Borges2011 JAP LungPerfusionByEIT

Regional lung perfusion estimated by electrical impedance tomography

in a piglet model of lung collapse

João Batista Borges,1,8 Fernando Suarez-Sipmann,1,2 Stephan H. Bohm,3 Gerardo Tusman,4

Alexandre Melo,5 Enn Maripuu,6 Mattias Sandström,6 Marcelo Park,7 Eduardo L. V. Costa,8,9

Göran Hedenstierna,10 and Marcelo Amato8

1Department of Surgical Sciences, Section of Anaesthesiology and Critical Care, Uppsala University, Uppsala, Sweden;2Instituto de Investigaciones Sanitarias ISS-FJD, Fundación Jiménez Díaz, Madrid, Spain, CIBERES; 3Swisscom AG,

Landquart, Switzerland; 4Department of Anesthesiology, Hospital Privado de Comunidad, Mar del Plata, Argentina;5Research and Development Department, Timpel SA, São Paulo, Brazil; 6Department of Nuclear Medicine, Uppsala

University Hospital, Uppsala, Sweden; 7Hospital das Clínicas, and 8Cardio-Pulmonary Department, Pulmonary Divison,

Hospital das Clínicas, University of São Paulo; 9Research and Education Institute, Hospital Sírio Libanês, São Paulo, Brazil;

and 10Department of Medical Sciences, Clinical Physiology, Uppsala University, Uppsala, Sweden

Submitted 13 September 2010; accepted in final form 26 September 2011

Borges JB, Suarez-Sipmann F, Bohm SH, Tusman G, MeloA, Maripuu E, Sandström M, Park M, Costa EL, HedenstiernaG, Amato M. Regional lung perfusion estimated by electricalimpedance tomography in a piglet model of lung collapse. J Appl

Physiol 112: 225–236, 2012. First published September 29, 2011;doi:10.1152/japplphysiol.01090.2010.—The assessment of the re-gional match between alveolar ventilation and perfusion in criticallyill patients requires simultaneous measurements of both parameters.Ideally, assessment of lung perfusion should be performed in real-timewith an imaging technology that provides, through fast acquisition ofsequential images, information about the regional dynamics or re-gional kinetics of an appropriate tracer. We present a novel electricalimpedance tomography (EIT)-based method that quantitatively esti-mates regional lung perfusion based on first-pass kinetics of a bolus ofhypertonic saline contrast. Pulmonary blood flow was measured in sixpiglets during control and unilateral or bilateral lung collapse condi-tions. The first-pass kinetics method showed good agreement with theestimates obtained by single-photon-emission computerized tomogra-phy (SPECT). The mean difference (SPECT minus EIT) betweenfractional blood flow to lung areas suffering atelectasis was �0.6%,with a SD of 2.9%. This method outperformed the estimates of lungperfusion based on impedance pulsatility. In conclusion, we describea novel method based on EIT for estimating regional lung perfusion atthe bedside. In both healthy and injured lung conditions, the distri-bution of pulmonary blood flow as assessed by EIT agreed well withthe one obtained by SPECT. The method proposed in this study hasthe potential to contribute to a better understanding of the behavior ofregional perfusion under different lung and therapeutic conditions.

first-pass kinetics; pulmonary blood flow distribution

EFFICIENT GAS EXCHANGE, the primary function of the lung, is theresult of an intimate matching of regional ventilation andperfusion across alveoli. Critically ill patients under mechani-cal ventilation often present gross disturbances of pulmonarygas exchange. To determine the regional matching of alveolarventilation and perfusion during pathologic conditions, amethod capable of measuring ventilation and perfusion, bothglobally and regionally, is needed. Such measurements gobeyond the structural and anatomical information provided byclassical imaging technologies and require functional informa-

tion like regional blood flow or the regional kinetics of aconvenient tracer at the bedside. In recent publications (15,16), the determination of regional perfusion through atelectaticlung units has been repeatedly pointed out as an importantmissing link in the assessment of the consequences of mechan-ical ventilation.

New developments in functional imaging methods such asmagnetic resonance imaging (MRI; Ref. 50), X-ray computer-ized tomography (CT; Refs. 5, 6, 23, 32), single-photon-emission computerized tomography (SPECT, Ref. 29), andpositron emission tomography (PET) have provided, to variousdegrees, interesting insights into the heterogeneity of lungaeration and the distribution of ventilation and perfusion. Fromthese methods, PET has established a quantitative relationshipbetween regional ventilation and perfusion and gas exchange(52). However, the above methods lack the ability to track thefrequent changes in the lung structure and function that occurin the critically ill because they do not allow for continuousmonitoring at the bedside. Novel methods capable of repeateddata acquisition and interpretation in real time are desirable.

Electrical impedance tomography (EIT) has emerged as anew functional imaging method potentially meeting manyclinical needs (1–3, 8, 25, 51). Subjecting the chest to minuteelectrical currents, this radiation-free, noninvasive techniquemeasures the electric potentials at the chest wall surface toproduce two-dimensional (2-D) images that reflect the imped-ance distribution within the thorax. Cyclic variations in pul-monary air and blood content are the major determinants forthe changes in thoracic impedance, the former usually of muchlarger magnitude. Because cyclic changes in local impedancemainly correspond to changes in lung aeration, recent studies(22, 27, 40, 51, 56) have shown that EIT can reliably assessimbalances in the distribution of regional ventilation in criti-cally ill patients. Besides other features like portability and thepossibility of around-the-clock monitoring, the high temporalresolution is an important aspect of bedside imaging thatallows for the study not only of ventilation but also of fasterphysiological phenomena, such as the pulsatility of the lungduring the cardiac cycle (7, 9, 17–20, 30, 37, 45, 53, 55). Themeasurement of lung pulsatility has been the principal methodfor assessing perfusion by EIT, as it correlates well with strokevolume measured by thermodilution (53). The pulsatility

Address for reprint requests and other correspondence: J. B. Borges: Dept.of Surgical Sciences, Section of Anaesthesiology & Critical Care, UppsalaUniv., 751 85, Uppsala, Sweden (e-mail: [email protected]).

J Appl Physiol 112: 225–236, 2012.First published September 29, 2011; doi:10.1152/japplphysiol.01090.2010. Innovative Methodology

8750-7587/12 Copyright © 2012 the American Physiological Societyhttp://www.jap.org 225

method measures the amplitude of the cyclic perturbations inlocal lung impedance caused by the passage of the stroke-volume through the lung (7, 17–19, 30, 37, 45, 53, 55). Lungpulsatility, however, is also influenced to a significant degreeby the distensibility of the pulmonary vessels as well as the sizeand patency of the pulmonary microvascular bed (42). There-fore, pulsatility-based methods might be misleading as a mea-sure of pulmonary perfusion should collapse of small pulmo-nary vessels occur, as well as during any important change inparenchyma architecture.

The concept of using hypertonic solutions as an EIT contrastagent to measure perfusion was proposed a long time ago (9).More recently, lung perfusion estimated by such a method hasbeen compared with regional static CT images (20). We hereinpresent a modified version of the EIT-based method to estimateregional lung perfusion based on the first-pass kinetics of abolus of hypertonic contrast. While the theoretical frameworkof such first-pass kinetics has been validated for CT and MRItechnology (26, 28, 34, 35, 41), it requires adaptations to meetthe particularities of EIT like the nature and timing of theperturbations caused by the hypertonic bolus as well as theabsence of precise information about the slice thickness ofthe EIT image, which encompasses large portions of the lungand heart. Our hypothesis was that the novel indicator dilutionmethod should outperform lung pulsatility as a surrogate forregional lung perfusion.

