Imaging the Injured Lung: Mechanisms of Action and Clinical Use

Plain chest radiography enabled the original description of acute respiratory distress syndrome (ARDS),¹  and—decades later—computed tomography revealed that diffuse lung densities are in fact heterogeneously distributed.²  In retrospect, this is easy to understand: plain radiographs provide a two-dimensional image of the chest (fig. 1A) but, except for classical pointers (e.g., silhouette sign), cannot reliably indicate where in the dorsal–ventral dimension the infiltrate is located, nor can it differentiate among combinations of superimposed aerated, consolidated, or atelectatic lung.³  The most obvious advance provided by the computed tomography scan was the identification, in patients managed in the supine position, of the massive loss of lung aeration predominating in dorsal regions, whereas ventral regions remained partially or fully aerated and received the most part of tidal volume (fig. 1B).^2,4,5  This concept of the “baby lung” enabled the “spatial” rationale for tidal volume (V_T) limitation^6,7  and provided important evidence for prone positioning for patients with ARDS,^8,9  as well as for the concept of recruiting nonaerated lung.^5,10,11 

The idea of the baby lung as the vulnerable region (receiving the bulk of each V_T) received support—decades later—from positron emission tomography demonstrating colocalization of ventilator-induced stretch in the baby lung (computed tomography) and ventral inflammation (positron emission tomography), in proportion to the degree of stretch.^12,13  Using sophisticated computed tomography imaging¹⁴  with high spatial resolution,¹⁵  recent studies suggest that the concept of the baby lung can be further refined, because the lung aeration in ARDS is inhomogeneous even at a far smaller scale—too small to be distinguished by computed tomography—such that microscale regions surrounded by atelectatic lung may undergo disproportionate degrees of (unrecognized) stretch, analogous to the macroscale baby lung.^16–18  This may mean that there are two levels of baby lung, one macroscopic and mostly ventral and the other microscopic and diffuse, and different imaging would be necessary to visualize and manage each.

Although plain radiography and computed tomography contribute to patient care in ARDS,³  risks of transport, radiation, and limitations of interpretation continue to be challenges. In addition, the complexity of image processing is an important obstacle.¹⁹  Lung ultrasound and electrical impedance tomography increasingly allow serial assessment of the lung (and thorax) at the bedside and are increasingly reported to assist in monitoring disease progression, response to intervention, and titration of ventilator settings.^20–26  However, challenges remain in terms of spatial resolution and tissue penetration. Magnetic resonance imaging had found limited use in acute respiratory medicine, but recent methodologies have advanced our knowledge of lung function^16,27–29  and metabolism.^30,31 

Overriding these developments is recognition of the inherent limitations of the current clinical criteria for ARDS.³²  Because these are based on oxygenation, end-expiratory pressure, and the presence (vs. absence) of pulmonary infiltrates on plain radiography, it is not possible to distinguish among several phenotypic³³  or physiologic characteristics³⁴  that likely affect outcome and response to intervention.³⁵  Indeed, such variability confounds the results of most clinical trials³⁶  and complicates patient selection for complex or high-risk treatments, such as extracorporeal lung assist or stem cell therapy. A major opportunity for the field will be incorporation of lung imaging—together with emerging biologic techniques—to help characterize patients or groups of patients and thereby facilitate clinical discovery and management in ARDS onto a progressively more biologic basis.³⁷ 

The chest radiograph has for decades been a cornerstone for diagnosis and management in acute medicine. All definitions to date^38–40  have included acute infiltrates on chest radiograph as requirements for diagnosis of ARDS (fig. 1). Although newer imaging modalities are emerging into clinical practice, the traditional chest radiograph has deep roots in the culture and practice of critical care.

Plain chest x-ray yields a two-dimensional projection image, typically acquired from frontal and sometimes additional lateral views. There are two basic elements: an x-ray generator and a detector. X-ray images display the contrast between organs based on the attenuation (i.e., absorption) of x-ray energy. For example, the lung mostly contains air and, compared with the rib cage, has minimum attenuation of the x-ray beam. Loss of lung aeration caused by edema or infiltration in the lung is contrasted against healthy (i.e., air-filled) parenchyma.

A standard chest radiograph delivers a radiation dose of approximately 0.1 mSv (comparable with the dose absorbed in a transoceanic flight). Because there are two (not three) dimensions, the absolute pixel value in x-ray cannot correspond to a specific tissue density, unlike the fixed scale in computed tomography; for this reason, quantitative analysis of plain chest radiograph is difficult. Radiologists instead use the relative contrast of images, and pattern recognition skills, to formulate diagnoses. Critically ill patients are imaged by portable machines, where the generator is at closer distance to the detector than in ambulatory radiography, and the x-ray beam originates anteriorly to the patient.

Chest x-ray is highly accessible; it is a standard in every intensive care unit, and aside from diagnosis (presence of bilateral infiltrates, and absence of suggestion of a cardiac cause), it is used in the course of ARDS to complement clinical assessment and parameters for gas exchange and mechanics, to determine overall trajectory, in addition to the identification of intercurrent issues such as pleural effusion, atelectasis, pneumothorax, and correct placement of endotracheal tubes and central venous catheters.

Although the Berlin definition allows using computed tomography in alternative, the chest radiograph is an integral component of the ARDS criteria.³⁹  Chest radiograph has been used to grade ARDS severity⁴⁰  and to monitor disease progression and treatment responses. The description of bilateral infiltrates in ARDS has not changed significantly since the original publication; indeed, the authors described patchy infiltrates that were “frequently confused with acute heart-failure.”¹  Opacities are initially hazy, mostly symmetric, and have ill-defined vascular margins that tend to disappear as the infiltrates become more confluent, leading to a loss of diaphragmatic and cardiac margins.⁴¹  Signs of fibrosis or extraalveolar air (interstitial emphysema, pneumothorax) may appear with prolonged ventilation, especially with elevated V_T.

The traditional x-ray is a century-old imaging tool. Modern image acquisition has shifted from analog films to fully digital images. Contemporary images are instantly produced and available on the device seconds after the examination has been performed. Current x-ray systems are becoming more compact and portable, are used in the intensive care unit with minimum care disruption, and are almost always instantly digitized.

