or EIT is

14

essential for fair comparisons between both technologies, or also to extract the most

reliable information from EIT.

Limitations of this study

Unlike the gold standard two-dimensional CT slice, with a homogenous thickness of

1 cm, EIT slice represents a less precise thickness of tissue, which is radius

dependent (56). Part of the electrical current commonly flows through planes above and

below the electrode plane, and the central part of the image is especially susceptible to

these out of plane influences, theoretically up to 10 cm above or below. Therefore, an

ideal comparison study should examine EIT against a thicker CT slicing (10-20 cm),

requiring more radiation and multislice tomography.

Nevertheless, the high within-subject coefficient of determination obtained with our

dynamic single-slice approach (R2 ≥ 0.92) suggests that even CT multislicing might not

provide much additional information. One possible explanation for this finding is that in

spite of theoretical assumptions, the amount of out of plane current may be negligible in

the human thorax. Another important consideration is that the lung may be relatively

homogeneous along the craniocaudal axis, behaving like a liquid body in patients under

mechanical ventilation (57). By assuming this isogravitational behavior, out of plane

changes would be similar to in-plane ones, minimally affecting our analysis (58).

Another limitation of our study might be related to the fact that EIT and CT acquisitions

were not simultaneous, and that the lung might behave slightly differently during each

slow-inflation (59). We tried to minimize this problem, contemplating procedures like the

exclusion of non-reproducible pressure-time tracings, the averaging of two EIT

15

acquisitions (before and after CT) for agreement analysis, and the use of intense

homogenizing maneuvers before each slow-inflation. Nevertheless, this intrinsic

limitation eventually precluded us from obtaining better agreement with CT slices.

Our final concern is that the presented results are only valid for the specific device

tested here and for ROIs not smaller than one fourth of the thoracic cross-sectional area.

These issues are linked, since technological improvements such as new image

reconstruction algorithms (60-67), larger number of electrodes (68), or higher precision in

current injection or voltage readings could all decrease errors in EIT imaging, improving

spatial resolution (69, 70). In fact, our reproducibility analysis suggests that we are close

to the resolution limits of the tested device and that any further decrease in ROI size

would impair reproducibility. As shown in this study, small differences in inter-electrode

spacing along the thoracic perimeter can have impact on EIT analysis (figure 7). Better

electrode-array handling (71) and new mathematical formulations to take into account

thoracic asymmetries are needed for the next years.

Implications of current data

Despite the limitations cited above, we think that the reported performance of EIT was

good enough for certain clinical applications, especially bedside adjustments of

mechanical ventilation with immediate feedback. A similar EIT device could easily detect

selective intubation, large pneumothorax or lobar atelectasis. Additionally, as already

reported by our group and others, subtle changes in PEEP level can produce large

imbalances in regional ventilation along the gravity axis, usually by the same order of

magnitude observed in the present study (27, 36).

16

Despite the low spatial resolution of current EIT devices, the high temporal resolution of

EIT looks promising. In our study, technical limitations forced us to use slow-motion

inflation of the lung, which, in turn, allowed us to detect transient and usually

imperceptible phenomena occurring during normal tidal breaths. For instance,

dependent zones in most patients presented complete blockage of ventilation during

significant part of inspiration (figure 9). Suddenly, 20-30 seconds later, some regional

ventilation could be precisely and simultaneously detected by EIT and CT – without

detectable perturbation in simultaneous pressure-time tracings. The clinical relevance of

such “inflation-delays” is a matter for future studies, but faster temporal resolutions in

new EIT devices would allow us to monitor such phenomena without any especial

maneuver. In the context of evidences suggesting deleterious effects of tidal

recruitment (72, 73), such sensitive detection at bedside is encouraging (26).

In conclusion, even at its current stage of development, EIT can reliably assess

imbalances in distribution of tidal volume in critically ill patients. When comparing

regional ventilation across different thoracic regions, the quantitative information

provided by EIT carries good proportionality to changes in air-content - as calculated by

dynamic CT scanning – but not with CT gas/tissue ratio or CT mean-densities.

17

ACKNOWLEDGMENTS:

We are grateful to the EIT Study Group (especially to Prof. Raul Gonzalez and the

team of the Polytechnic Institute and to Dra. Joyce Bevilacqua from the Applied

Mathematic Institute - University of São Paulo), for their valuable input, criticisms and

discussions during the experiments and data analysis.

18

REFERENCES:

1. Gattinoni L, Mascheroni D, Torresin A, Marcolin R, Fumagalli R, Vesconi S, Rossi

GP, Rossi F, Baglioni S, Bassi F, Nastri G, Pesenti A. Morphological response to

positive end expiratory pressure in acute respiratory failure: computerized

tomography study. Intensive Care Med 1986; 12: 137-142.

2. Gattinoni L, D'Andrea L, Pelosi P, Vitale G, Pesenti A, Fumagalli R. Regional effects

and mechanism of positive end-expiratory pressure in early adult respiratory distress

syndrome. JAMA 1993; 269: 2122-2127.

3. Pelosi P, D’Andrea L, Vitale G, Pesenti A, Gattinoni L. Vertical gradient of regional

lung inflation in adult respiratory distress syndrome. Am J Respir Crit Care Med 1994;

149: 8-13.

4. Malbouisson LM, Busch CJ, Puybasset L, Lu Q, Cluzel P, Rouby JJ. Role of the heart

in the loss of aeration characterizing lower lobes in acute respiratory distress

syndrome. CT Scan ARDS Study Group. Am J Respir Crit Care Med 2000; 161:

2005-2012.

5. Albert RK, Hubmayr RD. The prone position eliminates compression of the lungs by

the heart. Am J Respir Crit Care Med 2000; 161: 1660-1665.

6. Pelosi P, Cereda M, Foti G, Giacomini M, Pesenti A. Alterations of lung and chest

wall mechanics in patients with acute lung injury: effects of positive end-expiratory

pressure. Am J Respir Crit Care Med 1995; 152: 531-537.

7. Ranieri VM, Brienza N, Santostasi S, Puntillo F, Mascia L, Vitale N, Giuliani R,

Memeo V, Bruno F, Fiore T, Brienza A, Slutsky AS. Impairment of lung and chest wall

mechanics in patients with acute respiratory distress syndrome: role of abdominal

distension. Am J Respir Crit Care Med 1997; 156: 1082-1091.