METHODS

EIT

EIT data were acquired using the enlight platform for impedancetomography, developed by the Experimental Pulmonology Laboratoryand Polytechnic Institute of the University of São Paulo, in a partner-ship with Dixtal Biomédica (São Paulo, Brazil; Refs. 12–14). Theprototype is capable of producing 50 real-time images per second.After the thoracic perimeter was measured, 32 adhesive electrodeswere placed equidistantly around the circumference of the thorax justbelow the level of the axilla. Small electrical currents (5–12 mA; 125kHz) were injected in a rotating sequence through pairs of electrodes,with one noninjecting electrode interposed between two injectingelectrodes. During an injection pattern, the noninjecting electrodeswere used to measure 29 differential voltages between interleavedelectrode pairs in parallel. One complete acquisition cycle of 32current patterns produced 928 voltage measurements comprising one“raw voltage frame” used as an input to construct a relative EITimage. These images were generated by a reconstruction algorithm fora cross section of the thorax, which is based on a sensitivity matrixderived from a three-dimensional (3-D) finite element model with�6,000 elements, 6-cm thick, and with the approximate dimensions ofa piglet’s thorax (21 � 22 cm). The relative impedance changesestimated for the midlayer of the finite element mesh (midway alongthe craniocaudal axis) were plotted in a matrix containing 860 validpixels from a total of 1,024 (32 � 32). Thus the size of each pixelcorresponded to �0.7 � 0.7 cm in the axial plane. A “primary relativeimage” was created by comparing the most recent raw voltage framewith a reference frame recorded at the beginning of each dataacquisition or at any other selected moment. Output pixel valuesrepresented percent changes in local impedance compared with suchreference. For the present analysis, the reference was taken either at thebeginning of each ECG-triggered data acquisition or at the beginning ofinspiration for mechanical ventilation-gated images. In the case of ournovel perfusion analysis, however, the reference frame was always takenat apnea during a preceding expiratory pause. For further details on thetechnical aspects of EIT technology, see Refs. 4, 12–14.

The average or simple sum of all pixel values within each outputimage was plotted against time, producing a global EIT signal. Suchsignal is linearly related to changes in parenchymal air content (39,51). This concept was applied regionally within regions of interest(ROIs) producing regional EIT signals.

Concept of Estimating Regional Lung Perfusion by EIT

Due to its high conductivity, 20 % NaCl acts as an EIT contrastagent (9, 20, 49), which after injection into the right atrium, duringapnea, passes through the pulmonary circulation thereby producinga dilution curve that follows typical first-pass kinetics. The result-ing regional time-impedance curves are then analyzed to quanti-tatively assess regional perfusion. After inverting these curves(multiplying by �1), an appropriate time window lasting from themoment of bolus infusion until just before some impedance re-bound was detected (usually at the middecay of the curve) wasselected to avoid recirculation artifacts. This window was re-stricted further by using data points from the beginning of theperturbation until its peak (48), thereby making recirculationartifacts even more unlikely. The resulting curve was then fitted ona pixel-by-pixel basis by a corresponding gamma-variate function.(33, 48).

Before using such routine, however, the following additional signalprocessing at the pixel level had to be performed.

Subtraction of the Right Cardiac Phase

During contrast bolus injection, some EIT pixels, especiallythose in the right heart region, typically show a two-valley behav-ior in which the earlier local minimum is related to the passage ofthe bolus through the right heart and the later one to its passagethrough the lung parenchyma, with both phenomena representedwithin the same pixel (Fig. 1). Such behavior, known as partialvolume effect, is well described for CT technology. Given therelatively large pixels and comparatively poor spatial resolution ofEIT, especially along the z-axis (craniocaudal), this effect requiresproper treatment.

We assumed that each pixel in the image was composed of a doublecompartment, an earlier and faster one related to the right heart and alater and slower one to the lung. Each compartment exhibited its ownfirst-pass kinetics and a respective gamma function. The fitting pro-cess was then performed in such a way that the sum of the twoindependent gamma functions had to best fit the raw curve for eachpixel, allowing a certain temporal overlap between both curves. Wethen discarded the right heart component and continued the analysiswith the net lung-perfusion signal corresponding to the expectedbehavior of the slow compartment and its respective gamma function(Fig. 1).

Left Cardiac Phase and Recirculation

Perfusion of the bronchial arteries and systemic recirculationcause a typical and late (8 to 12 s after bolus injection) rebound ofimpedance perturbation of very low magnitude. For most measure-ments it was easy to select a time window long enough to obtain anadequate curve fitting but short enough to avoid contamination bythe above phenomena. In addition, a transient and localized im-pedance rebound was sometimes observed due to accumulation ofcontrast within the left cardiac chambers. This effect happenedbefore recirculation and just a few frames (0.5 to 1.0 s) after themain lung-perfusion phase. Although causing impedance perturba-tions of much lower magnitude than the right heart, we reasonedthat such phenomenon could also cause partial volume artifacts inthe few pixels corresponding to the left heart region leading to anoverestimation of the net lung-perfusion components. Therefore,especially because of those pixels with a very slow perfusion andthose showing an early left heart perturbation, we used a method to

Innovative Methodology

226 REGIONAL LUNG PERFUSION BY EIT

J Appl Physiol • doi:10.1152/japplphysiol.01090.2010 • www.jap.org

estimate regional blood flow that relies solely on the part of thedilution curve before its peak. For details see below, and APPENDIX

A AND B.

Calculating Regional Flow

After extracting the net contrast curve representing lung perfu-sion within each pixel the following parameters were derived fromthe fitted gamma function: peak-value, time-to-peak, area above

the curve (i.e., integral of the corresponding gamma function curve

over time, also called gamma area; Ref. 48), and maximum slope

within the time interval from time-to-appear to the time-to-peak.

The most widely described method for calculating regional

blood flow in CT analysis, MRI, or focal angiograms is based on

the central volume theorem (57), according to which the regional

perfusion of a parenchymal organ is expressed by the following

formula:

Fig. 1. Typical behaviors of the signal in a lungand a heart/lung pixel observed after injectionof a bolus of hypertonic saline. A: typical lungpixel shows 1 visible valley only, representingthe passage of the contrast through the lung.Fitting of the early cardiac component wasnegligible and cannot be seen. Note that thenadir of the curve occurs �3–4 s after bolusinjection. B: typical pixel behavior at thoracicregions close to the heart exhibiting a 2-valleyshape. Note that the first and steeper valleyoccurs earlier, immediately after the bolus ofcontrast (arrow) corresponding to the passageof the bolus through the cardiac chambers(right atrium and ventricle). A second valleycan be observed later along this same pixeltracing (4–5 s after bolus injection) corre-sponding to the later passage of the contrastthrough lung regions also represented withinthis same pixel. Such lung regions are proba-bly located at a different position along thecraniocaudal z-axis , but at a similar location inthe x-y-plane. Although fitting and integralcalculations were performed after inverting thecurves above, we presented the negativecurves directly derived from electrical imped-ance tomography (EIT) images, illustrating thenegative perturbation in impedance caused bythe hypertonic bolus. For both pixels, the esti-mates for net lung perfusion (i.e., the maxi-mum slope of the initial moment of lung per-turbation) were calculated from the fitted graycurve.


227REGIONAL LUNG PERFUSION BY EIT


relative blood flow�pixel�

� relative blood volume�pixel� ⁄ mean transit time�pixel�;

where the relative blood volume(pixel) is directly estimated from thefitted gamma area for each pixel.

In our calculations, however, the accurate estimation for the mean

transit time would require not only the removal of the signal-compo-nents related to recirculation and left cardiac phase but also thesimultaneous measurement of the time-impedance curves of the feed-ing pulmonary artery (to be used in a deconvolution process, to correctfor the fact that the bolus is not fast enough).

