Intensive Care Med (2001) 27: 602±608 DOI 10.1007/s001340100860
Luiz M. Malbouisson Françoise PrØteux Louis Puybasset Philippe Grenier Pierre Coriat Jean-Jacques Rouby
Received: 17 July 2000 Final revision received: 9 October 2000 Accepted: 14 November 2000 Published online: 22 February 2001 Springer-Verlag 2001 This work was presented in part at the 28th Congress of La SociØtØ de RØanimation de Langue Française, Paris, 19±21 January 2000. Dr. Malbouisson was the recipient of a scholarship provided by the Ministre des Affaires Etrangres Français (Ref. 2334471)
L. M. Malbouisson ´ L. Puybasset ´ P. Coriat ´ J.-J. Rouby ( ) RØanimation Chirurgicale Pierre Viars, DØpartement d'AnesthØsie-RØanimation, Hôpital de la PitiØ-Salptrire, University of Pierre et Marie Curie Paris VI, 47±83 boulevard de l'Hôpital, 75013 Paris, France E-mail:
[email protected] Phone: +33-1-42 17 73 00 Fax: +33-1-42 17 73 26
)
P. Grenier Department of Radiology, Hôpital de la PitiØ-Salptrire, INSERM U494, University of Pierre et Marie Curie Paris VI, 47±83 boulevard de l'Hôpital, 75013 Paris, France F. PrØteux Institut National des TØlØcommunications, 9 rue Charles Fourrier, 91011 Evry, France Present Address: Dr. Malbouisson, Department of Anesthesiology, Hospital Dos Clinicas, Universidade de SaÄo Paulo, Brazil
TECHNIQUE
Validation of a software designed for computed tomographic (CT) measurement of lung water
Abstract Objective: The in-vitro validation of a computed tomographic (CT) software specifically designed for quantifying the volume of water contained in the lung. Design: An in-vitro, ex-vivo study. In 1993, a postmortem left pneumonectomy was performed in a patient who died from acute respiratory distress syndrome. The lung was fixed, inflated and dried according to a technique proposed by Markarian and Dailey in 1975 aimed at producing a lung specimen spongy in texture and suitable for radiography. Measurements and results: In 1999, 13 CT scans of this lung specimen were performed corresponding to different bronchial instillations of known volumes of water and albumin 4 %. The different lung weights resulting from the successive bronchial instillations were calculated using a specially designed software, Lungview, adapted for CT measurements and compared with the actual lung weight measured by an electronic scale. The increase in lung weight measured by Lungview was closely correlated with the actual increase in lung weight resulting from bronchial instillation of water and albumin (y = 0.99x 23, r = 1 for water and y = x 17, r = 1 for albumin 4 %) and the precision of the bias was 7 g for water and 3 g for albumin 4 %.
Conclusions: This study shows that the CT software Lungview accurately measured the volume of lung water present within an air-dried exsanguine human lung. Key words Acute respiratory distress syndrome ´ Computed tomography ´ Lung edema
603
Introduction A spiral CT scanner allows an accurate measurement of the volume of many human organs [1] and provides the possibility of calculating the respective volumes of gas and tissue present in the lungs. This calculation is based on the linear correlation existing between the physical density of various phantoms and their CT attenuation [2, 3, 4] and on the assumption that lung tissue has a physical density very close to hydric density. Recently we developed a software ± Lungview ± that served for measuring lung volumes from raw CT data in a series of 48 patients with acute respiratory distress syndrome (ARDS) [5, 6, 7]. The present study was undertaken to validate the ability of Lungview to measure the volume of lung water. An air-inflated exsanguine human lung was filled with known amounts of water or albumin and actual lung weights were compared with lung weights calculated using Lungview.
dry left lung with spongy texture was obtained. The lung was stored in a hermetically sealed bag between 1993 and 1999 without detectable deterioration. Experimental protocol Before filling the lung with fluid, a CT scan was performed. The lung was then filled with known volumes of water or albumin 4 % through the left main stem bronchus and carefully shaken. In the first part of the experiment, the lung was filled with increasing amounts of water : four times 100 ml followed by 300 ml once. After completion of the experiment, the lung was dried with a hair drier until its weight returned to the initial dry weight. The second part of the experiment was performed 1 week later and was similar to the first one, except that human albumin 4 % was used to fill the lung in order to mimic high permeability type lung edema. Six 100 ml aliquots were intrabronchially instilled. Thirteen CT scans of the lung specimen were performed, corresponding to the different bronchial instillations of water and albumin. Before each CT scan acquisition, corresponding to a given bronchial instillation, the isolated lung was placed on an electronic scale and weighed (Electronic Scale, Teraillon BE 201, France).