8. Lewis JF, Jobe AH. Surfactant and the adult respiratory distress syndrome. Am Rev

Respir Dis 1993; 147: 218-233.

9. Vieira SR, Puybasset L, Richecoeur J, Lu Q, Cluzel P, Gusman PB, Coriat P, Rouby

JJ. A lung computed tomographic assessment of positive end-expiratory pressure-

induced lung overdistension. Am J Respir Crit Care Med 1998; 158: 1571-1577.

19

10. Gattinoni L, Pesenti A, Bombino M, Baglioni S, Rivolta M, Rossi F, Rossi G,

Fumagalli R, Marcolin R, Mascheroni D, Torresin A. Relationship between lung

computed tomographic density, gas exchange, and PEEP in acute respiratory failure.

Anesthesiology 1988; 69: 824-832.

11. Dambrosio M, Roupie E, Mollet JJ, Anglade MC, Vasile N, Lemaire F, Brochard L.

Effects of positive end-expiratory pressure and different tidal volumes on alveolar

recruitment and hyperinflation. Anesthesiology 1997; 87: 495-503.

12. Amato MBP, Marini JJ. Barotrauma, Volutrauma, and the Ventilation of Acute Lung

Injury. In: Marini JJ, Slutsky AS, editors. Physiological Basis of Ventilatory Support, 1

ed. New York: Marcel Dekker; 1998, p. 1187-1245.

13. Hedenstierna G, Tokics L, Strandberg A, Lundquist H, Brismar B. Correlation of gas

exchange impairment to development of atelectasis during anaesthesia and muscle

paralysis. Acta Anaesthesiol Scand 1986; 30: 183-191.

14. Malbouisson LM, Muller JC, Constantin JM, Lu Q, Puybasset L, Rouby JJ. Computed

tomography assessment of positive end-expiratory pressure- induced alveolar

recruitment in patients with acute respiratory distress syndrome. Am J Respir Crit

Care Med 2001; 163: 1444-1450.

15. Katz JA, Ozanne GM, Zinn SE, Fairley HB. Time course and mechanisms of lung-

volume increase with PEEP in acute pulmonary failure. Anesthesiology 1981; 54: 9-

16.

16. Bryan AC, Milic-Emili J, Pengelly D. Effect of gravity on the distribution of pulmonary

ventilation. J Appl Physiol 1966; 21: 778-784.

17. Brazelton TB, 3rd, Watson KF, Murphy M, Al-Khadra E, Thompson JE, Arnold JH.

Identification of optimal lung volume during high-frequency oscillatory ventilation

using respiratory inductive plethysmography. Crit Care Med 2001; 29: 2349-2359.

18. Tusman G, Böhm SH, Tempra A, Melkun F, Garcia E, Turchetto E, Mulder PGH,

Lachmann B. Effects of recruitment maneuvers on atelectasis in anesthetized

children. Anesthesiology 2003; 98: 14-22.

19. Puybasset L, Cluzel P, Gusman P, Grenier P, Preteux F, Rouby JJ. Regional

distribution of gas and tissue in acute respiratory distress syndrome. I. Consequences

20

for lung morphology. CT Scan ARDS Study Group. Intensive Care Med 2000; 26:

857-869.

20. Brismar B, Hedenstierna G, Lundquist H, Strandberg A, Svensson L, Tokics L.

Pulmonary densities during anesthesia with muscular relaxation - a proposal of

atelectasis. Anesthesiology 1985; 62: 422-428.

21. Gattinoni L, Pelosi P, Crotti S, Valenza F. Effects of positive end-expiratory pressure

on regional distribution of tidal volume and recruitment in adult respiratory distress

syndrome. Am J Respir Crit Care Med 1995; 151: 1807-1814.

22. Barber DC, Brown BH. Applied Potential Tomography. J Phys E Sci Instrum 1984; 17:

723-733.

23. Brown BH, Barber DC, Seagar AD. Applied potential tomography: possible clinical

applications. Clin Phys Physiol Meas 1985; 6: 109-121.

24. Adler A, Shinozuka N, Berthiaume Y, Guardo R, Bates JH. Electrical impedance

tomography can monitor dynamic hyperinflation in dogs. J Appl Physiol 1998; 84:

726-732.

25. Adler A, Amyot R, Guardo R, Bates JH, Berthiaume Y. Monitoring changes in lung air

and liquid volumes with electrical impedance tomography. J Appl Physiol 1997; 83:

1762-1767.

26. Borges JB, Okamoto VN, Tanaka H, Janot GF, Victorino JA, Tucci MR, Carvalho

CRR, Barbas CSV, Amato MBP. Bedside detection of regional opening/closing

pressures by electrical impedance vs. computed tomography (CT) (abstract). Am. J.

Respir. Crit. Care Med. 2001; 163 (Suppl): A755.

27. Kunst PW, Vazquez de Anda G, Bohm SH, Faes TJ, Lachmann B, Postmus PE, de

Vries PM. Monitoring of recruitment and derecruitment by electrical impedance

tomography in a model of acute lung injury. Crit Care Med 2000; 28: 3891-3895.

28. Kunst PW, Bohm SH, Vazquez de Anda G, Amato MB, Lachmann B, Postmus PE, de

Vries PM. Regional pressure volume curves by electrical impedance tomography in a

model of acute lung injury. Crit Care Med 2000; 28: 178-183.

29. Frerichs I, Hahn G, Schroder T, Hellige G. Electrical impedance tomography in

monitoring experimental lung injury. Intensive Care Med 1998; 24: 829-836.

21

30. Frerichs I, Hahn G, Hellige G. Gravity-dependent phenomena in lung ventilation

determined by functional EIT. Physiol Meas 1996; 17(Suppl 4A): A149-157.

31. Hinz J, Hahn G, Neumann P, Sydow M, Mohrenweiser P, Hellige G, Burchardi H.

End-expiratory lung impedance change enables bedside monitoring of end- expiratory

lung volume change. Intensive Care Med 2003; 29: 37-43.

32. Frerichs I, Schiffmann H, Oehler R, Dudykevych T, Hahn G, Hinz J, Hellige G.

Distribution of lung ventilation in spontaneously breathing neonates lying in different

body positions. Intensive Care Med 2003; 29: 787-794.