Due to such practical difficulties, we opted for an adaptation of anolder and less frequently used method for measuring regional perfu-sion called the maximal slope method (28, 34). The method is basedon the Fick principle of the conservation of mass to a given region ofinterest (28, 36). Assuming there is no venous drainage of contrastbefore the peak of the pulmonary artery input function, the accumu-lated mass of the contrast within a pixel can be calculated as theproduct of regional blood flow and the time integral of the pulmonaryartery contrast concentration:

m�t�pixel

� blood flowpixel · �0

tpulmonary artery concentration�� · d�

(1)

Differentiating Eq. 1 leads to an expression of the slope or dm(t)/dt:

dm�t�dt

� blood flowpixel · pulmonary artery concentration�t� (2)

Consequently, the slope will be maximal when the pulmonaryartery concentration is maximal. Regional blood flow can then becalculated as:

blood flowpixel �

�dm�t�dt

�max

pulmonary artery concentration�t�max

(3)

The maximal slope method is thus based on the concept that thespeed of accumulation of contrast within a pixel, represented by theinitial slope of the gamma curve, reflects perfusion as the flow ofblood into that pixel or the wash-in function to that compartment. Anassumption of this method is that there is no venous outflow ofcontrast during the period used for estimating the slope, thus requiringa high rate of contrast injection in relation to the transit times foundin the lung. Another assumption is that the blood flow is constant,instead of pulsatile. For EIT, this method is especially convenient forseveral reasons: 1) the high temporal resolution of EIT delivers up to50 points per second with a signal-to-noise ratio higher than in CT orMRI imaging at their common resolutions, making the curve fittingsand slope determinations more reliable; 2) the bolus of hypertoniccontrast is relatively small (5 ml) and can be infused in �1 s; 3) weobserved that pixel perturbations caused by the given amount ofhypertonic contrast were well above background noise (the signal tonoise ratio per pixel is higher than in CT studies of lung perfusion, forinstance); 4) there was no need to measure the input function as wewere not interested in absolute values of perfusion (it was enough tocompare pixels within the image); and 5) we could focus on the veryinitial part of the gamma curve for each pixel thereby avoidingrecirculation and left-heart artifacts. These artifacts can markedlyaffect the gamma-area calculations, but their effect on the initial slopeof such gamma curve is negligible. For each pixel, we searched for themaximum slope of the gamma function from the time-to-appear to thetime-to-peak of the fastest pixel (i.e., after removing the right heartperturbations, we identified the first pixel in the image to reach itspeak and assumed that it represented the compartment with the fastesttransit time within the lung). The relative proportion of this fastest

slope, against the slopes for all other pixels, represented the relativeperfusion to each lung region.

To evaluate the predicted performance of such adapted algorithm,we performed heuristic mathematical simulations (see APPENDIX A).

EIT Perfusion Analysis Based on Local Lung Pulsatility

At each protocol condition, an EIT perfusion analysis based onlocal lung pulsatility (�Z) (7, 17–19, 30, 37, 45, 53, 55) was per-formed and compared with the hypertonic saline bolus first passkinetics method.

Briefly, this method is based on the fact that the ejection of thestroke volume from the right heart during systole causes an increasein intravascular pulmonary blood volume and pressures, which leadsto the distension of the elastic pulmonary vascular bed, such bulgingof blood within septa purportedly causes a decrease in local electricalimpedance. It is assumed that such cyclic changes in regional imped-ance reflect the proportion of the stroke volume directed to suchregions. Thus, by measuring the amplitude of the cyclic changes inlung impedance for each pixel, one can estimate the local distributionof right ventricular stroke volume (see supplemental video 2) or thelocal distribution of pulmonary blood flow. The pulsatility measure-ment did not require a contrast agent and was made using anECG-gated acquisition comprising 50 consecutive cardiac cyclesduring a 20-s apnea. Additionally, the resulting gated signal waslow-pass filtered at a frequency of 15 Hz to avoid overestimation inthe �Z due to noise (24).

Experimental Protocol

The study was approved by the Animal Ethics Committee ofUppsala University (Uppsala, Sweden). Six piglets (2–3 mo oldweighing 28.4 � 2.6 kg) of mixed Hampshire, Yorkshire, and Swed-ish country breeds obtained from a local breeder were used andstudied during various experimental conditions. All animals under-went the same routine instrumentation, intravenous anesthesia using acombination of fentanyl, ketamine and midazolam, and monitoring aspreviously described (46). Animals were tracheotomized and mechan-ically ventilated using a cuffed 7-mm inner diameter endotrachealtube (Mallinckrodt, Athlone, Ireland). Baseline ventilation was deliv-ered in a volume controlled mode using a Servo-i ventilator (MaquetCritical Care, Solna, Sweden) with the following settings: tidal vol-ume of 6 ml/kg, respiratory rate of 30 breaths/min, positive end-expiratory pressure of 10 cmH2O, inspiratory-to-expiratory ratio of1:2, and fraction of inspired oxygen of 1. All protocol steps wereperformed with the animals lying in the supine position. After prep-aration, two major experimental conditions were studied.

Unilateral lung collapse in healthy animals. With the use of thehealthy animals, the following conditions were sequentially applied:1) bilateral lung ventilation, 2) unilateral lung ventilation due tocomplete contralateral lung collapse after selective intubation of themain bronchus of the ventilated lung, and 3) unilateral lung ventilationwith sodium nitroprusside infusion to attenuate hypoxic pulmonaryvasoconstriction.

Bilateral-dependent lung collapse in injured animals. Lung injurywas induced by repeated lung lavages as previously described (31).After lung injury was established, three situations were monitoredsequentially during bilateral lung ventilation: 1) ventilation usingbaseline settings, 2) after sodium nitroprusside infusion, and 3) afteran effective lung reexpansion maintaining an “open lung condition”by a positive end-expiratory pressure of �16 cmH2O (46).

Each experimental condition lasted 45 min, and measurementswere made at the end of each one.

EIT measurements of perfusion based on first-pass kinetics wereperformed after switching the ventilator to continuous positive airwaypressure (CPAP) maintaining the same level of expiratory pressurebut eliminating tidal ventilation. A 20% NaCl bolus was injected




within �1 s through a central venous line into the right atrium 5 s afterstarting CPAP.

SPECT Imaging

Pulmonary blood flow distribution was assessed by intravenousinjection of 99mtechnetium-labeled macro-aggregated albumin(99mTc-MAA, Pulmocis; CISbiointernational, Gif sur Yvette, France).At the beginning of each phase, a reference SPECT was performedbefore each new injection of 99mTc-MAA to subtract the remainingtechnetium activity from the previous scan. A CT scan (covering thesame volume as the SPECT) was taken immediately after eachSPECT and used for attenuation correction. Images were acquiredusing a SPECT/CT dual-head gamma camera (Millennium; GeneralElectric Systems, Milwaukee, WI) with an all-purpose medium-energy collimator. SPECT acquisition was made in 60 projections (30per head) and stored in a 128 � 128 matrix, resulting in a pixel sizeof 4.42 mm. The overall scan time for SPECT and CT was �45 min.Data were reconstructed first on an eNTEGRA workstation and lateron a Xeleris workstation (Millennium; General Electric Systems). Thereconstruction was performed with an iterative model (OSEM, 4iterations and 8 subsets) and a Hann filter (cut-off 0.85) for thepostreconstruction filtering on both workstations. The 128 transverseslices, each of thickness 4.42 mm, were corrected for radiationspillover and for baseline subtraction using a HERMES workstation(Hermes Medical Solution, Stockholm, Sweden). For each recon-structed slice, the contents were analyzed by custom-made software.Before calculating activity distribution, a background subtraction of10% of the global maximum was performed.

As opposed to EIT, SPECT measurements had to be performedalong the entire experimental condition for 45 min to improve signalto noise ratio. The initial bolus of labeled albumin, however, wasinjected after switching the ventilator to CPAP maintaining the samelevel of expiratory pressure but eliminating tidal ventilation, similarlyto the injection of hypertonic saline during EIT acquisitions.

ROIs

For EIT and SPECT image comparisons at the different experi-mental conditions we analyzed pulmonary blood perfusion in differentROIs.

During unilateral lung collapse in healthy animals, the lung wasdivided by a vertical line into two ROIs: the left and the right lung(Fig. 2).

During bilateral-dependent lung collapse in injured lungs, the lungwas divided by a horizontal line into two ROIs delimiting a superior(nondependent and aerated) from an inferior (dependent and col-lapsed) one. Based on the corresponding CT images, we identified thehorizontal plane that best separated a nondependent region encom-passing an open lung region from a fixed dependent one encompass-ing most of atelectatic tissue (Fig. 3).

Finally, for both healthy and injured lungs, we analyzed foursymmetrically delimited quadrants of the lung.