Compared with the standard posteroanterior technique, portable chest radiograph has technical limitations decreasing image quality. These include geometric distortion, scattered dose fraction, lower energy, presence of wires and tubes, and motion artifacts attributable to breathing.⁴²  Agreement on interpretation of chest radiograph is surprisingly poor. Chest radiograph has limited accuracy (39 to 70% sensitivity and 70 to 100% specificity when using computed tomography as reference) in detecting radiologic abnormalities characteristic of ARDS.^42,43  Although bilateral infiltrates on plain radiograph (fig. 1A) are required to diagnose ARDS,³⁹  it is often impossible to reliably differentiate among atelectasis, pleural fluid, consolidation, and hydrostatic or permeability edema.^3,44  In fact, bilateral infiltrates do not help accurately identify diffuse alveolar damage.^45–47  The agreement among experts as to the presence of pulmonary infiltrates (without regard to what the infiltrates represent) in patients being evaluated for ARDS is low (κ statistic for agreement is 0.38 to 0.55).^48,49  This has not improved (κ = 0.50) with the more recent Berlin definition,⁵⁰  despite the radiographic criteria being more explicit (bilateral opacities—not fully explained by effusions, lobar/lung collapse, or nodules), the availability of a set of example images,³²  and the adoption of a training program.⁵¹ 

The diagnostic accuracy of plain radiography can be improved with the aid of semiautomatic algorithms,⁵²  becoming fully automated with machine learning and deep learning.⁵³  In deep learning methodologies, computers are trained to recognize features directly from images in a fully automated manner. In a recent study, the trained deep learning algorithm outperformed practicing radiologists for the detection of pneumonia and other pathologies in chest x-ray images,⁵³  notwithstanding the fact that relevant patient information was not made available to the computer programs. Further improvements in artificial intelligence and deep learning algorithms will be possible with the availability of extensive data sets, including clinical information. To facilitate this progress, the National Institutes of Health have recently shared a new database (ChestX-ray14, https://www.nih.gov/news-events/news-releases/nih-clinical-center-provides-one-largest-publicly-available-chest-x-ray-datasets-scientific-community. Accessed January 9, 2019.)⁵⁴  consisting of more than 100,000 images of more than 30,000 unique patients, together with the radiology reports. Furthermore, the limitations of computation cost and algorithm complexity will almost certainly be easy to overcome over time. Although these technologic improvements are likely to augment the reliability of chest radiography, it is uncertain whether the specificity (e.g., for lung inflammation or other biologic processes) will be improved, and the limitations of consensus ARDS definitions may persist.

Computed tomography represents thoracic structures in three dimensions; it refines the morphologic assessment of the lung aeration in ARDS and, combined with quantitative analysis of regional tissue density,⁵⁵  it powerfully measures the severity of lung injury. Recent image processing yields spatial resolution approaching the acinar level and provides detailed maps of regional ventilation⁵⁶  and lung stretch.^57,58 

Computed tomography provides three-dimensional representations of anatomic structures by reconstructing images of the same object obtained from multiple directions as the x-ray generator and the detector rotate. Computed tomography represents each element of tissue volume (i.e., each voxel) within the reconstructed image by quantifying its density according to attenuation of x-ray energy in relation to typical values for air and water (fig. 2A). The spectrum of density is visualized with an arbitrary grayscale that ranges from white to black and is adjusted by the radiologist.

In a modern (multidetector spiral computed tomography) scanner, the x-ray generator and detector are rigidly fixed on opposite sides of the subject, rotating in the same direction (i.e., both clockwise or both counter-clockwise), creating a helical path around the subject. This substantially reduces acquisition time, motion artifacts, and radiation exposure (≈7 mSv for a high-resolution spiral thorax computed tomography, compared with ≈0.1 mSv for a standard chest radiograph and ≈12 mSv for computed tomography pulmonary or coronary angiography).

Computed tomography has multiple uses in assessment, intervention, and research.

Compared with plain radiography, computed tomography is more accurate for the detection and localization of pleural effusion, pneumothorax, alveolar–interstitial syndrome, atelectasis, and lung consolidation.^59,60  Although not necessary for a diagnosis of ARDS,³⁹  a ground glass appearance on computed tomography (fig. 1C; i.e., partial, but not abolished, aeration) is characteristic.⁶¹  Computed tomography might improve the accuracy of ARDS diagnosis versus chest radiograph,⁴³  but correlation with pathologic standard is unclear.

In the supine position, complete opacification of posterior and caudal lung regions is typical, attributable to alveolar atelectasis, infiltration, and flooding. In ARDS, predominant nondependent⁶²  opacities suggest lung superinfection, whereas in the absence of ARDS, “tree-in-bud”⁶³  opacifications or consolidation of lower lobes⁶⁴  suggests community-acquired and ventilator-associated pneumonia. Important ARDS mimickers such as interstitial lung disease or bronchiolitis obliterans with organizing pneumonia are readily distinguished.⁶⁵ 

Computed tomography impacts clinical decisions in more than 20% of cases.^59,66  Chronic phases of ARDS are characterized by progressive fibrosis,⁶⁷  fibrotic reticulations, traction bronchiectasis (i.e., bronchial dilatation generated by parenchymal loss), and cysts that evolve into a “honeycomb” pattern.⁶¹  In addition, bullous lesions and extrapulmonary air reflect barotrauma and hyperinflation attributable to mechanical ventilation (fig. 1D).^67,68  Adverse prognosis is associated with greater proportions of abnormal tissue (e.g., consolidation, ground glass opacification)⁶⁹  and fibroproliferative changes.⁷⁰ 

For quantitative analysis beyond routine interpretation, further image processing is needed, as follows: the lungs are segmented (from nonpulmonary structures), and the tissue densities are averaged and analyzed within regions of interest, such as horizontal slices.¹⁰  In ARDS, segmentation requires time and experience to distinguish high-density parenchyma from effusion or chest wall, but new algorithms automate this task.^71–73 

X-ray attenuation (i.e., absorption) in each voxel is expressed in Hounsfield units after calibration against reference standards (i.e., density of water yields 0 Hounsfield units, and density of air is –1,000 Hounsfield units; fig. 2A). Assuming that completely nonaerated (degassed) lung tissue has density similar to water (0 Hounsfield units), it can be stated that the density of each voxel reflects its relative proportions of gas versus tissue.^74,75  Thus, normal lung tissue is –700 Hounsfield units (corresponding to 70% air, 30% tissue).⁷⁶  Within the normal inflation range (–500 to –900 Hounsfield units), higher Hounsfield units values are more frequent at low lung volumes (e.g., functional residual capacity), and lower Hounsfield units values are more common at higher volumes (e.g., approaching total lung capacity). Atelectasis, edema, and infiltrates have Hounsfield units values close to 0, indicating absence of aeration, and hyperinflated lung (emphysema, or ARDS during mechanical ventilation) is less than –900 Hounsfield units.⁷⁷  Densities from –100 to –500 Hounsfield units (e.g., ground glass) indicate decreased (but not abolished) gas content.⁷⁸ 

Each voxel (≈1 mm³) contains up to 170 alveoli,⁷⁹  and therefore computed tomography necessarily involves a degree of tissue averaging within each voxel, resulting in increased levels of intermediate density (–100 to –500 Hounsfield units) when ventilated and nonventilated airspaces are mixed (fig. 2B).