33. Neumann P, Berglund JE, Andersson LG, Maripu E, Magnusson A, Hedenstierna G.

Effects of inverse ratio ventilation and positive end-expiratory pressure in oleic acid-

induced lung injury. Am J Respir Crit Care Med 2000; 161: 1537-1545.

34. Neumann P, Berglund JE, Mondejar EF, Magnusson A, Hedenstierna G. Effect of

different pressure levels on the dynamics of lung collapse and recruitment in oleic-

acid-induced lung injury. Am J Respir Crit Care Med 1998; 158: 1636-1643.

35. Kunst PW, Vonk Noordegraaf A, Hoekstra OS, Postmus PE, de Vries PM. Ventilation

and perfusion imaging by electrical impedance tomography: a comparison with

radionuclide scanning. Physiol Meas 1998; 19: 481-490.

36. Sipmann FS, Borges JB, Caramez MP, Gaudencio AMAS, Barbas CSV, Carvalho

CRR, Amato MBP. Selecting PEEP during mechanical ventilation by using electrical

impedance tomography (EIT) (abstract). Am J Resp Crit Care Med 2000; 161(Suppl):

A488.

37. Genderingen HR, Vught AJ, Jamsen JRC. Estimation of regional volume changes by

electrical impedance tomography during a pressure-volume maneuver. Intensive

Care Med 2003; 29: 233-240.

38. Frerichs I, Hinz J, Herrmann P, Weisser G, Hahn G, Dudykevych T, Quintel M,

Hellige G. Detection of local lung air content by electrical impedance tomography

compared with electron beam CT. J Appl Physiol 2002; 93: 660-666.

39. Victorino JA, Okamoto VN, Borges JB, Janot GF, Tucci MR, Carvalho CRR, Amato

MBP. Disturbed regional ventilation in ALI/ARDS assessed by EIT: functional CT

validation (abstract). Am J Respir Crit Care Med 2003; 167(Suppl.): A618.

22

40. Dijkstra AM, Brown BH, Leathard AD, Harris ND, Barber DC, Edbrooke DL. Clinical

applications of electrical impedance tomography. J Med Eng Technol 1993; 17: 89-

98.

41. Grundy D, Reid K, McArdle FJ, Brown BH, Barber DC, Deacon CF, Henderson IW.

Trans-thoracic fluid shifts and endocrine responses to 6 degrees head- down tilt.

Aviat Space Environ Med 1991; 62: 923-929.

42. Barber DC. Quantification in impedance imaging. Clin Phys Physiol Meas 1990;

11(Suppl A): 45-56.

43. Harris ND, Suggett AJ, Barber DC, Brown BH. Applied potential tomography: a new

technique for monitoring pulmonary function. Clin Phys Physiol Meas 1988; 9(Suppl

A): 79-85.

44. Harris ND, Suggett AJ, Barber DC, Brown BH. Applications of applied potential

tomography (APT) in respiratory medicine. Clin Phys Physiol Meas 1987; 8(Suppl A):

155-165.

45. Nopp P, Harris ND, Zhao TX, Brown BH. Model for the dielectric properties of human

lung tissue against frequency and air content. Med Biol Eng Comput 1997; 35: 695-

702.

46. Barber DC. A review of image reconstruction techniques for electrical impedance

tomography. Med Phys 1989; 16: 162-169.

47. Santosa F, Vogelius M. A backprojection algorithm for electrical impedance imaging.

SIAM J Appl Math 1990; 50: 216-243.

48. Hounsfield GN. Computerized transverse axial scanning (tomography). 1. Description

of system. Br J Radiol 1973; 46: 1016-1022.

49. Gattinoni L, Caironi P, Pelosi P, Goodman LR. What has computed tomography

taught us about the Acute Respiratory Distress Syndrome? Am J Respir Crit Care

Med 2001; 164: 1701-1711.

50. Thomas DC, Disall-Allum JN, Sutherland IA, Beard RW. The use of EIT to assess

blood volume changes in the female pelvis. In: Holder D, editor. Clinical and

physiological applications of electrical Impedance tomograph. London: UCL Press;

1993, p. 227-233.

23

51. Thomas DC, Siddall-Allum JN, Sutherland IA, Beard RW. Correction of the non-

uniform spatial sensitivity of electrical impedance tomography images. Physiol Meas

1994; 15(Suppl 2a): A147-A152.

52. Mantha S, Roizen MF, Fleisher LA, Thisted R, Foss J. Comparing methods of clinical

measurement: reporting standards for Bland and Altman analysis. Anesth Analg

2000; 90: 593-602.

53. Bland JM, Altman DG. Measurement error. BMJ 1996; 312: 1654.

54. Hinz J, Neumann P, Dudykevych T, Andersson LG, Wrigge H, Burchardi H,

Hedenstierna G. Regional ventilation by electrical impedance tomography. A

comparison with ventilation scintigraphy in pigs. Chest 2003; 124: 314-322.

55. Bland JM, Altman DG. Calculating correlation coefficients with repeated observations:

Part 1 - Correlation within subjects. BMJ 1995; 310: 446.

56. Blue RS, Isaacson D, Newell JC. Real-time three-dimensional electrical impedance

imaging. Physiol Meas 2000; 21: 15-26.

57. Borges JB, Janot GF, Okamoto VN, Caramez MPR, Barros F, Souza CE, Carvalho

CRR, Amato MBP. Is there a true cephalocaudal lung gradient in ARDS? (abstract).

Am J Respir Crit Care Med 2003; 167(Suppl): A740.

58. Borges JB, Janot GF, Okamoto VN, Barros F, Souza CE, Carvalho CRR, Barbas

CSV, Amato MBP. Aggressive recruitment maneuvers assessed by helical CT: is it

necessary multiple slices? (abstract). Am J Respir Crit Care Med 2002; 165(Suppl):

A680.

59. Suki B. Fluctuations and power laws in pulmonary physiology. Am J Respir Crit Care

Med 2002; 166: 133-137.

60. Cohen-Bacrie C, Goussard Y, Guardo R. Regularized reconstruction in electrical

impedance tomography using a variance uniformization constraint. IEEE Trans Med

Imaging 1997; 16: 562-571.