Images Analysis

The perfusion of each ROI was calculated as its respective per-centage of total blood flow.

For SPECT (using 99mTc-MAA � SPECT):

ROIa% � ROIa counts/total counts,

where ROIa % is percentage of total pulmonary blood flow directed toROIa, ROIa counts is sum of absolute values of counts of all pixelsencompassed in ROIa, and total counts is sum of absolute values ofcounts of all pixels encompassed in all ROIs.

To obtain the corrected SPECT count values used in the calcula-tions above, a reference SPECT was taken before each new injection

of 99mTc-MAA to subtract the remaining technetium activity from theprevious scan.

For EIT-gamma variate modeling:

ROIa% � ROIa�maximal slope�/total�maximal slope�

where ROIa % is percentage of total pulmonary blood flow directed toROIa estimated by EIT-gamma variate modeling, maximal slope ispixel by pixel maximal slope of the gamma function of the net lungperfusion component from the time-to-appear to the time-to-peak ofthe fastest pixel, ROIa (maximal slope) is the sum of (maximal slope)values of all pixels encompassed in ROIa, and total (maximal slope)is the sum of (maximal slope) values of all pixels encompassed in allROIs.

For EIT-pulsatility modeling:

ROIa% � ROIa��Zpulsatile�/total��Zpulsatile�

where ROIa % is the percentage of total pulmonary blood flowdirected to ROIa estimated by EIT pulsatility; �Zpulsatile is the ampli-tude (difference between maximum and minimum values) of �Zcardiac oscillations for each pixel, observed in ECG-gated measure-ments; ROIa (�Zpulsatile) is the sum of �Zpulsatile values of all pixels

Fig. 2. Regions of interest in healthy lungs. Regions of interest in healthy lungsfor the perfusion analysis. Three situations were studied sequentially: bilaterallung ventilation (A), unilateral lung collapse (B; left lung atelectasis), andunilateral lung collapse (C) with sodium nitroprusside infusion. After the lungborders were determined by the computerized topography (CT) scan, the leftand the right lungs were separated by a vertical line, drawn from the sternumto the spinal column. Corresponding vertical line was also drawn in the EITimages after identifying the corresponding positions of electrodes on the CTand within the EIT finite-element mesh. Note the decreased intensity of thesignal in the left lung after atelectasis (B), representing a decreased perfusion,which was partially recovered during the infusion of nitroprusside. SPECT,single-photon-emission computerized tomography.




encompassed in ROIa; and total (�Zpulsatile) is the sum of �Zpulsatile

values of all pixels encompassed in all ROIs.For EIT pulsatility analysis, all pixels corresponding to the heart

region were previously removed. The selection of such pixels wasbased on a phase analysis: pixels presenting a predominant positivevariation in �Z between two consecutive QRS complexes wereconsidered “heart” pixels (21, 47) (Fig. 4).

RESULTS

A total of six animals are included in this analysis: three inthe unilateral lung collapse protocol (complete left lung col-lapse in two and right lung collapse in one) and three in thebilateral dependent lung collapse protocol.

EIT-Perfusion Analysis Based On The First-Pass Kinetics

The distributions of pulmonary blood flow assessed bySPECT and EIT (first-pass kinetics) showed good agreement,as exemplified by the qualitative inspection of the images(Figs. 2 and 3) and also by the quantitative analysis using twoROIs (Fig. 5), one of them always representing the collapsed

part of the lung. This was true for both healthy and injuredpiglets. After induction of atelectasis, both techniques consis-tently demonstrated an initial redistribution of blood flow tonon collapsed lung areas, especially in the healthy lung prep-arations (see supplemental videos 1 and 3). For instance, in oneof the healthy animals, the fractional pulmonary blood flowdirected to the left lung decreased from �50% down to �20%after complete atelectasis of this lung (Figs. 2 and 5). Thisredirection of blood flow was attributed to hypoxic pulmonaryvasoconstriction, which was later attenuated by nitroprussideinfusion. We also observed an immediate increment in regionalflow to dependent lung regions after recruitment of the col-lapsed areas in injured lungs. This increment exceeded theincrement caused by nitroprusside in all animals, as demon-strated by both EIT and SPECT analysis (see Fig. 5).

Altogether, the mean difference (bias) between the estimatesfor the blood flow towards the atelectatic ROI in relation to theblood flow to the whole lung (henceforth called fractionalblood flow), as measured by SPECT vs. EIT, was �0.6% witha SD of 2.9% (range: �6.7 to � 3.8%).

Figure 6 shows the Bland-Altman plots of the SPECT vs.EIT measurements of fractional blood flow to four lung quad-rants of the transverse images. The overall agreement was good(limits of agreement � �10.9% to � 10.5%) although asystematic overestimation of perfusion to the top-left quadrantwas observed in EIT measurements (P � 0.05). We did notobserve any dependence between the difference seen in SPECTand EIT-NaCl measurements and their average magnitude(P � 0.14).

EIT-Perfusion Analysis Based On Local LungPulsatility (�Z)

Significant differences between the estimates of pulmonaryblood flow distribution by SPECT and the correspondingestimates based on EIT pulsatility were observed (Figs. 5and 6).

The discrepancy was especially magnified in the presenceof whole lung atelectasis or after the infusion of nitroprus-side. In all animals, SPECT analysis indicated that atelec-tasis caused a systematic decrease in local perfusion (meandecrease � �23% of total pulmonary blood flow; range:�14 to �29%; P � 0.028, Wilcoxon rank test). Thepulsatility analysis, however, showed contradictory results:atelectasis of the injured, dependent lung zones caused somedecrease in �Z pulsatility, but atelectasis of one whole lungcaused a marked increase in local �Z pulsatility in twoanimals. In Figs. 4B and 5, we show one extreme examplein which EIT’s local pulsatility estimated a fractional per-fusion of 74% to the atelectatic left lung, whereas theSPECT indicated just 19%.

The infusion of nitroprusside increased the blood flowthrough atelectatic areas in all animals. The pulsatility analysis,however, systematically suggested a decreased blood flowthrough atelectatic areas after nitroprusside.

Figure 6 shows the Bland-Altman plots of the SPECT vs.EIT pulsatility measurements of fractional blood flow to fourlung quadrants (of axial images). There was a systematicoverestimation of perfusion to the top left quadrant (P �

0.001) and a systematic underestimation of perfusion to thebottom right quadrant (P � 0.01). There was no dependence of

Fig. 3. Regions of interest in injured lungs. Regions of interest in the pigletswith acute lung injury. Three situations are represented sequentially: lunginjury condition (A; dependent lung atelectasis), lung injury in conjunctionwith sodium nitroprusside infusion (B), and open lung condition after lungrecruitment (C). One superior and one inferior region of interest were separatedby a horizontal line based on the corresponding CT images, where thedependent region encompassed nearly all atelectatic tissue in condition (A).Corresponding horizontal line was also draw in the EIT images after objec-tively identifying and matching the positions of electrodes on the CT imagesand within the EIT finite-element mesh. Note the zones of maximum intensityof perfusion, which are above the horizontal line in A. This was evident on bothEIT (first-pass kinetics) and SPECT images. Such zones of maximum intensitywere displaced downward in B and C, where the highest intensity can bedetected in dependent lung zones.




the difference between SPECT and EIT pulsatility in relation totheir average magnitude (P � 0.16).

DISCUSSION

We herein report a novel EIT-based method for estimatingregional lung perfusion at the bedside. The proposed approachis based on first-pass kinetics of an EIT-indicator dilution

curve, using hypertonic saline bolus and overcoming majorlimitations of a pulsatility-based method. This method showeda good agreement with estimates of lung regional perfusionobtained by SPECT. Our findings also support the notion thatEIT pulsatility may be a less suited method for estimatingpulmonary perfusion.

