Mechanical Ventilation Redistributes Blood to Poorly Ventilated Areas in Experimental Lung Injury*

Supplemental Digital Content is available in the text.

were taken before and after the PEEP trial sequence.
Three-material differentiation algorithm DECT images were post-processed using a bespoke algorithm that takes as inputs CT images of the same region of lung taken at two energy levels and six 'DECT coefficients', which represent the CT attenuation coefficients of three materials of interest at each energy level.
For this study, the materials of interest were chosen as gas, soft tissue and iodinated blood.
The DECT coefficients for gas were fixed at -1000 Hounsfield Units (HU) for each energy level as per the definition of HU. The coefficients for soft tissue were chosen as the mean attenuation of the liver during each scan series and those for iodinated blood as the mean attenuation of the descending aorta in the last frame of the series. These two structures were chosen because the liver represents a well-perfused soft-tissue organ with a reasonably radiologically homogenous parenchyma similar to the lung with the notable exception that it does not contain gas, and the contents of a large blood vessel can be assumed to contain only blood. These choices of coefficients ensure that any parenchymal and intravascular accumulation of iodine across ventilatory conditions was accounted for in the calculations.
The algorithm then attempts to find the optimum values for the fraction of each material that comprise each voxel of the source images with the constraints that each fraction is between 0 and 1 and all three fractions sum to 1. The algorithm used is available under a permissive licence (3).

CT segmentation and analysis
Following three-material differentiation, images were manually segmented in 3D Slicer version 4.8.1 (4) to include lung parenchyma and exclude extra-thoracic contents, the heart, mediastinal contents, inferior vena cava and major airways and bronchi. Each individual image was then classified as recorded either in inspiration or expiration, or marked as intermediate based upon the mean gas volume fraction of the total slice. Specifically, for each ventilatory condition in each animal, the mean gas volume fraction for each 1 Hz frame was linearly scaled to lie between 0% and 100% where 0% represented the mean gas volume fraction of the frame with the least amount of gas and 100% represented that of the frame with the greatest. All frames with a gas volume fraction of ≤30% on this scale were deemed to represent expiration and all with a gas volume fraction ≥70% were identified as inspiratory frames. Images were automatically segmented further into three regions of equal height aligned along the gravity vector and regional volume fractions of gas and iodinated blood calculated.

Calculation of normalized gas and blood volumes
Individual region gas or blood volumes within each slice (V or Q respectively) were calculated based upon the known volume fractions of gas or blood within each region and the volume of the region: A scale factor was calculated for each individual ventilatory condition per animal for both inspiration and expiration based upon the total volume of both the slice in the dynamic series and the entire lung (including gas, parenchyma and blood) taken at breath holds in either inspiration or expiration respectively: = ! !ℎ ! Whole lung scans taken in expiration at each PEEP setting were also used to determine regional lung tissue mass. The volume was manually segmented to include only the lung itself and exclude extra-thoracic contents and then automatically segmented into three gravitational regions as described above for a single slice. The mass of each region was calculated for each breath hold by assuming that the water density of the slice was 1 minus the gas volume fraction, and then mean lung mass of each region was calculated. 'Wholelung equivalent' gas and blood volumes were calculated for each region such that they represented the expected volume of gas or blood within the entire lung should the composition of the lung be the same as that of the slice. These were then normalized to the mean tissue mass in that region for that particular animal.
Where VN and QN represent normalized gas and iodinated blood volumes within a particular region of the lung respectively.
The entire process is summarised in Fig. 1c.

Measurement of atelectatic mass fractions
Volume DECT scans of the whole lung were obtained during both end-expiratory and endinspiratory breath holds after acquisition of dynamic images in each ventilatory condition.
Each breath hold was 10 s in duration and the scan sequence was started once airway flow fell to zero. Volumes were manually segmented to exclude major intra-pulmonary airways and blood vessels, and extra-thoracic and mediastinal contents. The three-material differentiation algorithm was then applied. Atelectatic subregions were defined as those regions with gas volume fraction ≤ 0.1 (equivalent to regions ≥ -100 Hounsfield Units (HU) on single-energy non-contrast scans as previously described (5)). The mass of this subregion and of the whole lung were calculated based upon region volume and mean tissue density (defined as 1 -gas density). This density is equivalent to the sum of the soft tissue and iodinated blood fractions, assuming each has the density of water. This calculation was chosen, rather than soft tissue alone, to allow comparison with published single-energy CT results where it is impossible to distinguish between soft tissue and blood volume (5).
Finally, the fractional atelectatic mass (FAM) was determined as the ratio of the mass within the atelectatic subregion to that of the whole lung and reported in both end-expiration (FAMexp) and end-inspiration (FAMinsp). The difference between these two measurements for each ventilatory condition in each animal was reported as cyclical recruitment/derecruitment (R/D).