61. Vauhkonen M, Lionheart WR, Heikkinen LM, Vauhkonen PJ, Kaipio JP. A MATLAB

package for the EIDORS project to reconstruct two-dimensional EIT images. Physiol

Meas 2001; 22: 107-11.

24

62. Molinari M, Cox SJ, Blott BH, Daniell GJ. Adaptive mesh refinement techniques for

electrical impedance tomography. Physiol Meas 2001; 22: 91-96.

63. Dehghani H, Barber DC, Basarab-Horwath I. Incorporating a priori anatomical

information into image reconstruction in electrical impedance tomography. Physiol

Meas 1999; 20: 87-102.

64. Avis NJ, Barber DC. Image reconstruction using non-adjacent drive configurations.

Physiol Meas 1994; 15(Suppl 2a): A153-A160.

65. Blott BH, Daniell GJ, Meeson S. Electrical impedance tomography with compensation

for electrode positioning variations. Phys Med Biol 1998; 43: 1731-1739.

66. Edic PM, Isaacson D, Saulnier GJ, Jain H, Newell JC. An iterative Newton-Raphson

method to solve the inverse admittivity problem. IEEE Trans Biomed Eng 1998; 45:

899-908.

67. Mueller JL, Siltanen S, Isaacson D. A direct reconstruction algorithm for electrical

impedance tomography. IEEE Trans Med Imaging 2002; 21: 555-559.

68. Tang M, Wang W, Wheeler J, McCormick M, Dong X. The number of electrodes and

basis functions in EIT image reconstruction. Physiol Meas 2002; 23:129-140.

69. Cheng KS, Simske SJ, Isaacson D, Newell JC, Gisser DG. Errors due to measuring

voltage on current-carrying electrodes in electric current computed tomography. IEEE

Trans Biomed Eng 1990; 37: 60-65.

70. Cook RD, Saulnier GJ, Gisser DG, Goble JC, Newell JC, Isaacson D. ACT3: a high-

speed, high-precision electrical impedance tomograph. IEEE Trans Biomed Eng

1994; 41: 713-722.

71. McAdams ET, McLaughlin JA, Mc CAJ. Multi-electrode systems for electrical

impedance tomography. Physiol Meas 1994; 15(Suppl 2a): A101-A106.

72. Muscedere JG, Mullen JBM, Slutsky AS. Tidal ventilation at low airway pressures can

augment lung injury. Am J Respir Crit Care Med 1994; 149: 1327-1334.

73. Taskar V, John J, Evander E, Robertson B, Jonson B. Surfactant dysfunction makes

lungs vulnerable to repetitive collapse and reexpansion. Am J Respir Crit Care Med

1997; 155: 313-320.

25

TABLE AND FIGURE LEGENDS:

Table 1: Patient characteristics at entry

M = male; F = female; COPD = chronic obstructive pulmonary disease; AIDS = acquired

immunodefiency syndrome; PCP = Pneumocystis carini pneumonia; VT = tidal volume;

CST = respiratory system compliance (average slope of the inflation P-V curve, from zero

to 30 cmH2O)

Exclusion criteria: contraindications for sedation, paralysis or hypercapnia, and presence

of bronchopleural fistula.

Figure 1: Sketch of thoracic plane and theoretical effects of different electrode

positioning. During EIT imaging, impedance changes occurring in real trapezoid domain

(left) are projected over a circular EIT domain (right), deforming lung areas. The

perimeter of the EIT circle necessarily corresponds to the skin with electrodes. By using

standard electrode-positioning (top), mid-electrode 5 is frequently placed over the skin

close the posterior lung (LLL illustrates an atelectatic left lower lobe), and not at mid-lung

height. Since the EIT imaging algorithm assumes that electrode 5 is at mid-thoracic-

height, midway between electrodes 1 and 9, there is some shrinking of nondependent

lung representation, with expansion of LLL. This is because EIT back-projection

assumes that every lung tissue above electrode 5 must be projected over the anterior

26

part of the circular EIT representation, whereas every tissue below electrode 5 (LLL) has

to occupy the whole posterior part of the circular EIT representation.

The test positioning used in this study is illustrated at the bottom. Electrode-5 is

displaced ventrally (3 cm), closer to mid-lung height. Electrodes 1-5 have now a shorter

inter-electrode distance than electrodes 5-9 (where subcutaneous tissue is abundant).

We hypothesized that such positioning would avoid the over-representation of LLL.

Figure 2: Schematic regions of interest (ROI) on CT and EIT. Each one of the 12 ROIs

embraced a portion of the chest wall plus part of the lung. On EIT images, the portions

were selected automatically by a special software splitting the original circle with 788

pixels into subsets shown. On CT image, the skin border was manually designed,

forming the outer boundary of a cross section of the thorax with approximately 140.000

pixels. Subsequently, uppermost and lowest pixels of contour were taken as references,

and four evenly spaced layers, each one corresponding to one fourth of antero-posterior

thoracic diameter, were drawn. Similarly, the crossing between skin contour and the

horizontal line at mid thoracic height defined references to split the thorax in its middle

(left and right halves). URQ = upper right quadrant; ULQ = upper left quadrant; LRQ =

lower right quadrant; LLQ = lower left quadrant.

Figure 3: Bland-Altman plots of the differences in regional distribution of tidal volume,

estimated by EIT and CT. (for brevity, only 6 representative lung regions, and only the

standard electrode-positioning is presented). The overall span of Y-axis represents our a

priori limits of agreement. Representation: dotted lines - limits of agreement of the

observed sample; gray line - mean of observed differences; downward triangles -

27

measurements taken before CT exam; upwards triangles - measurements taken after

CT exam; circles - the average difference; gray square - patient #9, presenting acute

cardiogenic pulmonary edema soon after CT scan, resulting in the largest disagreement.

Figures 4 and 5: EIT and CT images obtained at start (left) and at end (right) of slow-

inflation maneuver. Both images were obtained at the same thoracic plane (5th

intercostal space). The relative EIT images express only variations in impedance. Note

that the back-projection algorithm projects impedance changes (bright colors represent

increased impedance) onto the same quadrants suffering higher aeration in CT images

(regions getting darker shades of gray). Atelectatic zones, the mediastinum and the

pleural effusion zones remain silent. F.R.C.= Functional residual capacity.