Methods
Acquisition of the computed tomography
Preparation of the inflated and fixed human lung
Spiral volumetric acquisition of the entire inflated and fixed lung specimen was performed with a Tomoscan SR 7000 (Philips, Eindhoven, The Netherlands). The exposures were taken at 120 kV and 250 mAs. The rotation time was 1 s, collimation 10 mm and pitch of 1. The volume of the voxel was 3.6 mm3. Contiguous axial images 10 mm thick were reconstructed from the volumetric data obtained during a 15-s period using a 180 interpolation algorithm and a soft tissue filter. Images were displayed at lung window settings (1600 HU width and ±600 HU level).
The lung that served for the experiment was prepared according to a technique proposed by Markarian and Dailey in 1975 [8, 9]. This simple and easy-to-implement method is aimed at producing a lung specimen which can be stored for over 10 years without damage [10] and is suitable for histopathology, radiography and CT examinations. In 1993, a postmortem left pneumonectomy was performed in a 65-year-old man who died due to ARDS complicating postoperative bronchopneumonia 5 days after the surgical resection of a thoracoabdominal aortic aneurysm. The pneumonectomy was performed according to French legislation (law no. 781181, December 22, 1976, followed by the statutory order no. 78501 of March 31, 1978 and the implementation order of April 3, 1978) and after obtaining informed consent from the patient's relatives. A thoracotomy was performed in the 5th left intercostal space at the bedside under surgical conditions within 20 min following death. After cessation of mechanical ventilation, both lungs were then removed from the thorax, the trachea being sectioned immediately beneath the larynx. After a careful dissection avoiding lung laceration, the two lungs were separated by a tracheal section at the carina level leaving a long portion of the left main stem bronchus, pulmonary vessels were tied with strings and the left main stem bronchus was cannulated with an endotracheal tube no. 7.5. The left lung was then inflated via the endotracheal tube by a fixative composed of polyethylene glycol 400 (25 %), ethyl alcohol 95 % (10 %), formaldehyde 37 % (10 %) and water (55 %). The fixative was instilled by gravity at a pressure of 30 cmH2O until the lung surface was firmly distended and there were small amounts of fixative weep through the pleural surface. The endotracheal tube was clamped to prevent loss of fluid and the lung specimen was floated in a container filled with the same fixative for 7 days. The lung was then suspended from a ring stand over a drip basin and the endotracheal tube was connected to a source of air equipped with a continuous positive airway pressure system set at a pressure of 30 cmH2O. The air pressure, causing the fixative to weep from the pleural surface, was maintained for 3 days and a
Measurements of lung volumes by Lungview The software Lungview, specifically designed for calculating volumes of lung tissue from the measurement of CT numbers and total lung volume, was developed at the Institut National des Telecommunications, Evry, France. The calculations by the software were based on the following principles of measurement: the volume of the total lung was measured as the total number of voxels present in a given region of interest times the volume of the voxel. The respective volumes of water and albumin were calculated using a method previously described [11, 12] and based on the principle that CT attenuation (expressed as the CT number) and physical density are closely correlated. The CT number characterizing each individual voxel was expressed in Hounsfield units (HU) and was defined as the attenuation coefficient of the X-ray by the material being studied minus the attenuation coefficient of water divided by the attenuation coefficient of water. By convention, the CT number of water was 0 HU. The CT number was scaled by a factor 1000, the CT number of gas being ±1000 HU. According to the exact linear relationship existing between physical density and CT attenuation, a lung area characterized by a mean CT number of ±500 HU was considered as being composed of 50 % gas and 50 % water or albumin. A lung area characterized by a mean CT number of ±200 HU was considered as being composed of 20 % gas and 80 % water (or albumin). Using this analysis, it was possible to compute the volume of water or albumin present in the lung.