A fundamental reason for the poorer performance of the EITpulsatility, as evidenced in Fig. 4, is probably related to the factthat such method basically measures the pulsatile changes inpulmonary blood volume instead of real forward flow of blood.Changes in vascular tone and synchronous changes in aircontent (in opposite phase) as previously described (44, 54)might potentially affect the relationship between the cyclicpulsatile changes in regional impedance and the proportion ofthe stroke volume directed to those regions. Our results suggestthat EIT-pulsatility data may be strongly influenced by thedownstream pulmonary vascular resistance and by the disten-sibility of the small pulmonary vessels (19, 43). Probably,different lung conditions and airway pressures also modulatedthe pulsatility behavior of small pulmonary vessels. Anotherpotential reason for the poor performance of the pulsatilitymethod would be the inclusion of the heart region in themeasurements, which we tried to avoid by identifying the heartregion and excluding it from the analyses. However, theidentification of pixels corresponding to the heart region mighthave been less than ideal in some cases, which could partiallyexplain the overestimation of perfusion to the top left quadrantby the pulsatility method. The review of individual casessuggested that it is hard to completely exclude residual com-ponents of the left atrium from the images. The phases of theleft atrium and lungs are similar enough to preclude theexclusion of the former based on phase analysis. More sophis-ticated methods would be required.

The biggest difference between local blood flow measuredby SPECT and EIT-pulsatility images was found during uni-lateral collapse. Stronger pulsatile changes in impedance werefound within the collapsed lung, while the amount of bloodflowing to this region was significantly reduced according to

Fig. 4. EIT-pulsatility (ECG gated) images of 1 animal, before removing pixelscorresponding to the heart region, during 3 conditions: bilateral lung ventila-tion (A), unilateral healthy left lung collapse (B), and unilateral lung collapse(C) plus sodium nitroprusside infusion. A gray scale was applied, with thelighter color corresponding to higher pulsatility signal. Time-synchronizedECG plus impedance waveforms of three representative pixels are shown, withequivalent scales for both lungs (but halved for heart region). Orange pixelrepresents the left lung, gray represents heart, and yellow represents the rightlung. After creation of atelectasis, note the global decrease in pulsatility, butespecially in the right (open) lung (B), where the fractional blood flow at thismoment, according to SPECT and EIT-first-pass kinetics, was maximal.Concomitantly, the left atelectatic lung gets relatively “brighter” (because ofthe auto-scaling of the gray images) producing the false impression of anincreased perfusion. As explained in the discussion, this phenomenon may becaused by a diastolic redistribution of blood flow (pendelblut) from the left tothe right pulmonary artery. By addition of nitroprusside infusion (C), a markedattenuation of the difference between the open and collapsed lung becomesevident, approaching the normal lung pattern with bilateral ventilation (A).Diastolic emptying of the pulmonary vessels toward the left atrium is charac-terized by a marked rise in �Z signals (amplitude � difference betweenmaximum and minimum values of EIT signal oscillations) before the next QRScomplex. Note the delay of such process within the right lung in situation (C),which is absent in A (arrows). Note also the complete opposition of phasesbetween the heart and the lungs waveforms.




both SPECT and EIT-first-pass kinetics (Figs. 4B and 5). Ourdata are in agreement with the findings of Newell et al. (38),who reported a retrograde flow from atelectatic lung zones tohealthy ones during the diastolic phase. Within the atelectaticlung, the elastic compartment represented by small pulmonaryvessels seems to accumulate a great part of the stroke volumeat the end of systole, expelling it back to the contralateral lungduring diastole. The much later nadir of EIT signal oscillationsin the healthy lung (suggesting a later surge in local bloodvolume), compared with the atelectatic one (Fig. 4C), sug-gested this redistribution of perfusion (“pendelblut” in analogywith the German term “pendelluft”) between both lungs. Thispendelblut was not present during bilateral ventilation. Wehypothesize that, in the presence of massive atelectasis, the pen-delblut effect could be so marked and prolonged that the normaldiastolic emptying of pulmonary vessels might be attenuated,decreasing the magnitude of �Z pulsatility in the healthy areas. Itworks as if the healthy lung was receiving an almost constant flowrather than a pulsatile, variable flow. Thus despite a higher flowrate (as evidenced by SPECT and EIT-first-pass kinetics), thehealthy lung had lower pulsatility.

Interestingly, nitroprusside infusion always attenuated therelative pulsatility within the collapsed lung (Figs. 4C and 5),

although systematically increasing the amount of blood shunt-ing through these areas. This seemed to be associated with anincreased pulsatility within the remaining open lung areas,rather than a true attenuation within collapsed areas.

It is important to keep in mind that, although the pulsatilitymethod has been appealing because of its ease, it does not relyon solid mathematical foundation to really represent perfusionin a quantitative manner. In contrast, the theories behind thefirst-pass kinetics of a tracer and the maximum-slope approachdescribed in this study are concepts that relate to classicphysiological studies (11, 28, 34, 57) about parenchymal per-fusion, having also a proven mathematical justification (see themathematical model described in the APPENDIX A). The presentdata may unravel a potential limitation of the pulsatilitymethod, especially in conditions where atelectasis may play animportant role, which will have to be confirmed in furtherstudies.

Besides the small number of animals, which prevented usfrom presenting definitive statistics, but rather a descriptiveproof of concepts, some other limitations of this study deservementioning. First, due to technical reasons, SPECT and EITimage acquisitions could not be done simultaneously but se-quentially. To circumvent this limitation, we always waited for

Fig. 5. Distribution of pulmonary blood flow in healthy lungs and in lungs with acute lung injury. Percentage of total pulmonary blood flow flowing throughthe left atelectatic lung as measured by EIT and SPECT. All animals are shown. Each panel refers to one studied animal. EIT perfusion according to the first-passkinetics was based on the maximum slope of the first-pass of contrast through the lung. EIT pulsatility represents the local amplitude of impedance swings duringECG-gated acquisition (see METHODS for details). Three sequential conditions in three noninjured lung preparations are shown: bilateral lung ventilation, unilaterallung collapse, and unilateral lung collapse with sodium nitroprusside infusion. Three sequential conditions in three injured lung preparations are shown: lunginjury (bilateral, dependent lung atelectasis), lung injury plus sodium nitroprusside infusion, and open lung condition after lung recruitment (total reversal ofatelectasis). Note the good match between the first-pass kinetics method and SPECT. Note also the opposite, artifactual changes in EIT pulsatility observed duringatelectasis and after the infusion of nitroprusside.




steady-state conditions, including the effects of nitroprusside,whereby a continuous monitoring of arterial blood gases wasessential. A second limitation comes from the fact that SPECTis essentially a 3-D imaging procedure, while EIT analysis issomething between 2-D and 3-D (although the finite mesh is3-D, the electrodes were placed within a single plane). Al-though reasonably large, the effective thickness of the EITcross-sectional slice (�5–10 cm) varies with the size and shapeof the animal (14), and one could never guarantee that most ofthe lung is being represented in all animals. Clinically, thislimitation has to be balanced against the 20-s acquisition in EITvs. the 45-min acquisition in SPECT. The subtraction of thecardiac component of perfusion from mixed pixels while main-taining the net lung component (Fig. 1) by using the EIT-gamma variate algorithm was an essential feature of thismethod. However, at the edges of the heart, some challengingoverlap between the behavior of lung and heart tissue may haveremained, sometimes causing uncertainties in the double-func-tion fitting process. This phenomenon may be responsible forthe lower agreement between SPECT and EIT in the top leftquadrant. Although the location of the cardiac chambers inhumans could make this differentiation easier in the clinicalsetting, more studies on this subject are needed. Fourth, theassumption of the maximal slope method that no tracer leavesthe ROI before the peak artery concentration is reached couldbe violated in the presence of the combination of high bloodflow and low blood volume. As discussed in the APPENDIX A,however, such violations in the assumption would lead to smallestimation errors, provided that blood volume decreases by nomore than one order of magnitude. Another potential limitationof the proposed method is the possibility of some diffusion ofsodium chloride to the outside of the vessels in the lungs. Inthis situation, the solute that remains in the vessel will leave thelungs through the venous drainage while the diffused solutewill tend to stay in the lungs, violating the conservation ofmass principle (unless an extravascular compartment is ac-

counted for). For the calculations of the maximal slope, how-ever, the conservation of mass could still be applied correctly,since one of the assumptions is that there is no outflow ofhypertonic saline before the peak of the pulmonary artery inputfunction. In this case, what the conservation of mass implies isthat all the solute that reaches the ROI (feeding vessel andextravascular compartment lumped together), irrespective ofwhether it remains inside the vessels or not, came through thefeeding artery. Finally, one last important theoretical concern isthat, due to the low spatial resolution of EIT, the maximumslope time point might be slightly displaced (in time) amongsubregions within the ROI, likely causing wrong estimates ofspatially averaged maximum slopes. The extent of this poten-tial limitation deserves future studies.