Figure 6: Box-plot representing distributions of tidal volume estimated by EIT (white

boxes) and CT (gray boxes) in 9 patients, when using standard electrode-positioning.

Boxes indicate 25% and 75% percentiles, with median line inside. Error bars represent

5% and 95% percentiles. Left panel points out ventilation imbalances between left and

right thoracic areas. Right panel points out imbalances between upper and lower parts of

the thorax.

*: P = 0.04 using asymptotic approximation for Wilcoxon Signed-rank test.

†: P = 0.02 using asymptotic approximation for Wilcoxon Signed-rank test.

Figure 7: Box-plot representing distributions of tidal volume estimated by EIT (white

boxes) and CT (gray boxes). Electrode-positioning was standard in the top panel and

28

test in the bottom. Boxes represent 25% and 75% percentiles, with median line inside.

Error bars represent 5% and 95% percentiles. There is an overall trend for progressively

lower ventilations from ROI 1 to ROI 4, either in EIT (P = 0.001, Friedman test) or in CT

(P = 0.003). The test positioning of electrodes resulted in better match with CT.

Significant differences between CT and EIT estimates were detected only for the

standard positioning.

*: P = 0.018 using asymptotic approximation for Wilcoxon Signed-rank test.

†: P = 0.028 using asymptotic approximation for Wilcoxon Signed-rank test.

Figure 8: Scattered plots illustrating adjusted multiple regression for local impedance

changes during slow-inflation (standard electrode-positioning), when projected over

synchronized changes in CT images. Dependent variable in each plot represents

integral of pixel values over a certain ROI in EIT, calculated for each image in the

inflation sequence. Independent variables are, respectively: regional air-content,

regional gas/tissue ratio (middle), and regional mean-density (bottom), all calculated

from corresponding ROIs in CT images. Plots for all ROIs, patients and trials (pre or post

CT) are superposed, with approximately 9600 data points per graph. R2 represents

within-subject coefficient of determination.

Figure 9: Temporal sequence of EIT and CT estimates during slow-lung inflation. The

arrow indicates the moment when the lower half of the thorax started to ventilate, almost

30 seconds later than the upper thorax.

29

FOOTNOTES:

1: EIT image distortions tend to produce plots with different slopes for each region.

Consider an EIT slice with adjacent regions A and B. Consider also X-axis

representing true changes in air-content for regions A or B (measured by CT) versus

Y-axis representing the measured changes in impedance. If part of a true impedance

change in region “A“ was wrongly projected over region “B” (characterizing an image

distortion), the slope of plot “A” would decrease, whereas the slope of plot “B” would

increase in the same system of coordinates. The result would be a poor R2.

30

Table 1

Patient Age Gender Diagnosis APACHE IIPaO2/FiO2

PEEPVT

( mL )CST

( mL/cmH2O )

Days on Mech.Vent.

1 61 M COPD, right lobectomy, sepsis

18 289 5 540 61 3

2 44 M stroke, alcohol abuse, sepsis

35 222 15 480 35 8

3 52 M Hemothorax, sepsis, pneumonia

8 114 14 460 59 8

4 43 F COPD, AIDS, PCP 15 176 15 500 60 3

5 36 M AIDS, miliary tuberculosis, PCP

12 270 15 290 17 10

6 43 M Hystoplasmosis, tuberculosis

13 188 16 250 56 8

7 39 M Pulmonary carcinoma, sepsis

22 192 18 400 17 9

8 36 M Non-Hodgkin lymphoma, sepsis

11 237 20 350 47 4

9 60 F Congestive heart failure, lung edema

22 240 15 500 68 8

10 31 M Systemic lupus, miliary tuberculosis

19 227 13 370 22 20

31

Figure 1:

1 2

3

6

4

7

5

89

LLL LLL

1 2

3

4

5

6

789

CT EIT

LLL

1 2

3

4

5

6

789

1 23

6

4

7

5

89

LLL

Test positioning

Standard positioning

13

13

32

Figure 2:

Layer 1

Layer 2

Layer 3

Layer 4

URQ ULQ

LRQ LLQ

UPPER

LOWER

RIGHT LEFT

CT EIT

RIGHT LEFT

UPPER

LOWER

URQ ULQ

LRQ LLQ

Layer 1

Layer 2

Layer 3

Layer 4

33

Figure 3:

RIGHT LUNG Layer 2UPPER LUNG

LOWER LUNG Layer 4LEFT LUNG

Mean VT Distribution (EIT and CT)

0.50

0.25

0

-0.25

-0.50

0.50

0.25

0

-0.25

-0.50

0.80.60.40.2 1.00.0 0.80.60.40.2 1.00.0 0.80.60.40.2 1.00.0

Diff

eren

ce

( E

IT -

CT

)D

iffer

ence

(

EIT

-C

T )

0.25

0

-0.25

0.25

0

-0.25

Mean VT Distribution (EIT and CT) Mean VT Distribution (EIT and CT)

34

Figure 4:

F.R.C. FULL INSPIRATION

35

Figure 5:

F.R.C. FULL INSPIRATION

36

Figure 6:

0.0

0.2

0.4

0.6

0.8

1.0

0.0

0.2

0.4

0.6

0.8

1.0

CTEIT ( Standard Positioning )

0.0

0.2

0.4

0.6

0.8

1.0

RIGHT LEFT UPPER LOWER

RE

GIO

NA

L V

EN

TIL

AT

ION

(

perc

ent

of v

ent il

atio

n p

er s

lice

)

†

*

37

Figure 7:

0.0

0.2

0.4

0.6

0.8R

EG

ION

AL

VE

NT

ILA

TIO

N

( pe

rce n

t of

ven

tilat

ion

per

slic

e ) Roi 1 Roi 2 Roi 3 Roi 4

0.0

0.2

0.4

0.6

0.8

RE

GIO

NA

L V

EN

TIL

AT

ION

(

perc

e nt

of v

entil

atio

n p

er s

lice

)

CTEIT

NON

DEPENDENTDEPENDENT

†*

Standard( equally spaced )

positioning

Testpositioning

38

Figure 8:

-0.05 0.00 0.05 0.10 0.15 0.20 0.25-10000

0

10000

20000

30000

40000

50000

6000060

40

20

0

0 0.05 0.15 0.25-0.05

Changes in Gas – Tissue ratio

R2 =0.560.200.10

∆Z

(pi

xeli

nteg

ral )