604
Fig. 1 The left part of the figure shows three CT sections performed at the same level in the three following conditions: dry lung (upper panel), lung filled with 300 ml of albumin 4 % (middle panel) and lung filled with 600 ml of albumin 4 % (lower panel). On the lower figure, some fluid is seeping out and can be seen as a meniscus which was not taken into consideration in the analysis by Lungview. The corresponding distributions of the CT attenuations are shown on the right of each CT section
In a first step, using a manually controlled cursor, the outside limits of the lung were delineated excluding the part of the hilum containing the trachea and the two main stem bronchi. Voxels were sorted into 256 groups according to the range of CT attenuations from ±1200 HU to +200 HU, each group separated by 5.47 HU. For each compartment of a known number of voxels, the total lung volume and the volume of water or albumin were computed using the following equations: Volume of the voxel = (size of the pixel)2 section thickness (1) Total volume = number of voxels volume of the voxel (2) Volume of water (or albumin) = 1 + (CT/1000) total volume (3)
if the compartment considered has a CT number below 0 or (3'), Volume of water (or albumin) = number of voxels volume of the voxel (4) if the compartment considered has a CT number 0 or more, where CT is the mean CT number of the compartment analyzed. In a second step, the volumes of water or albumin of the left lung were calculated by adding the values of all the compartments analyzed. To validate the measurement of the lung volume with Lungview, the CT scan of a 3 l syringe (MedGraphics, St Paul, Minn.) filled with air was also performed. Statistical analysis The actual and the measured lung weights were compared by linear regression analysis and by the Bland and Altman method [13]. The statistical analysis was performed using Statview 5.0 statistical software (SAS Institute, Cary, N. C., USA). The statistical significance level was fixed at 0.05.
605
Table 1 Measurement by Lungview of lung volumes following bronchial instillations of known volumes of water (HU Hounsfield units) Lung weight measured by electronic scale (g)
Volume of each instilled aliquot (ml)
Mean CT attenuation (HU)
Overall lung volume (ml)
Volume of each instilled aliquot calculated by Lungview (ml)
192 (dry lung) 291 391 495 589 836
± 99a 100 104 94 247
±872 ±804 ±735 ±676 ±614 ±484
1356 1365 1382 1449 1448 1469
± 93b 105 103 89 244
a
The volume of instilled water was measured as the difference between the weight of the lung after the first instillation of water (100 ml) and the weight of the dry lung (192 g)
b
The volume of instilled water was calculated by Lungview as the difference between the calculated volume of water after the first instillation of water (99 ml) and the calculated volume of water in the dry lung (168 ml)
Table 2 Measurement by Lungview of lung volumes following bronchial instillations of increasing volumes of albumin 4 % (HU Hounsfield units) Lung weight measured by electronic scale (g)
Volume of each instilled aliquot (ml)
Mean CT attenuation (HU)
Overall lung volume (ml)
Volume of each instilled aliquot calculated by Lungview (ml)
224 (dry lung) 324 421 521 631 725 825
± 100a 97 100 110 96 100
±842 ±742 ±711 ±651 ±594 ±536 ±480
1320 1350 1397 1454 1518 1531 1557
± 98b 96 104 108 94 100
a The volume of instilled albumin was measured as the difference between the weight of the lung after the first instillation of albumin (100 ml) and the weight of the dry lung (224 g)
Results Figure 1 shows three CT sections with the corresponding distribution of CT attenuations performed at the same anatomical level. The first CT section was performed on the dry lung, the second on the lung filled with 300 ml albumin 4 % and the third on the lung filled with 600 ml albumin 4 %. The distribution of CT attenuations of the dry lung was monophasic with a peak localized at ±929 HU and no CT attenuations greater than ± 600 HU. The distribution of CT attenuations was shifted to the right with increasing volumes of albumin 4 %. Tables 1 and 2 show the lung volumes and the mean CT attenuations of the inflated and fixed lung filled with successive amounts of water and albumin. As shown on Fig. 2, the weight of the lung measured by Lungview was closely correlated to the actual weight measured by the electronic scale, whether the lung was filled with water or albumin. The bias was ±29 g and the precision of the bias was 7 g when water was instilled; the bias was ±15 g and the precision of the bias
b
The volume of instilled albumin was calculated by Lungview as the difference between the calculated volume of albumin after the first instillation of albumin (100 ml) and the calculated volume of water in the dry lung (210 g)
was 3 g when albumin 4 % was instilled. In both cases, a slight underestimation of the actual lung weight was observed. The volume of the 3 l syringe computed by Lungview was 2947 ml.