Although lung structure and aeration are deduced from standard volumetric images, functional characteristics can be determined using specialized imaging protocols. Tidal variation in structure or aeration may be assessed between (pairs of) volumetric images that are acquired at end-expiratory and end-inspiratory breath-holds.¹⁴  The deformation and motion of thoracic tissue between the image pairs may be estimated using three-dimensional image registration,^56,80–82  an image processing technique that aligns two or more images using the same spatial coordinates (fig. 3). Specialized registration functions for lung image processing^83,84  correct for changes in tissue density associated with extremes of lung deformability (e.g., attributable to atelectasis). After registration, each matching voxel can be tracked across aligned images when the shape of the entire lung changes as a result of inspiration, expiration, or progression of injury (see also Supplemental Digital Content 1, http://links.lww.com/ALN/B848).

Acquisition of multiple volumetric images, characterizing thoracic motion throughout the breathing cycle, is possible using respiratory-gated image reconstruction, based on a surrogate lung volume signal or fixed respiratory rate.^81,84–86  As shown in figure 4 and video Supplemental Digital Content 2 (http://links.lww.com/ALN/B849), tidal recruitment results essentially from the transformation of poorly aerated into normally aerated lung regions and marginally from reaeration of nonaerated lung regions. The same behavior is observed concerning positive end-expiratory pressure (PEEP)–induced lung recruitment.⁷⁸  Four-dimensional image registration, estimating the motion of thoracic structures across space and breathing phase, expresses variation in regional lung strain and aeration,⁸⁷  as well as out-of-phase lung motion. The accuracy of image registration was evaluated in an international challenge, although ARDS was not included.⁸⁸  However, image segmentation and registration remain complex tasks for automated processing,⁸⁹  especially in lung injury where consolidated or atelectatic lung limits structural reference points or contrast against the local chest wall.^56,71 

Regions of low ventilation may be identified using wash-in of a radiopaque tracer gas, such as xenon or krypton.^58,90–92  Although image quality for tracer gas techniques is low, initial reports suggest that it may be enhanced by dual-energy computed tomography.^92–95  Dual-energy computed tomography relies on the contrast produced by sudden increases in photon absorption at specific energy levels—so-called “K-edges”—that denote the binding energy of K-shell electrons of atoms interacting with photons. Subtraction of images acquired using photon energies just above and below the K-edge of a tracer gas provides a high-contrast image of gas distribution.^95–98 

Computed tomography has provided major insights into our understanding of ARDS, in the following areas.

In the supine position, hyperdensities predominate in dependent (dorsal) lung, whereas aerated tissue predominates in the nondependent (ventral) lung.²  It is possible to estimate the weight of each horizontal slab of tissue (product of the density multiplied by the tissue height) yielding the superimposed pressure applied on each lung region^99,100  that contributes to the vertical gradient of pleural pressure.¹⁰¹  This gradient is increased in lung injury and governs the distribution of regional density, supporting a model where dependent loss of aeration is explained, in part, by compression from overlying edematous lung,¹⁰⁰  mediastinum,¹⁰²  and abdominal pressure,⁵  in addition to the constrained shape matching of the lung and the thorax.¹⁰³ 

In ARDS, aerated lung is typically reduced to less than 50% of normal capacity,¹⁰⁴  and this is mostly located in the ventral baby lung—or as multiple smaller aerated areas scattered within the injured lung.⁴  The baby lung (fig. 1B) receives all the inhaled gas, and tidal stretch in the baby lung is therefore disproportionally large. Such hyperinflation in ventral lung⁶⁸  explains why this region is especially susceptible to ventilator-induced injury; it also underscores why high V_T so readily causes lung injury, providing the rationale for current ventilator management of ARDS⁶ 

However, computed tomography scans demonstrate that the size of the baby lung is variable among patients with ARDS; thus, a fixed low tidal volume may expose patients with very small lung capacity to overdistension.¹⁰⁵  Conversely, it explains why very low V_T could cause underdistension and atelectasis in those with very large lung capacity. Such insight has led to development of driving pressure instead of V_T as a potential ventilation target.^106,107 

Recruiting poorly and nonaerated lung with PEEP or recruitment maneuvers increases aeration,^10,99,108  and this can be expressed as decrease in weight of nonaerated lung¹⁰⁹  or as increase in gas content within poorly and nonaerated lung regions.⁷⁸  Computed tomography studies show that lung reaerated during inflation might not remain aerated during expiration unless adequate PEEP is provided (fig. 5).¹¹⁰  This is clinically important given the experimental evidence that unstable recruitment causes substantial intrapulmonary shunt¹¹¹  and may worsen ventilator-induced injury.¹¹²  Furthermore, patients with ARDS who have massive loss of aeration (on computed tomography) have higher mortality,¹⁰⁹  making loss of aeration either a marker of severity or a treatable factor. However, globally applied strategies to maintain lungs aerated have not increased survival.¹¹³  This could be related to the fact that patients with focal versus nonfocal loss of aeration do not respond to recruitment maneuvers.¹¹⁴ 

In the supine position, the vertical gradients of pleural pressure are such that higher airway pressure (e.g., PEEP) preferentially distributes gas to the nondependent (rather than the dependent) lung, causing preferential nondependent hyperinflation rather than dorsal recruitment. This was noticed using plain radiography and led to the hypothesis that prone positioning favors more homogeneous aeration by decreasing the vertical pleural pressure gradient.⁸  Later, studies using computed tomography confirmed that the vertical gradient of computed tomography density (superimposed pressure) is attenuated when prone⁹  and that this lessens atelectasis, consolidation, cyclic derecruitment, and hyperinflation,^115,116  which together may explain the lower mortality associated with prone positioning in ARDS.¹¹⁷ 