-300 -200 -100 0

0

20000

40000

60000

Changes in Mean Density (HU)

60

40

20

0

-300 -200 -100 0

R2 =0.57

∆Z

(pi

xeli

nteg

ral )

0 20 40 60 80

0

20000

40000

6000060

40

20

0

20 40 60 800

Changes in Air - Content (mL)

R2 =0.92∆Z

(pi

xeli

nteg

ral )

39

Figure 9:

0 10 20 30 40 50 60

0

5

10

15

20

25

30

Inflation Time (seconds)

0

10

20

30

40

50

60

Air

(mL)

ent

erin

g th

e C

T s

lice R

elative impedance change

(pixelintegral)

EIT scan

CT scan

70 80

UPPER HALF

LOWER HALF

or

LU

NG

VO

LU

ME

AIRWAY PRESSURE

0 10 20 30 40

" ON LINE DATA SUPPLEMENT "

IMBALANCES IN REGIONAL LUNG VENTILATION: A VALIDATION

STUDY ON ELECTRICAL IMPEDANCE TOMOGRAPHY

AUTHORS:

VICTORINO, JOSUÉ A. 1

BORGES, JOÃO B. 1

OKAMOTO, VALDELIS N.1

MATOS, GUSTAVO F. J. 1

TUCCI, MAURO R. 1

CARAMEZ, MARIA P. R. 2

TANAKA, HARKI 1

SUAREZ SIPMANN, FERNANDO 3

SANTOS, DURVAL C. B. 3

BARBAS, CARMEN S. V. 1

CARVALHO, C. R. R. 1

AMATO, MARCELO B. P. 1

1

METHODS:

This study took place at the Respiratory and Medical ICU’s of Hospital das Clínicas,

University of São Paulo, Brazil. Patients were transported to the Radiology Department

for CT scanning. Data was collected from July to December 2001.

PATIENT SELECTION

The study protocol was approved by the hospital ethics committee.

Hemodynamically stable patients, under controlled mechanical ventilation were eligible

for study. Patients were kept supine, under sedation and paralysis during the study

period. They were monitored with ECG, pulse oxymetry and non-invasive blood

pressure. An independent team of physicians was responsible for patient care, with

autonomy to withdraw patients from the study.

MONITORING AND EQUIPMENT

Disposable flow/presssure sensors were connected to proximal airways, between the Y-

piece of the ventilator circuit and the endotracheal tube (CO2SMO-Plus - Novametrix

Medical Systems, Wallingford, CT). The CO2SMO-Plus monitor was connected to a

Notebook through a serial RS 232 output (DELL PII, 266mHz), where a special program

recorded proximal pressure and flow tracings at 100 Hz.

A customized flow-generator (Intermed – São Paulo, Brazil) drove the slow lung

inflation. It provided adjustable flow rates from 0.2 to 10 L/min, with 1% precision (for

intra-airways pressure up to 60 cmH2O). The flow-generator was previously calibrated at

2

1L/minute, through the displacement of reference volumes observed at a dry spirometer

(Vitatrace® Pró-doctor; Rio de Janeiro).

For EIT images, we used a 16-electrode Sheffield Applied Potential Tomograph (DAS-

01P Portable Data Acquisition System, IBEES, Sheffield, UK).

During transportation to the CT room, patients were ventilated with a Servo-Siemens

300 C ® ventilator (Siemens-Elema AB, Sweden).

For CT images, we used a GE HighSpeed equipment (GE, Milwalkee, WI), storing data

in an optic disk for subsequent analysis.

PROTOCOL DESIGN

Because it was impossible to obtain simultaneous EIT and CT images due to excessive

electromagnetic interference (signals from EIT got corrupted, with systematic failure in

reciprocity tests), we compared both technologies in sequential order (EIT→CT→EIT),

with repeated measures for EIT (pre and post CT) trying to control for eventual carry

over effects. Each EIT measure was composed by 3 sequential EIT acquisitions during 3

independent slow-inflations, each one preceded by a homogenizing lung maneuver. In

between EIT acquisitions, patients were mechanically ventilated according to baseline

parameters described below, for approximately 5-10 minutes.

For the first and second EIT acquisitions, two different electrode-positioning

arrangements (standard and test) were selected in random order. All electrodes were

discarded and replaced in between. An additional slow-inflation was repeated at the

same positioning of the second acquisition, without replacement o electrodes. In all

trials, electrode arrays were placed exactly at the same cross-sectional plane of the

3

thorax, but with different inter-electrode spacing according to the randomized positioning

(figure 1). Reference electrodes 1 and 9 were kept in the same spots for both

arrangements (anterior and posterior midline, respectively). Electrode 5 was always at

the left side of the thorax. The thoracic perimeter was previously assessed, and the

target skin positions for the electrodes were marked with ink. Those marks were later

used to guide CT scanning.

SPECIAL PROCEDURES AT THE ICU (BEFORE AND AFTER CT SCANNING):

At baseline, all patients were submitted to volume controlled ventilation with VT = 8

mL/kg, respiratory rate of 14 breaths/minute, PEEP = 10 cmH2O, at 100% oxygen. The

constant flow generator was also fed by 100% oxygen. Upper airways and trachea were

previously aspirated. Leak tests at a pause pressure of 25 cmH2O were performed to

adjust internal cuff pressure of endotracheal tube.

Patients were placed in supine position and 16 ECG electrodes (Meditrace 200

electrode, foam adhesive gel 3.6 cm 100/pch 1000/cs Danlee Medical Products, Inc)

were placed on the thorax at the level of the 5th inter-costal space at midclavicular line,

defining a transverse plane crossing the thorax 2 cm below the carina. An additional

ground electrode, for common mode feedback, was placed on the inferior abdomen.

The EIT system was based on a personal computer (Pentium III – 600 MHz), receiving

real-time voltage information from electrodes. Images were reconstructed off-line. At its

maximum speed, a complete voltage set (formed by the average of 10 rotating cycles of

voltage readings) is collected in approximately 1 second. In parallel, a second computer

(notebook, DELL PII, 266mHz) acquired pressure/flow signals at 100 Hz, through serial

4

connection to the CO2SMO-Plus device. Before data acquisition, we ran a customized

software for synchronizing the internal clocks of both computers (error < 0.005 second).