Discussion This study shows that the CT software Lungview can accurately measure increasing amounts of water or albumin instilled in an inflated and fixed human lung. Methodological considerations The principles on which Lungview relies are based on the linear correlation existing between the CT attenuation and the physical density [2, 3, 4] and on the fact that lungs are exclusively composed of gas and tissue, the human lung tissue being characterized by a density close to hydric density. In addition, CT is very accurate for measuring the volume of any physical structure,
606
Fig. 2 Correlations between lung weight measured by Lungview and the actual lung weight measured by an electronic scale for the dry lung and increasing amounts of extravascular lung water (upper panels) and albumin 4 % (lower panels). In each case, the left panels show the linear regression between the two parameters. The dotted line represents the identity line and the solid line represents the linear regression line. The right panels show the same data according to Bland and Altman's method. The solid line represents the mean value of the difference between the two methods (bias) and the dotted lines represent the precision of the bias ( 2 SD)
the only possible error being related to the error in the delineation of the structure's outlines. Volume measurement by CT is based on the computation of the voxels present in the structure times the volume of the voxel. In the present study, Lungview had a precision of 1 % for measuring the volume of a 3-l syringe and it can be assumed that it was very accurate for measuring the overall lung volume of the experimental dry lung. The ability of Lungview to measure the volume of lung water was in a phantom the structure of which was honeycombed. To take into account the complexity of the anatomy of human airways and alveolar structures, a specially prepared aerated human lung instilled with known amounts of water and albumin was used. However, it has to be pointed out that this model differs in many ways from a lung in situ. The tissue of the dry lung had a density higher than 1, the density of living lung tissue; it did not include blood in pulmonary vessels and the actual air-liquid interface determined by surfactant properties was absent. The lung was no longer contained in the rib cage and had lost its contact with the beating heart. For all these reasons, the excellent correlation found in the present study between the actual
and measured volumes of lung water might not be so close in vivo. First, the presence of the thoracic cage can induce ªbeam hardeningº related to the absorption of X-rays of low energy by radiodense material and in particular by bones [14]. Indeed, the CT X-ray beam covers a range of energies up to a maximum determined by the operating kilovoltage peak. The low-energy components of the spectrum are selectively absorbed by all tissues, but particularly by tissue with high atomic numbers such as bones, thus producing a shift of the spectrum towards high energies, also called ªbeam hardeningº. Second, the complexity of the anatomical shape of the lung increases the risk of ªpartial volumeº effects [15]. The partial volume effect occurs when a single voxel contains both the region of interest (the lung tissue) and also some adjacent dense tissue in the slice (mediastinal structures or rib cage). In this case, the CT attenuation measured will be the mean CT attenuation of all the tissues that occupied the voxel. This source of error increases with the thickness of the CT section and is particularly common at the apex and costodiaphragmatic angles of the lung. Third, the cardiac and vascular motions generate the
607
equivalent of a ªpartial volume effectº in the areas where the lung and the heart and vessels are contiguous. These three factors tend to artefactually increase the mean CT number at the anatomical limits between bones, heart and lung. Last but not least, the CT numbers characterizing a given object are not only related to the physical density of the object but can also be influenced by technical characteristics of the scanner [16]. However, with the most recent generation of scanner this potential error seems of marginal importance [17]. Nevertheless, each CT scanner should be ªcalibratedº before using absolute CT numbers for measuring lung volumes of tissue. Clinical implications The present study demonstrates that software such as Lungview can accurately measure the volume of lung water in excess and, by extension, of lung tissue. In vivo, the physiological meaning of the volume of lung tissue raises some problems of interpretation. This calculation takes into account the normal anatomical struc-
tures present in the healthy lung (lung parenchyma and blood contained in the pulmonary circulation) as well as additional tissue structures related to the ªpulmonary inflammationº: inflammatory cells, cellular debris, hyaline membranes and extravascular lung water. The difference between the volume of tissue measured in a given patient and the volume of tissue normally expected in a normal subject of a similar height and weight allows the determination of the amount of tissue in excess, which could be considered as one of the markers of ªpulmonary inflammationº in patients with ARDS [5]. One limit inherent in the CT assessment of tissue in excess is its inability to separate intrapulmonary blood from lung edema because of a lack of spatial resolution. The excess of lung tissue could also reflect the excess of the amount of extravascular lung water [18, 19, 20, 21] and has been recently used by physiologists to assess prolonged strenuous exercise-induced increase in extravascular lung water [22]. Acknowledgements We are indebted to the technicians of the Department of Radiology for their collaboration during the radiological manipulations and to VØronique Connan for typing the manuscript.