Some patients share similar radiologic appearance, treatment responses, or biologic characteristics.^35,114,118  For example, the prominence of symmetric ground glass instead of focal opacification¹¹⁹  suggests a nonpulmonary cause, reflecting blood-borne mediation of inflammation. In contrast, nonfocal density distribution versus prominent dependent loss of aeration suggests a favorable response to PEEP,¹²⁰  higher mortality,¹²¹  and increased levels of circulating marker of alveolar cell injury such as the soluble form of the receptor for advanced glycation end product.¹¹⁸  Such features point to a high-severity phenotype with widespread pulmonary edema.¹⁰⁹  With this rationale, the recently concluded LIVE Study (lung imaging morphology for ventilator settings in acute respiratory distress syndrome) tested the hypothesis that an imaging-guided ventilator strategy (targeting the different phenotypes) improves outcomes compared with a conventional (standardized) approach.¹²² 

Measurements of pressure and flow at the airway opening have been associated with distributed mechanical phenomena throughout the lungs, including nonlinear deflections in the dynamic pressure–volume curve associated with recruitment and overdistension.^123–126  However, computed tomography imaging at end expiration and end inspiration revealed intratidal variations in recruitment and overdistension that conflicted with predictions from pressure–volume data in injured lungs (despite there being good agreement under healthy conditions).¹²⁷  Instead, minimizing dynamic compliance in injured lungs by PEEP titration was found to be associated with reductions in both overdistension and intratidal recruitment. These findings highlight the value of medical imaging for heterogeneous lung injury and indicate that dynamic mechanical alterations in the lung are difficult to quantify using only aggregate pressure–volume data, especially where recruitment and overdistension coexist.^128,129 

Measuring regional lung function at very high resolution (i.e., subsegmental, acinar, alveolar) enables more accurate biologic characterization. Injured lung inflates nonuniformly; although this can cause highly localized extremes of mechanical stress,¹⁷  it can only be visualized with high spatial resolution. Computational calculation indicates the greatest inhomogeneity surrounding each voxel at interfaces between aerated and nonaerated tissue and at anatomical structures (“stress raisers”).¹³⁰  Preliminary reports of this method in patients with ARDS suggests that the extent of stress raisers reflects severity of injury.¹⁵ 

Photon-counting computed tomography produces an image based on unique spectral signatures¹³¹  yet relies on only a single x-ray source and specialized detectors capable of distinguishing among individual photons, whereas conventional detectors integrate all photon energies. Photon counting may therefore enable contrast-based functional imaging similar to dual-energy computed tomography, while involving reduced radiation exposure as well as enhanced ability to distinguish multiple contrast agents simultaneously.

ARDS results from the propagation of lung inflammation, initially localized (to one or more areas) and then generalized. Computed tomography images suggest that this propagation may be driven or amplified by inspiratory stretch (see video in Supplemental Digital Content 3, http://links.lww.com/ALN/B850).¹³²  Image analysis using parametric response maps can analyze inflation with voxel-by-voxel precision.¹³³  Parametric response maps are plots of density distribution obtained from coregistered inspiratory and expiratory computed tomography images (fig. 6, upper panels). A pattern of suboptimal aeration and large tidal density swings (termed “unstable inflation”) is associated with increased propagation of experimental injury¹⁴ ; this was attenuated by prone positioning,¹³⁴  and unstable inflation may predict mortality in patients with ARDS (fig. 6, lower panels).¹⁴  Thus, unstable inflation may be a treatable target in ARDS.

The coupling between ventilation and perfusion in response to regional oxygenation tension is closely regulated by a variety of mechanisms.¹³⁵  In ARDS, regional perfusion may be highly variable,¹³⁶  and impaired hypoxic pulmonary vasoconstriction may worsen hypoxia because of increased shunt and ventilation-to-perfusion mismatch. Although there are several techniques that may quantify regional perfusion using computed tomography,¹³⁷  few are applied clinically. Most clinical measurements of computed tomography perfusion are limited to detecting large defects (e.g., pulmonary emboli) or substantial enhancement (e.g., malignancies).¹³⁷ 

Regional perfusion may be determined by a sequence of cardiac-gated images acquired during intravenous infusion of an iodinated contrast agent.^138,139  After lung segmentation, computed tomography images of the parenchyma can be partitioned into tissue and blood components. Using such a technique, Dakin et al.¹³⁸  demonstrated that during ARDS, a greater proportion of blood flow is directed toward less aerated regions of the lung, and the amount of blood flow to consolidated lung is correlated with the severity of hypoxemia. It is uncertain, however, whether such physiologic correlates of perfusion and hypoxemia are present in human ARDS,^140,141  or whether such measurements of blood flow in ARDS can aid the titration of mechanical ventilation.¹³⁸ 

Computed tomography densitometry provides averages for gas or tissue content in each voxel. In human scanners, the dimensions (i.e., spatial resolution) of each voxel are ≈1 mm³, and computed tomography cannot differentiate among different alveolar units within each voxel (fig. 2B). Radiation exposure and dose accumulation limit its use, but doses can be reduced to ≈1 mSv while allowing accurate analysis.¹⁴²  Furthermore, interpolation allows whole lung quantitation from a limited set of computed tomography slices.¹⁴³  Finally, although transport of critically ill patients to a scanner is problematic,¹⁴⁴  portable scanners are becoming more readily available.

Quantitative computed tomography analysis will reveal mechanisms of injury and treatment responses, but clinical assessment currently relies on subjective interpretation. Quantitative computed tomography lacks a standard reference for clinical use, and postimage processing is complex, nonuniform, and time-consuming. Rapidly evolving computational techniques are streamlining such processing, and deep learning—after model training—involves minimal time or computational resource. If supported by clinical trials, these approaches alone, or together with biomarkers, will improve management, risk stratification, and trial selection for patients with ARDS.

Positron emission tomography is a form of functional imaging that allows visualization of a physiologic or pathologic process by marking, with radioactive isotopes, one (or more) of the substances involved in its pathways.

Positron emission tomography employs atoms in which a proton is converted into a neutron (by spontaneously losing a positron, a positive β particle [β⁺]) and an electron neutrino; the chemical element changes to one with a lower atomic number and increased nuclear energetic stability. The biologic molecule containing the radionuclide is introduced into the body and concentrates according to biochemical avidity of individual tissues and cells. After positrons are emitted from the tracer, they rapidly interact with electrons belonging to the local tissue, causing an annihilation that produces two photons traveling in opposite directions (fig. 7).¹⁴⁵  The positron emission tomography scanner contains a ring of detectors surrounding the structure of interest, and the simultaneous detection of the two photons in opposite parts of the ring represents a true signal (“true coincidence”). To define the location of the emitting region, the positron emission tomography software notes the time frame of the photons’ arrival in the detector and the angle between their trajectories. The quality of the image is directly proportional to the time resolution of the detector.