In conclusion, we describe a novel method based on EIT forestimating regional lung perfusion at the bedside. In bothhealthy and injured lung conditions, the distribution of pulmo-nary blood flow as assessed by EIT agreed well with the oneobtained by SPECT. The method proposed in this study has thepotential to contribute to a better understanding of the behaviorof regional perfusion under different lung and therapeuticconditions.

APPENDIX A

Heuristic Model For Regional Blood Flow

A mathematical model representing three compartments, the rightheart in series with two lung compartments in parallel (compartmentsA and B), was simulated using reference conditions found in normalphysiology: right heart chamber volume of �150 ml; infusion time of�2 s; baseline blood volume for each lung compartment of �150 ml;and total cardiac output of �5 l/min. Each lung compartment had aninflow equal to its outflow, adjusted independently from other com-partments and thus a fixed blood volume over time. This fixed bloodvolume could be different among compartments, depending on sim-ulations. Equations simulating mass balance and concentration curvesinside each compartment were solved numerically by assuming that

Fig. 6. Bland-Altman plots of perfusion for EIT and SPECT. Bland-Altman plots of the SPECT vs. EIT measurements based on the first-pass kinetics andpulsatility method are shown. For both healthy and injured lungs, we analyzed four symmetrically delimited quadrants of the lung. Left: Bland-Altman plotsof the SPECT vs. EIT pulsatility measurements of fractional blood flow to 4 lung quadrants (of axial images). There was systematic overestimation of perfusionto the top left quadrant (P � 0.001), and systematic underestimation of perfusion to the bottom right quadrant (P � 0.01). There was no dependence of thedifference between SPECT and EIT pulsatility in relation to their average magnitude (P � 0.16). Right: Bland-Altman plots of the SPECT vs. EIT measurementsbased on the first-pass kinetics of fractional blood flow to four lung quadrants of the transverse images. Overall agreement was good (limits of agreement �

�10.9% to � 10.5%) although a systematic overestimation of perfusion to the top left quadrant was observed in EIT measurements (P � 0.05). We did notobserve dependence between the difference between SPECT and EIT-NaCl measurements and their average magnitude (P � 0.14).




1) the contrast was infused at the right heart chamber in a short bolus(2 s); 2) the bolus suffered immediate and ideal dispersion; and3) blood flow was constant and nonpulsatile.

We calculated the regional percentage change in resistivity accord-ing to a modification of the formula proposed by Brown et al. (9)based on the volume fraction (V) of the indicator (the hypertonicsaline) contained within the compartment at that particular moment,the conductivity of blood (1 � 0.62 Sm/m), and the conductivity ofthe hypertonic saline (2 � 32 Sm/m).

� � �0

�0

�3 · V · ��2 � �1�

2 · �1 � �2 � 2 · V · ��2 � �1�(A1)

This means that the instantaneous intensity of the EIT signal repre-sented both the volume of the compartment and its final concentrationof sodium.

The estimated regional lung flow was calculated as the maximalslope derived from the fitted gamma curve (34). As in our experi-ments, we searched for the maximal slope from the start to thetime-to-peak perturbation. Percent flow directed to compartment Awas calculated as the maximal slope of that compartment A divided bythe sum of the slopes of compartments A and B.

We simulated changes in the perfusion to compartment A from100% down to a minimum of 10% of the value in compartment B,which was kept constant (Fig. A1, left). The maximal slope of theperturbation found in compartment A (derived from the gamma-fittedcurve) showed a strong linear correlation with the true perfusionsimulated in the model (R2 � 1.00). Additionally, we varied thevolume of blood contained in compartment A from 100% down to10% of the value found in compartment B, while keeping the perfu-sion (input flow) fixed in both compartments, to assess if the estima-tion of the perfusion was affected by changes in the residing bloodvolume (Fig. A1, right). For values of blood volume 30% of theinitial value, the maximal slope measurement was not affected by thedecrease in blood volume. For greater decreases in blood volume(�30% of the initial value), perfusion was underestimated, albeit to anerror of �10%.

Therefore, we concluded that the maximum slope was a robustestimate of perfusion even in the presence of changes in bloodvolume, provided that those changes are not greater than one order ofmagnitude. Previous studies on blood pooling at dorsal, gravity-

dependent lung regions have revealed that a difference of 10 times inblood volume can be found in normal lungs, when comparing the 2most extreme lung regions along the gravity axis. Under such condi-tions, the maximum slope approach would underestimate the perfu-sion of these lung regions with very low blood volume by �10%.

APPENDIX B

New Dilution Curve Fitting Method

We developed an automatic/heuristic solution that ran for every pixelin Fig. A1 in a batch process. Before the gamma fitting, we used afinite-impulse response filter in the time domain to reduce the high-frequency noise. To the filtered signal, we applied the fitting process,described in detail elsewhere (48). Of note, the fitting was noniterativeand made use of only three relevant data points: the time to appear, themiddle-height point, and the peak point. If the curve had two peaks, as inFig. 1B, we localized the two peaks, even when the first one was smallerthan the second one (by using the second and third derivatives), andapplied this gamma fitting to the three data points obtained till the firstpeak. This procedure intrinsically assumed that no contrast entered thelung before the first peak in the right heart. Then we calculated thesubtraction between the raw contrast curve and the gamma-fitted curve ofthe right heart. We subsequently performed a second gamma fitting tothis subtracted curve, obtaining the gamma-fitted curve related to thelung. This procedure was performed on a pixel-by-pixel basis. Badfittings occurred in 5–10% of the cases and could be detected as pixels towhich there was either zero or abnormally high perfusion (outliers).When this happened, we visually detected the right heart area as well asthe typical-lung areas, using videos of the saline passage through the heartand lungs, and used the time to peak information of each of these tworegions to establish boundaries for fitting. For instance, if the first peak ofa two-peaked curve occurred close to the time to peak of the typical-lungpixel, the second peak would be ignored, as opposed to considering thefirst and second peaks as related to the right heart and lung, respectively.

Limitations

In our images, due to the low spatial resolution, it was very difficultto separate the lung parenchyma from the pulmonary vessels ofsecond or third generation. Thus our perfusion data was likely con-

Fig. A1. Model simulations for the influence of blood flow and blood volume on the estimates of lung perfusion based on the maximum-slope of contrastperturbation. Model simulations for the impedance perturbation caused by changes in instantaneous sodium concentration within 2 lung compartments(compartments A and B) after hypertonic bolus injection. Timing and amount of sodium chloride infusion (5 ml of 20% sodium chloride infused within 2 s,injected directly into a third compartment upstream, with 150 ml, simulating the right heart) was kept constant throughout the simulations. Situation ofcompartment B was kept constant (regional blood flow � 2.5 l/min, and regional blood volume � 150 ml), while the blood flow of compartment A changedfrom 2.5 down to 0.25 l/min (left). At right, we simulated a reduction in the blood volume of compartment A, from 150 ml down to 15 ml. As shown, the calculusof the maximum slope in impedance perturbations found in compartment A kept a quasilinear relationship with the true blood flow (relative perfusion) to thiscompartment. As also expected, pure changes in blood volume within this compartment (but with sustained blood flow) caused only a minor change (�5%)in the maximal slope. Ideally, we should see no changes related to such pure blood-volume variations.




taminated by the perfusion of the large pulmonary arteries. This is alimitation when analyzing very small regions of interest, since theinclusion or not of an artery can cause large differences in theestimates (especially if the cross section of the artery belongs to oneregion of interest but its feeding territory belongs to another one).However, when analyzing large regions of interest, as in our study, itis very likely that the perfusion of big arteries is mixed up with thesurrounding parenchyma perfusion, and both are strongly related tothe perfusion of the respective region of interest.