References 1. Breiman RS, Beck JW, Glenny R, Akwari OE, Heaston DK, Moore AV, Ram PC (1982) Volume determination using computed tomography. Am J Roentgenol 138: 329±333 2. Mull RT (1984) Mass estimates by computed tomography: physical density from CT numbers. Am J Radiol 143: 1101±1104 3. Denison DM, Morgan MD, Millar AB (1986) Estimation of regional gas and tissue volumes of the lung in supine man using computed tomograph. Thorax 41: 620±628 4. Wandtke JC, Hyde RW, Fahey PJ, Utell MJ, Plewes DB, Goske MJ, Fischer HW (1986) Measurement of lung gas volume and regional density by computed tomography in dogs. Invest Radiol 21: 108±117 5. Puybasset L, Cluzel P, Gusman P, Grenier P, Preteux F, Rouby J-J and the CT scan ARDS Study Group (2000) Regional distribution of gas and tissue in acute respiratory distress syndrome ± Part 1: Consequences on lung morphology. Intensive Care Med 26: 847±863
6. Rouby JJ, Puybasset L, Cluzel P, Richecoeur J, Lu Q, Grenier P and the CT Scan ARDS Study Group (2000) Regional distribution of gas and tissue in acute respiratory distress syndrome ± Part 2: Physiological correlations and definition of an ARDS severity score. Intensive Care Med 26: 1046±1056 7. Puybasset L, Muller JC, Cluzel P, Coriat P, Rouby JJ and the CT Scan ARDS Study Group (2000) Regional distribution of gas and tissue in acute respiratory distress syndrome ± Part 3: Consequences on the effects of positive endexpiratory pressure. Intensive Care Med 26: 1215±1227 8. Markarian B, Dailey ET (1984) Preparation of inflated lung specimens. In: Heitzman ER (ed) The lung. Radiologic-pathologic correlations, 2nd edn. Mosby, St. Louis Princeton Toronto, pp 4±12 9. Markarian B (1975) A simple method of inflation-fixation and air drying of lungs. Am J Clin Pathol 63: 45±58 10. Bessis L, Callard P, Gotheil C, Biaggi A, Grenier P (1992) High-resolution CT of parenchymal lung disease: precise correlation with histologic findings. Radiographics 12: 45±58
11. Vieira SR, Puybasset L, Richecoeur J, Lu Q, Cluzel P, Gusman PB, Coriat P, Rouby JJ (1998) A lung computed tomographic assessment of positive endexpiratory pressure-induced lung overdistension. Am J Respir Crit Care Med 158: 1571±1577 12. Vieira SR, Puybasset L, Lu Q, Richecoeur J, Cluzel P, Coriat P, Rouby JJ (1999) A scanographic assessment of pulmonary morphology in acute lung injury. Significance of the lower inflection point detected on the lung pressure-volume curve. Am J Respir Crit Care Med 159: 1612±1623 13. Bland JM, Altman DG (1986) Statistical methods for assessing agreement between two methods of clinical measurement. Lancet I:307±310 14. Drummond GB (1998) Computed tomography and pulmonary measurements. Br J Anaesth 80: 665±671 15. Williams G, Bydder GM, Kreel L (1980) The validity and use of computed tomography attenuation values. Br Med Bull 36: 279±287 16. Levi C, Gray JE, McCullough EC, Rattery RR (1982) The unreliability of CT numbers as absolute values. Am J Roentgenol 139: 443±447
608
17. Kemerink GJ, Lamers RJ, Thelissen GR, Van Engelshoven JM (1995) Scanner conformity in CT densitometry of the lungs. Radiology 197: 749±752 18. Hedlund LW, Effmann EL, Bates WM, Beck JW, Goulding PL, Putman CE (1982) Pulmonary edema: a CT study of regional changes in lung density following oleic acid injury. J Comput Assist Tomogr 6: 939±946
19. Hedlund LW, Vock P, Effmann EL, Lischko MM, Putman CE (1984) Hydrostatic pulmonary edema. An analysis of lung density changes by computed tomography. Invest Radiol 19: 254±262 20. Forster BB, Muller NL, Mayo JR, Okazawa M, Wiggs BJ, Pare PD (1982) High-resolution computed tomography of experimental hydrostatic pulmonary edema. Jpn Heart J 23: 697±709
21. Scillia P, Delcroix M, Lejeune P, Melot C, Struyven J, Naeije R, Gevenois PA (1999) Hydrostatic pulmonary edema: evaluation with thin-section CT in dogs. Radiology 211: 161±168 22. Manier G, Duclos M, Arsac L, Moinard J, Laurent F (1999) Distribution of lung density after strenuous, prolonged exercise. J Appl Physiol 87: 83±89