A positron emission tomography scanner is a large a ring-shaped structure that contains the positron detectors, through which the patients moves as in a computed tomography scanner (they are often combined). Recently, smaller and portable devices have been developed but are limited to specific areas of the body¹⁴⁶  or for veterinarian use.¹⁴⁷  The cost of a positron emission tomography scan largely depends on the cost of the tracer and the length of the exam (can follow a metabolic pathway for several hours). For example, a 2-h lung exam using the radiolabeled glucose analog [¹⁸F]-fluoro-2-deoxy-d-glucose costs ≈€1,500 (plus personnel), and the cost-effectiveness depends on the pathology and the tracer. A ¹⁸F-fluoro-2-deoxy-d-glucose positron emission tomography exam delivers ≈14.0 mSv¹⁴⁸  and more if combined with a computed tomography scan.

Any biologic pathway can be assessed by positron emission tomography provided a positron-emitting version of a core pathway molecule can be administered. Many tracers have been used and more are being identified.^149,150  In studying lung injury, studies focusing on ventilation use inhaled or injected [¹³N]-N₂; on perfusion, use injected [¹³N]-N₂ or [¹⁵O]-H₂O, and on lung cell metabolic activity use [¹⁸F]-fluoro-2-deoxy-d-glucose. More than one tracer can be used simultaneously.

Inhaled nitrogen as [¹³N]-N₂ has been studied in experimental lung injury and interrupted at steady-state inhalation; emission images obtained during tracer washout¹⁵¹  delineated aerated lung volume and alveolar recruitment; in addition, aerated lung volume reflected closely traditional pressure–volume analysis. This approach was also used to determine regional specific lung volume change (V_T divided by the end-expiratory gas volume, but per region), which approximated to the regional specific ventilation estimated by [¹³N]-N₂ washout.¹⁵²  When it is not possible to use a combined computed tomography–positron emission tomography scanner to identify the lung border (inhaled [¹³N]-N₂ is distributed only in ventilated lung regions), and then the simultaneous injection of [¹⁵O]-H₂O (which has an intravascular distribution) will map the organ.

Intravenous injection of dissolved [¹³N]-N₂ gas (in a saline solution) has yielded key insights^153–155  and provides regional information about perfusion, ventilation, and shunt. After injection it will immediately reach a peak positron emission tomography activity in the pulmonary circulation, which during apnea lowers to a plateau as a result of redistribution in the circulation. After the plateau, the washout curve during ventilation gives information on the level of alveolar ventilation. By this technique, it has been possible to demonstrate that sustained inflation can displace perfusion from aerated regions toward nonaerated lung and temporarily increase intrapulmonary shunt.¹⁵⁶ 

The use of the glucose analog [¹⁸F]-fluoro-2-deoxy-d-glucose permits assessment of the metabolic activity of cells employing a glycolytic pathway.¹⁵⁷  As with glucose, [¹⁸F]-fluoro-2-deoxy-d-glucose is transported into cells and is phosphorylated. The [¹⁸F]-fluoro-2-deoxy-d-glucose, which cannot progress through the Krebs cycle, is trapped in the cells in which dephosphorylation activity is low. As analog of glucose, the [¹⁸F]-fluoro-2-deoxy-d-glucose tracer has been extensively used for tissues relying on glucose metabolism, such as brain and tumors.

Neutrophil activation is heightened in ARDS,¹⁵⁸  and increased polymorphonuclear neutrophils energy production and glucose consumption is prominent.¹⁵⁹  The neutrophil activity can be monitored using positron emission tomography with [¹⁸F]-fluoro-2-deoxy-d-glucose, because neutrophils are largely responsible for its uptake, although persistence of a [¹⁸F]-fluoro-2-deoxy-d-glucose signal in neutrophil-depletion suggests that other cells play a role.¹⁶⁰ 

The number of counts in the positron emission tomography signal during [¹⁸F]-fluoro-2-deoxy-d-glucose administration aggregates the total number of neutrophils and their metabolic avidity for glucose. The quantification was initially developed for solid organs or tumors, and its use in (aerated) lung tissue may be subject to flaws. The standardized uptake value measures the [¹⁸F]-fluoro-2-deoxy-d-glucose uptake in a region of interest, corrected for the injected dose and a distribution parameter.¹⁶¹  Although the standardized uptake value incorporates the [¹⁸F]-fluoro-2-deoxy-d-glucose signal from blood and tissue, lung physiopathology involves more complex kinetics including experimental data fitting within multicompartmental models and timed blood sampling while the positron emission tomography scanner acquires positron counts.¹⁶² 

Lung tissue simultaneously contains different tissue densities, and it may be necessary to discriminate between uptake per unit tissue and the effects of dense (e.g., atelectasis) tissue or regional hypoventilation; this may be mitigated by a correction for lung density^163,164  or the inclusion of a three-compartment (i.e., blood, tissue-precursor, tissue-metabolic) model^165,166  allowing separation of prephosphorylation from irreversible trapping of [¹⁸F]-fluoro-2-deoxy-d-glucose in the tissues. However, lung edema has a density that is close to that of tissue, and this further complicates such analysis.¹⁶⁷ 

Identifying the location of inflammation during lung injury makes the [¹⁸F]-fluoro-2-deoxy-d-glucose positron emission tomography an invaluable research tool with potential impact on the care of patients. For example, it is now clear that in patients with ARDS, active inflammation during mechanical ventilation (without spontaneous effort) is localized to the nondependent baby lung in short-term observations^12,168  (fig. 8A), and this is supported by longer-term experiments.¹³  These data confirmed older work demonstrating the same distribution of inflammation using localized biopsy,¹⁶⁹  further supporting low tidal volume ventilation to protect the baby lung. In contrast, positron emission tomography scan (in experimental animals) has revealed that the locus of injury during spontaneous effort appears to be in the dependent lung close to the diaphragm (fig. 8B).¹⁷⁰ 

Recently, [¹⁸F]-fluoro-2-deoxy-d-glucose positron emission tomography was used to study the relationship between a clinically relevant ventilator strategy and lung local neutrophil activity in an animal model of endotoxemia.¹⁷¹  At 24 h, [¹⁸F]-fluoro-2-deoxy-d-glucose uptake was increased to a greater extent in consolidated and moderate-high aeration regions than in normally aerated regions. Regional strain and pulmonary blood volume were both increased in high-phosphorylation areas. The approach raised not only the possibility that inflammation is induced by the mechanical ventilator stretch, but that the interaction of ventilator strain with local blood flow, by maximizing microvascular stress and the endothelial surface exposed to circulating inflammatory cells, could initiate or propagate injury. These concepts, supported also by recent animal studies,^172,173  suggest that the prevention of lung injury might involve careful management of hemodynamics, as well as ventilator management.