In theory, there might be to some extent diffusion of sodiumchloride to the outside of the vessels in the lungs (characterizing thesystem as “open” instead of closed). In this situation, the diffusedsolute would tend to stay in the lungs, while the solute that remainedin the vessel would leave the lungs through the venous drainage. Toapply the principle of conservation of mass, we would need toconsider a closed system and, therefore, we should consider the threecompartments: the arterial (input), the interstitial (diffused solute),and the venous (output). Since the calculation of the maximal slopeassumes that there was no outflow of hypertonic before the peak of thepulmonary artery input function, the conservation of mass would stillbe applied correctly if we only consider a system with two compart-ments. In this case, what the conservation of mass implies is that allthe solute that has reached the lungs came through the feeding artery,irrespective of whether the solute has remained inside the vessels ornot.

ACKNOWLEDGMENTS

The study was been performed with the help of Agneta Roneus, KarinFagerbrink, Pia Boström, Anna Eriksson, and Aseel Sherif, who are deeplyacknowledged.

GRANTS

Supported by for this study was provided by grants from Fundação deAmparo a Pesquisa do Estado de São Paulo (FAPESP; São Paulo StateResearch Support Foundation), Conselho Nacional de DesenvolvimentoCientífico e Tecnológico (National Council for Scientific and TechnologicalDevelopment), Financiadora de Estudos e Projetos (FINEP; Studies andProjects Financial Support Provider), and Coordenação de Aperfeiçoamento dePessoal de Nível Superior-Ministry of Education, Brazil. It was also supportedby grants from institutional funds from Uppsala University.

DISCLOSURES

S. H. Bohm has consulted for Dixtal Biomedica and is the inventor andowner of patents for EIT technology. A. Melo received honoraria for support-ing the EIT project as an expert electronic engineer. In addition to financialsupport from FAPESP and FINEP, M. Amato received grant support fromDixtal Biomedica. The remaining authors have not disclosed any potentialconflicts of interest.

AUTHOR CONTRIBUTIONS

J.B.B., F.S.-S., S.H.B., G.T., A.M., M.P., E.L.V.C., G.H., and M.A.conception and design of research; J.B.B., F.S.-S., S.H.B., G.T., A.M., E.M.,M.S., M.P., E.L.V.C., G.H., and M.A. performed experiments; J.B.B., A.M.,E.M., M.S., M.P., E.L.V.C., and M.A. analyzed data; J.B.B., F.S.-S., S.H.B.,G.T., E.M., M.S., M.P., E.L.V.C., G.H., and M.A. interpreted results ofexperiments; J.B.B., E.M., M.S., M.P., E.L.V.C., and M.A. prepared figures;J.B.B., F.S.-S., S.H.B., G.T., A.M., E.M., M.S., M.P., E.L.V.C., G.H., andM.A. drafted manuscript; J.B.B., F.S.-S., S.H.B., G.T., A.M., E.M., M.S.,M.P., E.L.V.C., G.H., and M.A. edited and revised manuscript; J.B.B., F.S.-S.,S.H.B., G.T., A.M., E.M., M.S., M.P., E.L.V.C., G.H., and M.A. approvedfinal version of manuscript.

REFERENCES

1. Adler A, Amyot R, Guardo R, Bates JH, Berthiaume Y. Monitoringchanges in lung air and liquid volumes with electrical impedance tomog-raphy. J Appl Physiol 83: 1762–1767, 1997.

2. Adler A, Shinozuka N, Berthiaume Y, Guardo R, Bates JH. Electricalimpedance tomography can monitor dynamic hyperinflation in dogs. J

Appl Physiol 84: 726–732, 1998.

3. Barber DC, Brown BH. Applied potential tomography. J Phys E Sci

Instrum 17: 723–733, 1984.

4. Bodenstein M, David M, Markstaller K. Principles of electrical imped-

ance tomography and its clinical application. Crit Care Med 2009.

5. Borges JB, Carvalho CR, Amato MB. Lung recruitment in patients with

ARDS. N Engl J Med 355: 319–320; author reply 321–312, 2006.

6. Borges JB, Okamoto VN, Matos GF, Caramez MP, Arantes PR,

Barros F, Souza CE, Victorino JA, Kacmarek RM, Barbas CS,

Carvalho CR, Amato MB. Reversibility of lung collapse and hypoxemia

in early acute respiratory distress syndrome. Am J Respir Crit Care Med

174: 268–278, 2006.

7. Brown BH, Barber DC, Morice AH, Leathard AD. Cardiac and

respiratory related electrical impedance changes in the human thorax.

IEEE Trans Biomed Eng 41: 729–734, 1994.

8. Brown BH, Barber DC, Seagar AD. Applied potential tomography:

possible clinical applications. Clin Phys Physiol Meas 6: 109–121, 1985.

9. Brown BH, Leathard A, Sinton A, McArdle FJ, Smith RW, Barber

DC. Blood flow imaging using electrical impedance tomography. Clin

Phys Physiol Meas 13: 175–179, 1992.

10. Brudin LH, Rhodes CG, Valind SO, Wollmer P, Hughes JM. Regional

lung density and blood volume in nonsmoking and smoking subjects

measured by PET. J Appl Physiol 63: 1324–1334, 1987.

11. Clough AV, Linehan JH, Dawson CA. Regional perfusion parameters

from pulmonary microfocal angiograms. Am J Physiol Heart Circ Physiol

272: H1537–H1548, 1997.

12. Costa EL, Borges JB, Melo A, Suarez-Sipmann F, Toufen C Jr, Bohm

SH, Amato MB. Bedside estimation of recruitable alveolar collapse and

hyperdistension by electrical impedance tomography. Intensive Care Med

35: 1132–1137, 2009.

13. Costa EL, Chaves CN, Gomes S, Beraldo MA, Volpe MS, Tucci MR,

Schettino IA, Bohm SH, Carvalho CR, Tanaka H, Lima RG, Amato

MB. Real-time detection of pneumothorax using electrical impedance

tomography. Crit Care Med 36: 1230–1238, 2008.

14. Costa EL, Lima RG, Amato MB. Electrical impedance tomography.

Curr Opin Crit Care 15: 18–24, 2009.

15. Cressoni M, Caironi P, Polli F, Carlesso E, Chiumello D, Cadringher

P, Quintel M, Ranieri VM, Bugedo G, Gattinoni L. Anatomical and

functional intrapulmonary shunt in acute respiratory distress syndrome.

Crit Care Med 36: 669–675, 2008.

16. Di Marco F, Devaquet J, Lyazidi A, Galia F, da Costa NP, Fumagalli

R, Brochard L. Positive end-expiratory pressure-induced functional re-

cruitment in patients with acute respiratory distress syndrome. Crit Care

Med 38: 127–132.

17. Eyuboglu BM, Brown BH, Barber DC, Seagar AD. Localisation of

cardiac related impedance changes in the thorax. Clin Phys Physiol Meas

8: 167–173, 1987.

18. Fagerberg A, Stenqvist O, Aneman A. Electrical impedance tomography

applied to assess matching of pulmonary ventilation and perfusion in a

porcine experimental model. Crit Care 13: R34, 2009.

19. Fagerberg A, Stenqvist O, Aneman A. Monitoring pulmonary perfusion

by electrical impedance tomography: an evaluation in a pig model. Acta

Anaesthesiol Scand 53: 152–158, 2009.

20. Frerichs I, Hinz J, Herrmann P, Weisser G, Hahn G, Quintel M,

Hellige G. Regional lung perfusion as determined by electrical impedance

tomography in comparison with electron beam CT imaging. IEEE Trans

Med Imaging 21: 646–652, 2002.

21. Frerichs I, Pulletz S, Elke G, Reifferscheid F, Schadler D, Scholz J,

Weiler N. Assessment of changes in distribution of lung perfusion by

electrical impedance tomography. Respiration 77: 282–291, 2009.