After homogenizing lung history, the slow-inflation was started by manual actuation of a

3-way stopcock, directing the constant flow towards the lung and through the proximal

pressure/flow sensor. After training, we always started the first EIT scanning 1.0 to 1.5

seconds after starting the slow-inflation. Pressure/flow data acquisition always started a

few seconds before, in order to later precise the starting time of inflation (see below).

EIT electrodes were then removed and patients were transported to the CT room.

SPECIAL PROCEDURES AT THE CT ROOM:

Patients were kept in supine position inside the gantry, with a laser beam simulating the

X-ray emission. The table was positioned to guarantee that the CT slice was taken at the

same cross-sectional plane defined by the electrode marks. Before starting data

acquisition, we ran a customized software for synchronizing the internal clocks of the

notebook (collecting proximal pressure/flow) and the CT workstation within 0.005

second. Lung conditions were homogenized as before. After training, we were able to

start the first scanning 1.0 to 1.5 seconds after the very start of the slow-inflation. The

tube current *time product was set at 120 mAs and beam energy at 100 Kva.

IMAGING ANALYSIS – EIT:

Off-line, the two-dimensional EIT image reconstruction algorithm of the Sheffield device

projects the estimates of the regional changes in thoracic impedance over a 32 x 32

matrix. A central circular region with 788 pixels (diameter = 32 pixels) represents the

thorax (figure 1). The image output assigns one value per pixel to represent the relative

5

change (% change) in impedance for that pixel from the reference frame to the current

one. We pre-defined 12 ROIs for this study, represented on figure 2. Each one of the 12

ROIs embraced a portion of the chest wall plus part of the lung. The portions were

selected automatically by a special software splitting the original circle into the following

subsets: total (788 pixels), right and left (394 pixels each), upper and lower (394 pixels

each), layers 1 to 4 (154, 240, 240, 154, pixels respectively), and the four quadrants

(197 pixels each).

A Labview 5.1.1 routine (Labview 5.1.1 – National Instruments software) was

customized to calculate the integral of pixel value over the 12 ROIs.

IMAGING ANALYSIS – CT SCAN:

A quantitative analysis of the tomographic tissue densities was performed according to a

two-step procedure. Firstly, the Osiris Medical Imaging Software version 3.6 (University

Hospital of Geneva - available on the Web: www.expasy.ch/UIN) was used to generate

histograms of density distribution for manually determined regions of interest (figure 2).

The Osiris software produces histograms in which the frequency distribution is displayed

over 2000 arbitrary CT compartments (from –1000 HU to +1000 HU, each step being 1

HU), and the number of voxels included in each compartment is presented in absolute

numbers. Subsequently, all data gathered from the histograms were transferred and

analyzed by a customized software (Labview 5.1.1 – National Instruments software),

programmed to extract quantitative information from regions of interest.

Briefly, the quantitative analysis is based on a quasi-linear relationship between x-ray

attenuation in a given tissue volume – i.e. the voxel, considered as a CT volume unit -

and the physical density of that lung volume (i.e. the mass/volume ratio) (E1-3).

6

The tissue attenuation on x-ray is expressed in CT numbers, or Hounsfield Units (HUs).

This CT number is obtained, in any given voxel, through determining the percentage of

radiation absorbed in a given lung volume. The attenuation scale arbitrarily assigns a

value of 0 H.U for water and a value of –1,000 HU for air (no absorption). This scaling

results in values close to +1,000 HU for bone tissues. By approximating tissue density to

water density (with a specific weight = 1), the relationship between tissue composition

and CT numbers, in any lung region of interest (ROI), may be expressed as:

Volumegas / (Volumegas + Volumetissue) = mean CT numberobserved /(CT numbergas – CT

numberwater)

If the equation above is re-arranged, it is possible to assess, for any given lung voxel,

where the total volume is known, the gas volume inside, the tissue volume inside, and

the gas/tissue ratio (E3). For example, a voxel of –1,000 HU is uniquely composed by

gas, a voxel of 0 H.U. is uniquely composed by water (or “tissue”), and a voxel of –500

H.U. is approximately 50% composed by gas and 50% by water (or tissue). Therefore,

the amount of air inside each voxel may simply be calculated as:

Air-contentVOXEL = (- CT number / 1000) x (Voxel volume)

During slow-inflation - as only one fixed slice through time was assessed - the changes

in air-content refer to the volume of air entering a 1 cm-thick slice. For reference, we

usually observe the entrance of approximately 100 mL of air in a typical slice (at the

thoracic level used in this study) during vital capacity maneuvers in volunteers.

7

The overall accuracy of these calculations was previously validated by a bench study

with a super-syringe with known internal volume (100 mL), and later in a sponge

experiment (intended to represent a complex lung-like structure) during stepwise

changes in water content. For the syringe, the amount of air computed by the addition of

all adjacent slices (at a pitch = 1) was 98 mL (2% error). For the sponge, after the

injection of 500 mL water in its interior, a 476 mL reduction was observed in the

calculated air-content (4.8% error).

SYNCHRONIZING CT AND EIT IMAGES

As described above, the notebook acquiring proximal pressure/flow signals was

synchronized with the EIT personal computer, and later, with the CT workstation (error <

0.005 second). Retrospectively, we looked at airway flow tracings (100 Hz), identifying

the start of slow-inflation with sharp flow spikes (maximum error of ± 0.02 second). Off-

line, we referenced the very start of EIT or CT scans relative to this time origin. Since we

had a higher image rate in the EIT acquisition, we linearly interpolated EIT image data

(for all ROIs) to the same points in time where we had CT scans, getting 30-45

“synchronized” images per inflation.

REGIONAL VENTILATION ACCORDING TO CT AND EIT IMAGES

Since the data collected during the first synchronized EIT scan was considered as the

voltage reference set, all pixels in this initial image had zero value, and subsequent

images represented impedance variations relatively to this reference. Therefore, after

obtaining the last synchronized frame during slow inflation, the regional distribution of

8

tidal volume was simply defined as the ratio between the integral of pixel values over a

certain ROI, divided by the integral of pixel values over the whole slice.