The principal limitations for clinical use of positron emission tomography are cost and duration of the exam and the exposure to radiation from the chosen tracer. A typical lung positron emission tomography exam cannot be performed at bedside, and transport of critically ill patients remains a barrier.

Biomedical development of positron emission tomography involves new hardware and analytic algorithms, as well as engineering of new tracers. In principle, a greater space and time resolution, together with improved scanner sensitivity, may improve image quality. Application in oncology has advanced the search for new tracers, and with this, the ability to image almost any biologic process seems likely in the near future.¹⁴⁹ 

Magnetic resonance imaging of soft tissue structures has revolutionized how physicians view the structure of the musculature, skeletal system, and brain. This technique offers superb contrast between tissue textures, as well as flexible acquisition with a variety of pulse sequences that can highlight specific targets (e.g., pathology, hemorrhage, nodules, etc.). However, proton imaging of the lung is challenging because of the low tissue density and magnetic effects at air/tissue interfaces. Additionally, because patients can hold their breath for limited time, the requirement for prolonged immobility is often impractical. In recent years, however, improvements in proton image acquisition and development of hyperpolarized magnetic resonance imaging are raising the prospects for magnetic resonance imaging in the study of lung injury, with potential impact on clinical management of ARDS.

Magnetic resonance imaging measures signals that are generated by the rotation of nuclei immersed in a strong magnetic field. To obtain images, nuclei are excited by external radiofrequency energy (illustrated in fig. 9), and the signal is then captured while the nuclei recover their original state. This is characterized by two time constants: T1 is the time constant with which the nuclei return to equilibrium, and T2 is the transverse relaxation time, required for the nuclei to go out of phase with each other. Water and inflamed tissue appear bright in T2-weighted images because they have longer T2. Magnetic resonance imaging signal strength is a function of the difference (polarization) between the number of spins aligned with the magnetic field and those aligned in the opposite direction. This fractional difference is minimal, yet signal is high in solid tissue because proton density is very high. To increase signal in the lungs, where protons are less abundant, nuclei such as helium-3 (³He), xenon-129 (¹²⁹Xe), and carbon-13 (¹³C) are hyperpolarized (i.e., aligned with the magnetic field) and delivered to the subject in gaseous or liquid form.

Several technologic advances are making magnetic resonance imaging appropriate for lung imaging.

Hyperpolarization is produced by transferring the angular momentum of a beam of circularly polarized (laser) light on to the spins of tracer nuclei (i.e., ³He or ¹²⁹Xe).^174–177  Polarization rates of 30 to 50% are commonly achieved¹⁷⁸  and maintained for the time required for imaging. The hyperpolarized nuclei are delivered in a 20 to 79% concentration with the inspired gas; thus, the inspired O₂ level is adequate for respiration. A very large signal enhancement relative to proton magnetic resonance imaging is achieved (see video in Supplemental Digital Content 4 [http://links.lww.com/ALN/B851] showing hyperpolarized gas imaging of healthy lungs during progressive inflation). In addition to measuring tracer density, radiofrequency pulses and magnetic gradients are delivered to study specific lung functions.

Magnetic resonance imaging investigation of molecules such as [¹³C]-pyruvate allows the study of metabolic flux in tissues, thanks to the ability of magnetic resonance spectroscopy to distinguish molecular transformations. Because of low natural abundance and small nuclear spin of ¹³C, the signal must be increased through processes of nuclear polarization.¹⁷⁹  The hyperpolarized molecule is then intravenously injected, and downstream metabolites (e.g., lactate and alanine from hyperpolarized pyruvate) are regionally measured.

The edema and atelectasis present in inflamed lung tissue increase spin density, and this facilitates image acquisition in the lungs. Proton magnetic resonance imaging can thus be used to visualize atelectasis¹⁸⁰  and inflammatory changes.^181,182  In addition, oxygen enhanced magnetic resonance imaging exploits the enhancement by alveolar oxygen of proton signal. These phenomena have been exploited to study lung perfusion and ventilation in preliminary human studies.^183,184 

After delivery to the alveoli, inhaled nuclei of ³He or ¹²⁹Xe are excited with pulse sequences designed to map alveolar mechanics,¹⁸⁵  partial pressures of oxygen,¹⁸⁶  or with ¹²⁹Xe, capillary blood, and tissue gas uptake.²⁸  The effects of disease and mechanical ventilation on alveolar mechanics are studied through application of diffusion-sensitizing gradients, yielding a value for the so-called apparent diffusion coefficient for each voxel. The apparent diffusion coefficient measures the restriction imposed by the alveolar walls on the diffusion of inhaled tracer nuclei (fig. 10A), and lower values indicate smaller dimensions of intraacinar airspace (i.e., the alveoli and the alveolar ducts).¹⁸⁷  Helium is well suited for this purpose because its small nucleus diffuses rapidly, allowing for more detailed characterization of the spaces.¹⁸⁸  In addition to apparent diffusion coefficient, regional ventilation can be mapped by measuring voxel-wise signal build-up during consecutive hyperpolarized breaths.¹⁸⁹ 

Hyperpolarized gas magnetic resonance imaging interrogates microscopic structures that are far smaller than the imaged voxels.¹⁸⁷  This is because the apparent diffusion coefficient signal reflects the predominant dimensions of airspaces contained in each tissue unit, but without direct visualization. The information provided is complementary to computed tomography, overcoming the limitations of spatial resolution and of tissue averaging. This was appreciated where healthy and injured lungs were ventilated under conditions of suboptimal recruitment.^16,27  Hyperpolarized gas cannot reach nonventilated alveoli, and thus no signal is obtained (fig. 10B)¹⁹⁰ ; however, atelectasis causes inspired gas to concentrate in adjacent residual ventilated airspaces (fig. 10B), augmenting the apparent diffusion coefficient signal. Indeed, after surfactant depletion, apparent diffusion coefficient values were elevated (fig. 10C).^16,191  By contrast, computed tomography typically displayed intermediate grayscale images (fig. 10B). The high values of apparent diffusion coefficient likely reflected overdistended airspaces in which ventilated and atelectatic alveoli are intermingled (fig. 10B). Thus, the data support a model whereby atelectasis is closely associated with airspace overdistension,¹⁹²  and recruitment therefore decreased apparent diffusion coefficient.¹⁶  Regional ventilation was also high in poorly recruited regions.¹⁹³  These studies suggest that in lung tissue with mixed inflation, airspaces may undergo dynamic stretch during ventilation, which could explain why mixed inflation is associated with progression of lung injury^14,134  and tissue inflammation.¹³ 

Imaging hyperpolarized [¹³C]-pyruvate allows estimation of the impact of disease³¹  and treatment³⁰  on metabolic flux in lung tissue. Quantities of lactate, pyruvate, and other metabolic byproducts are mapped as the carbon spectrum shifts with each chemical reaction. In rodent studies after acid aspiration, carbon magnetic resonance imaging showed progressive increases in tissue lactate/pyruvate ratio, which were contained by lung recruitment.³⁰  With the ability to study an array of metabolites, this methodology could be a viable alternative to positron emission tomography in the in vivo study of regional lung metabolism.