22. Frerichs I, Schmitz G, Pulletz S, Schadler D, Zick G, Scholz J, Weiler

N. Reproducibility of regional lung ventilation distribution determined by

electrical impedance tomography during mechanical ventilation. Physiol

Meas 28: S261–267, 2007.

23. Gattinoni L, Pelosi P, Crotti S, Valenza F. Effects of positive end-

expiratory pressure on regional distribution of tidal volume and recruit-

ment in adult respiratory distress syndrome. Am J Respir Crit Care Med

151: 1807–1814, 1995.

24. Hahn G, Dittmar J, Just A, Quintel M, Hellige G. Different approaches

for quantifying ventilation distribution and lung tissue properties by

functional EIT. Physiol Meas 31: S73–S84.

25. Hedenstierna G. Using electric impedance tomography to assess regional

ventilation at the bedside. Am J Respir Crit Care Med 169: 777–778, 2004.




26. Heiland S. Diffusion- and perfusion-weighted mr imaging in acute stroke:

principles, methods, and applications. Imaging Decisions MRI 7: 4–12,

2003.

27. Hinz J, Moerer O, Neumann P, Dudykevych T, Frerichs I, Hellige G,

Quintel M. Regional pulmonary pressure volume curves in mechanically

ventilated patients with acute respiratory failure measured by electrical

impedance tomography. Acta Anaesthesiol Scand 50: 331–339, 2006.

28. Konstas AA, Goldmakher GV, Lee TY, Lev MH. Theoretic basis and

technical implementations of CT perfusion in acute ischemic stroke, part

1: theoretic basis. AJNR Am J Neuroradiol 30: 662–668, 2009.

29. Kozian A, Schilling T, Freden F, Maripuu E, Rocken C, Strang C,

Hachenberg T, Hedenstierna G. One-lung ventilation induces hyperper-

fusion and alveolar damage in the ventilated lung: an experimental study.

Br J Anaesth 100: 549–559, 2008.

30. Kunst PW, Vonk Noordegraaf A, Hoekstra OS, Postmus PE, de Vries

PM. Ventilation and perfusion imaging by electrical impedance tomog-

raphy: a comparison with radionuclide scanning. Physiol Meas 19: 481–

490, 1998.

31. Lachmann B, Robertson B, Vogel J. In vivo lung lavage as an experi-

mental model of the respiratory distress syndrome. Acta Anaesthesiol

Scand 24: 231–236, 1980.

32. 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 163: 1444–1450, 2001.

33. Meier P, Zierler KL. On the theory of the indicator-dilution method for

measurement of blood flow and volume. J Appl Physiol 6: 731–744, 1954.

34. Miles KA. Measurement of tissue perfusion by dynamic computed to-

mography. Br J Radiol 64: 409–412, 1991.

35. Montet X, Ivancevic MK, Belenger J, Jorge-Costa M, Pochon S,

Pechere A, Terrier F, Vallee JP. Noninvasive measurement of absolute

renal perfusion by contrast medium-enhanced magnetic resonance imag-

ing. Invest Radiol 38: 584–592, 2003.

36. Mullani NA, Gould KL. First-pass measurements of regional blood flow

with external detectors. J Nucl Med 24: 577–581, 1983.

37. Newell JC, Blue RS, Isaacson D, Saulnier GJ, Ross AS. Phasic three-

dimensional impedance imaging of cardiac activity. Physiol Meas 23:

203–209, 2002.

38. Newell JC, Levitzky MG, Krasney JA, Dutton RE. Phasic reflux of

pulmonary blood flow in atelectasis: influence of systemic PO2. J Appl

Physiol 40: 883–888, 1976.

39. 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 35: 695–702, 1997.

40. Richard JC, Pouzot C, Gros A, Tourevieille C, Lebars D, Lavenne F,

Frerichs I, Guerin C. Electrical impedance tomography compared with

positron emission tomography for the measurement of regional lung

ventilation: an experimental study. Crit Care 13: R82, 2009.

41. Schramm P. CT angiography and perfusion CT in acute stroke: principles

and applications. Imaging Decisions MRI 7: 31–41, 2003.

42. Schuster DP, Haller J. Regional pulmonary blood flow during acutepulmonary edema: a PET study. J Appl Physiol 69: 353–361, 1990.

43. Smit HJ, Vonk Noordegraaf A, Marcus JT, Boonstra A, de Vries PM,

Postmus PE. Determinants of pulmonary perfusion measured by electricalimpedance tomography. Eur J Appl Physiol 92: 45–49, 2004.

44. Smit HJ, Vonk Noordegraaf A, Roeleveld RJ, Bronzwaer JG, Postmus

PE, de Vries PM, Boonstra A. Epoprostenol-induced pulmonary vaso-dilatation in patients with pulmonary hypertension measured by electricalimpedance tomography. Physiol Meas 23: 237–243, 2002.

45. Smith RW, Freeston IL, Brown BH. A real-time electrical impedancetomography system for clinical use– design and preliminary results. IEEE

Trans Biomed Eng 42: 133–140, 1995.46. Suarez-Sipmann F, Bohm SH, Tusman G, Pesch T, Thamm O,

Reissmann H, Reske A, Magnusson A, Hedenstierna G. Use of dy-namic compliance for open lung positive end-expiratory pressure titrationin an experimental study. Crit Care Med 35: 214–221, 2007.

47. Tanaka H, Ortega NR, Galizia MS, Borges JB, Amato MB. Fuzzymodeling of electrical impedance tomography images of the lungs. Clinics

63: 363–370, 2008.48. Thompson HK Jr, Starmer CF, Whalen RE, McIntosh HD. Indicator

transit time considered as a gamma variate. Circ Res 14: 502–515, 1964.49. Trautman ED, Newbower RS. A practical analysis of the electrical

conductivity of blood. IEEE Trans Biomed Eng 30: 141–154, 1983.50. Tusman G, Bohm SH, Tempra A, Melkun F, Garcia E, Turchetto E,

Mulder PG, Lachmann B. Effects of recruitment maneuver on atelectasisin anesthetized children. Anesthesiology 98: 14–22, 2003.

51. Victorino JA, Borges JB, Okamoto VN, Matos GF, Tucci MR, Cara-

mez MP, Tanaka H, Sipmann FS, Santos DC, Barbas CS, Carvalho

CR, Amato MB. Imbalances in regional lung ventilation: a validationstudy on electrical impedance tomography. Am J Respir Crit Care Med

169: 791–800, 2004.52. Vidal Melo MF, Layfield D, Harris RS, O’Neill K, Musch G, Richter

T, Winkler T, Fischman AJ, Venegas JG. Quantification of regionalventilation-perfusion ratios with PET. J Nucl Med 44: 1982–1991, 2003.

53. Vonk-Noordegraaf A II, Janse A, Marcus JT, Bronzwaer JGF, Post-

must PE, Faes TJ, De Vries PM. Determination of stroke volume bymeans of electrical impedance tomography. Physiol Meas 21: 285–293,2000.

54. Vonk Noordegraaf A, Kunst PW, Janse A, Marcus JT, Postmus PE,

Faes TJ, de Vries PM. Pulmonary perfusion measured by means ofelectrical impedance tomography. Physiol Meas 19: 263–273, 1998.

55. Vonk Noordegraaf A, Kunst PW, Janse A, Smulders RA, Heethaar

RM, Postmus PE, Faes TJ, de Vries PM. Validity and reproducibility ofelectrical impedance tomography for measurement of calf blood flow inhealthy subjects. Med Biol Eng Comput 35: 107–112, 1997.

56. Zhao Z, Moller K, Steinmann D, Frerichs I, Guttmann J. Evaluation ofan electrical impedance tomography-based global inhomogeneity indexfor pulmonary ventilation distribution. Intensive Care Med 35: 1900–1906, 2009.

57. Zierler KL. Theoretical basis of indicator-dilution methods for measuringflow and volume. Circ Res 10: 393–407, 1962.




Borges2011 JAP LungPerfusionByEIT

Documents

Transcript of Borges2011 JAP LungPerfusionByEIT