For CT images, regional distribution of ventilation for a particular ROI was calculated

through the comparison of the last (FINAL) to the first synchronized scan (INITIAL), using

the formula below:

( Air-content ROI FINAL - Air-content ROI INITIAL )

( Air-content ALL PIXELS FINAL - Air-content ALL PIXELS INITIAL )

STATISTICAL ANALYSIS

Before assessing the reproducibility of EIT measurement, and since the slow-inflations

were performed at different time points (up to 2 hours difference), the stability of lung

mechanics was tested through the cross-correlation of the 7 pressure-time curves

obtained for each patient (each one with 30-60 data points - figure E1. The presence of

at least one coefficient of correlation below 0.9 was considered as an indication of large

changes in lung conditions, and in this case the whole study case was removed from the

analysis.

EIT ERROR AND REPRODUCIBILITY

For each pair of measurements, at each electrode positioning (standard and test) and

for each lung region (figure 2), the within-subject standard deviation between repeated

measures (Sw) of regional ventilation was calculated according to the method described

9

by Bland and Altman (E4). The same calculations were performed for the special cases

without electrode replacement.

In order to obtain an overall Sw (which represents a global error detected during

repeated measures, for all regions and for both electrode-positioning arrangements), we

considered all measurements after including the regions, patient-region interactions, and

electrode positioning as factors in the repeated measurements ANOVA model. The

square root of the residual variance was considered as SW.

In order to establish an a priori and clinically meaningful value for the reproducibility

required in our experiments, we reasoned the following:

When splitting the lung according to its regions (see Figure 2), we frequently observed a

silent half-lung, or a silent quadrant/layer, suggesting a complete block of regional

ventilation (regional ventilation = 0%) directed to one half or one fourth of the lung.

Considering that the expected regional ventilation for those regions should be 50% and

25%, respectively (supposing homogeneous lung ventilation), and also considering that

the chances of such an extreme result being observed just by chance had to be lower

than 5%, a maximum Sw was calculated to be accepted in repeated measurements (i.e.

the reproducibility limits) (E4):

For regions representing half lung section (right, left, upper and lower ROIs):

Sw = 50.0% / (2–2 x 1.96) = 18.0% (E4)

For regions representing quadrants and layers (layers 1-4, URQ, ULQ, LRQ, and LLQ):

Sw = 25.0% / (2–2 x 1.96) = 9.0%.

10

The inter-observer agreement was not assessed in this study, since most of the

measurements were performed by computer calculation (quantitative image analysis),

with very few subjective decisions.

In order to check if some bias was introduced by the interposed CT scans between the

repeated EIT measurements, we checked if the inclusion of an additional factor (the

pre/post within factor) in the ANOVA model significantly decreased the residual

variance. This procedure attested a non-significant bias (P > 0.20).

EIT AGREEMENT AND CORRELATION TO CT

We used the same rationale to define, a priori, the maximum limits of agreement

between the two methods: in order to reliably detect a complete blockage of regional

ventilation, the sampled bias ± 2 SD should not cross the ±25% boundaries for an area

corresponding to one fourth of the lung; or ±50% for any half lung area.

During multiple linear regression (integral of pixel values plotted against CT air-content,

CT gas/tissue ratio or CT mean-densities), adjustments for inter-subjects differences

were obtained through the inclusion of dummy variables on the model according to the

procedure described by Bland and Altman (E5). Therefore, the reported R2 values

correspond to within-subject determination coefficients.

The SPSS software, version 9.0, was used for all the tests.

11

REFERENCES for the WEB supplement:

E1. Drummond GB. Computed tomography and pulmonary measurements. Br J

Anaesth 1998; 80:665-71.

E2. Hounsfield GN. Computerized transverse axial scanning (tomography). 1.

Description of system. Br J Radiol 1973; 46:1016-22.

E3. Gattinoni L, Caironi P, Pelosi P, Goodman LR. What has computed tomography

taught us about the Acute Respiratory Distress Syndrome? Am J Respir Crit Care

Med 2001; 164:1701-11.

E4. Bland JM, Altman DG. Measurement error. BMJ 1996; 312:1654.

E5. Bland JM, Altman DG. Calculating correlation coefficients with repeated

observations: Part 1 - Correlation within subjects. BMJ 1995; 310:446.

12

Figure E1 (Web repository): Consecutive slow-inflation recordings (converted into

pressure-volume curves) obtained in two patients. Each curve lasted ± 2 minutes. Three

recordings (black filled symbols) were obtained before the recording at CT room (thick

gray line) and 3 recordings afterwards (open symbols). Patient #7 presented

reproducible patterns of lung inflation (cross correlation coefficients > 0.99) while patient

#2 presented very dissimilar behaviors across inflations (cross correlation coefficients

< 0.9). He was, therefore, subsequently excluded. Although discarded from our analysis,

the dynamics of lung inflation observed in this case was very illustrative of the spatial

resolution of EIT (see the animations #1 and #2).

Figure E2 (Web repository): Scattered plots of patient #6, illustrating the linear

regression analysis for impedance variations during slow-inflation as dependent

variable, projected over the air-content changes within CT slice. Electrode positioning

was standard. Each colored plot represents the behavior of a different ROI during both

trials (pre and post the CT scan). The superposed plots and slopes indicate that there

was very little image distortion. R2 represents the overall coefficient of determination.

Animation #1 and Animation #2: Simultaneous pressure-volume recordings vs. EIT

images (sequence #1), and synchronized CT vs. EIT images (sequence #2). The image

sequence suggests that the right lung was blocked during a significant part of lung

inflation, probably due to persistent secretions in the airways, despite our previous

aspiration attempt. This blockage was overcome at the end of inspiration after varying

periods of time for each slow-inflation.

13

Figure E1:

PATIENT #7

AIRWAY PRESSURE (cmH2O)

0 10 20 30 40 50

VO

LUM

E

abo

ve

FR

C (

mL)

0

500

1000

1500

2000

2500

3000

PATIENT #4

AIRWAY PRESSURE (cmH2O)

0 10 20 30 40 50

VO

LUM

E

abo

ve

FR

C (

mL)

0

500

1000

1500

2000

2500

3000

PATIENT #2PATIENT #7

14

Figure E2:

-10 0 10 20 30 40 50 60 70-10000

0

10000

20000

30000

40000

50000

60000

70000

60

40

20

0

∆Z

(pi

xel

inte

gral

)

Changes in Air content (mL)

R2 = 0.96