Hyperpolarized ¹²⁹Xe dissolves in tissue and therefore permits imaging of transfer across the alveolar–capillary barrier. Reflected in the spectral shift, xenon is then measured in the blood and in the interstitial space, in addition to the intraalveolar phase. This behavior permits estimates of interstitial edema and hyperemia due to inflammation.^29,194  Thus, hyperpolarized ¹²⁹Xe magnetic resonance imaging has the potential to track regional alterations of gas transfer and uptake in injured lungs. Hyperpolarized magnetic resonance imaging has advantages including shorter acquisition times and higher spatial resolution (than positron emission tomography), as well as the lack of radiation exposure; and increased rapidity of execution allows multiple acquisitions and therefore longitudinal analysis in a given imaging session.

The use of magnetic resonance imaging in critically ill patients raises safety issues because of prolonged time spent in the scanner and the need for magnetic resonance–compatible (i.e., nonferrous) monitoring and ventilation equipment.

Hyperpolarized magnetic resonance imaging is likely to become a clinical reality in the characterization of lung pathology, enhancing understanding of lung injury. In ARDS, combining modalities offers a unique opportunity to perform simultaneous, spatially correlated measurements of lung function and metabolism that are otherwise impossible in vivo.

Electrical impedance tomography is a noninvasive, bedside monitoring system that uses microelectric current to monitor the distribution of tidal ventilation; the data are usually presented at the bedside as a continuous illustration in the sagittal plane.

Electrical impedance tomography injects microcurrents (high frequency, low amplitude) using 16 or 32 electrodes placed in a transverse (sagittal) plane around the thorax to obtain a cross-sectional image of lungs as 7- to 10-cm lung slice.^22,23  Pairs of electrodes inject current while the remaining electrodes read the resultant voltages generated by current passing through the thorax; the sensed current varies according to the diameter of the chest wall and change in electrical conductivity (fig. 11A). This cycle is repeated using alternating electrodes and results in sets of raw (unprocessed) electrical impedance tomography images; the devices can produce 50 images/s (i.e., high time resolution: 0.02 s), and image reconstruction generates raw electrical impedance tomography images from the measured voltages through the electrode plane.²²  Local changes in impedance are plotted in a matrix containing 860 (from a total of 1,024; 32 × 32) pixels, and the reconstructed images represent relative impedance changes for each pixel (termed ΔZ, or change in impedance), which is compared with a reference value for Z taken at the beginning of the data acquisition.

Most impedance changes (i.e., ΔZ) in the thorax are caused by an increase or decrease in intrapulmonary gas volume (i.e., V_T), and because of this, electrical impedance tomography is an appropriate tool to map the distribution of ventilation. Regional values of ΔZ have been shown to be proportional to changes in regional tissue aeration (gas content) as measured by computed tomography images in the same cross-sectional planes,¹⁹⁵  where greater increase in volume (gas content) corresponds to higher impedance. Thus, the distribution of tidal ΔZ represents regional ventilation during each breath.

Electrical impedance tomography electrodes are imbedded in a distensible belt that is placed on the thorax, usually over the fifth to sixth intercostal space. Placement of the electrodes at more caudal levels risks encroachment of the diaphragm into the measurement plane during expiration.²²  The presence of major spinal or chest wall wounds, multiple chest tubes, nonconductive bandages, or conductive wire sutures will interfere with current transmission and distort the ΔZ; in addition, the electrical impedance tomography currents can potentially interfere with cardiac pacemaker or defibrillator function.²² 

The key advantage of electrical impedance tomography is the ability to detect real-time information regional ventilation at the bedside; such information cannot be obtained by global monitoring (e.g., airway pressure, flow monitoring, or blood gas measurement). Thus, electrical impedance tomography monitoring is important especially when lung is injured, and distribution of aeration becomes inhomogeneous.

Electrical impedance tomography images measure the relative impedance changes (ΔZ) for each pixel. This represents the regional tidal volume during each breath, and because the time resolution is high (0.02 s), it smoothly tracks the dynamic pattern of regional inflation and deflation (i.e., the spatial distribution of ventilation at the bedside; see video in Supplemental Digital Content 5, http://links.lww.com/ALN/B852).

To quantify the regional distribution of ventilation, arbitrary so-called “regions of interest,” such as quadrants or layers, are described.²²  Analysis of electrical impedance tomography images based on regions of interest is helpful to detect spatial heterogeneity. The most frequently used measurement is the ratio of ventral to dorsal ventilation. For patients with ARDS, this dimension is especially important because most patients are ventilated in the supine position. For example, clinicians may suspect dorsal atelectasis or consolidation if electrical impedance tomography indicates predominantly ventral (vs. dorsal) distribution of ventilation (fig. 11B).

A specific metric of distribution of ventilation is the center of ventilation; this is an index to quantify shifts in regional tidal ventilation along the ventrodorsal dimension. The range across which ventilation occurs is considered from 0% (all ventral) to 100% (all dorsal), such that perfectly homogeneous ventilation is represented as the bulk of the imaged ventilation at the axis midpoint (i.e., a 50% center of ventilation; fig. 11B).^196–198  Finally, the amount of lung overinflation and collapse can be estimated by electrical impedance tomography at the bedside by the number of pixel units in which compliance changed before and after passing the best pixel compliance, while PEEP is progressively lowered in decremental steps.²⁴ 