Validation experiments
Eleven 10 mL plastic syringes were prepared containing heparinized pig's blood mixed with iodine contrast agent at concentrations between 0 and 10 mg/mL in 1 mg/mL increments and DECT images obtained as previously described. The CT attenuation coefficients of noniodinated (0 mg/mL) and iodinated (10 mg/mL) blood were measured as the mean of 3 mm radius x 30 mm depth (six 5 mm slices) cylinders concentric with the relevant syringe and these values used as the DECT coefficients for non-iodinated and iodinated blood respectively. Mean voxel iodinated blood fraction for a similar sized cylindrical sample of each syringe content was then obtained and multiplied by 10 mg/mL to give a final DECTderived iodine concentration.
For in vivo validation of gas volumes a whole lung DECT scan (voxel size 0.5x0.5x5 mm) of pig thorax was obtained during breath holds at end-inspiration, followed by a release of lung volume to end-expiration, without any intervening breaths. Images were segmented to exclude extra-thoracic and mediastinal contents, and the inferior vena cava and included trachea up to the level of the tip of the tracheostomy tube. DECT coefficients for iodinated blood were taken as the mean CT densities of the mid-thoracic descending aorta at each energy level and soft tissue as the mean CT densities of the liver parenchyma. Mean gas volume fraction was multiplied by total segmented volume to give a gas volume and the difference between inspiratory and expiratory volumes taken as the DECT-derived tidal volume. Spirometry tidal volume was measured at the airway and converted to BTPS conditions prior to comparison with DECT-derived volumes. PEEP was varied between 0 and 20 cmH2O and approximately 1 minute prior to each scan a dose of iodine contrast was administered to ensure the algorithm gave consistent results for gas volume fraction over a wide range of cumulative iodine doses.
Validation experiments were analysed using simple linear regression and the method of Bland and Altman (6). In vivo gas volume experiments were also analyzed using a modification of the Bland-Altman technique where cumulative iodine dose and endexpiratory lung volume were also represented along the x-axis to exclude any non-constant bias due to these variables. For five animals tidal volume was fixed at 10 mL/kg and PEEP varied between 0 and 20 cmH2O, and tidal volume was also varied for another three animals.

Single-slice vs whole lung analysis
We sought to demonstrate that the single juxtadiaphragmatic slice chosen was representative of the whole lung in injured animals in both expiration and inspiration. The mean densities of the single slice and whole lung in the seven animals in the main experiment were calculated as the sum of the soft tissue and iodinated blood densities. This is equivalent to 1 minus the air density i.e. similar to the calculation of mean lung density in a twocompartment single-energy CT experiment (5). For each ventilatory condition in each animal mean densities of each slice during expiration and each end-expiratory whole lung volume scan were calculated (n=7) and in five of these animals equivalent volumes in inspiration/end-inspiration were also measured.
In order to assess the cranio-caudal displacement of a single slice of the lung between expiration and inspiration, single-energy volume CT scans were obtained during endexpiration and end-inspiration breath holds in 4 animals with voxel sizes of 0.5x0.5x0.6 mm.
The expiration images were then registered onto the inspiration images using NiftyReg software (7), which uses the free-form deformation technique to generate a mesh of B-spline transformations by minimizing a cost function defined as the difference between the moving (expiration) and fixed (inspiration) volumes. It has previously been used to assess regional strain in the lung (8,9). Following manual inspection of the accuracy of the registration, a 5 mm thick slice of lung at the level of the juxtadiaphragmatic slice used in the dynamic images was segmented in the expiration volume and the mean cranio-caudal displacement in each of