One contribution from the use of electrical impedance tomography has been the identification of a novel mechanism of effort-dependent lung injury. Electrical impedance tomography revealed that vigorous spontaneous effort draws gas from the nondependent lung (called “pendelluft”), or from the trachea and ventilator, toward the dependent lung. This causes a transient, early inspiratory local overdistension and tidal recruitment in the dependent lung during early inspiration, corresponding, in space and time, to maximal intensity of the diaphragm contraction.¹⁹⁹  This is consistent with the finding that the bulk of effort-dependent lung injury occurs in the dependent lung, the same region where strong effort causes a local overdistension and tidal recruitment.¹⁷⁰  In contrast, positive-pressure ventilation during muscle paralysis worsens lung injury in the nondependent lung (i.e., the regions that typically receive most of the V_T).^168,169  Taken together, the emerging picture is that in ventilator-induced lung injury, either from vigorous effort or only positive pressure, the injury occurs in the lung regions receiving the most stretch (or ventilation). In this sense, electrical impedance tomography has a substantial potential to identify regional vulnerability to injury and to guide important clinical choices, such as muscle relaxation to suppress diaphragm activity.²⁰⁰ 

There are several limitations of the technique. First, although temporal resolution is excellent, spatial resolution is less than with computed tomography.^22,24  Second, because electrical impedance tomography measures relative change in impedance, it cannot identify regions of abnormality in which tidal impedance changes do not occur (e.g., preexisting atelectasis, pleural effusion, large bullae).^22,23  In the same way, electrical impedance tomography cannot identify the anatomical border between lung and nonpulmonary tissues. Third, although electrical impedance tomography is a useful research tool and has immense potential to personalize ventilator strategy at bedside, there is as yet no proven outcome benefit with its use. Finally, much of the analysis is performed off-line, which may limit immediate implementation of the results.

Emerging approaches will likely have important patient relevance (e.g., detection of tidal recruitment, pneumothorax, calculation of ventilation/perfusion ratio and estimation of cardiac output).^22,23 

Lung ultrasound is a useful diagnostic tool that can be applied in real time at the bedside.^20,201  It is an accurate and reproducible technique for the diagnosis and monitoring of many pulmonary and pleural conditions seen in critically ill patients.

In ultrasound technology, piezoelectric materials generate high-frequency (MHz; i.e., millions of cycles/s) sound waves that travel through biologic tissues in a straight direction until they encounter boundaries between two tissues with different acoustic characteristics (i.e., acoustic impedance). At these boundaries, a portion of the ultrasound energy is reflected back to the transducer, while the remainder continues until another boundary of different acoustic impedance is reached or the ultrasound energy is completely absorbed by the tissues.^20,202–204 

Sonographic images are generated on two key principles: travel time and intensity of reflection. An ultrasound wave travels from the transducer to a reflector and back to the transducer to generate an image; the total elapsed time is called “time of flight” and is directly related to the distance (i.e., depth) traveled. The brightness of the generated image is proportional to the intensity of reflection that occurs at a tissue interface, with weaker reflections appearing as darker (gray) pixels and stronger reflections as white pixels. Areas that do not reflect ultrasound (i.e., no difference in acoustic impedance) appear as black.^20,202,204 

Distinctive interactions of ultrasound waves with tissues, physical properties of the ultrasound beam, and specific image reconstruction algorithms may generate erroneous images called artifacts, which impact image quality and interpretation.^202,204,205  Most ultrasound modalities, as echocardiography, vascular, or abdominal ultrasound, aim to avoid such artifacts. However, in lung ultrasound, the distinctive characteristics of aerated lung tissue produce artifacts that provide useful information; thus, lung ultrasound requires understanding and systematic analysis of both artifactual and anatomical images.^20,26 

In normally aerated lungs, ultrasound waves are almost completely reflected at the interface between the pleura and the aerated lung, generating a hyperechoic (i.e., bright) horizontal stripe, called the “pleural line” (fig. 12A). Deep to this, multiple regularly spaced reverberation artifacts are seen, called “A lines,” and focal lung densities at the level of the pleural line (i.e., interlobular septa, microatelectasis) are also seen as short bright vertical artifacts (formerly called “Z lines”).^206,207  Finally, lung movements from breathing or transmission of cardiac contractions result in “lung sliding” and “lung pulse,” respectively.²⁶ 

Loss of aerated lung attributable to increased tissue content (e.g., edema, consolidation) or atelectasis impacts transmission in specific patterns.^20,26,207  Partially deaerated lung is heterogeneous, and this results in penetration through (more dense) or reflection from (less dense) lung areas.²⁰⁸  The resulting hyperechoic “B lines” arise from the pleural line, extend through the screen without fading, and move in conjunction with tidal lung movements.^26,209,210  The severity of aeration loss is thus assessed according to consensus recommendations.²⁶  Aeration is normal if only A lines or fewer than three B lines are present. Three or more B lines in an intercostal space represent a region of decreased lung aeration (interstitial syndrome): B1 pattern is characterized by spaced B lines and denotes moderate loss of lung aeration, and B2 pattern demonstrates coalescent B lines and is seen in severe loss of lung aeration (fig. 12B). The term “consolidation” denotes absence of alveolar air associated with ultrasound propagation and within-lung reflection, generating anatomical tissue-like images (fig. 12C).^20,26,211  See also the video in Supplemental Digital Content 6 (http://links.lww.com/ALN/B853).

The most important component of an ultrasound system is the transducer; this contains the piezoelectric material converting electrical to mechanical (ultrasound) energy and vice versa. The type of transducer impacts image interpretation²¹²  and is chosen depending on the lung region and the clinical question (table 1).²¹³  Two main features characterize different transducers: shape of the footprint and frequency of ultrasound waves. Larger footprints (e.g., curvilinear) allow for broader scanning areas, whereas smaller footprints (e.g., microconvex, cardiac phased-array transducers) enable transducer manipulation in small anatomic areas (e.g., intercostal spaces). For lung ultrasound, either curvilinear, microconvex, or phased-array transducers may be used.

The choice of adequate ultrasound frequency is more important than the footprint. Piezoelectric material generates ultrasound waves in the MHz range, and the frequency affects ultrasound penetration and axial discrimination (i.e., resolution). High-frequency transducers allow higher axial resolution and therefore better quality images, but they are limited by low tissue penetration of ultrasound waves (and the opposite holds for low frequency transducers). Therefore, in choosing an ultrasound transducer for lung ultrasound examination, depth of structures and level of detail required are the most important considerations.²¹⁴  For example, the pleural line is best visualized with a high-frequency (10 MHz) linear probe, whereas assessment of interstitial syndrome is better with lower frequency (1 to 5 MHz; table 1).

Lung ultrasound can have substantial diagnostic impact in acute respiratory failure, with substantial capacity to reclassify lesions and change management²¹⁵ ; the sensitivity and specificity for various conditions is illustrated (table 2).^{213,216–226}  Detection of real-time changes of lung ultrasound patterns and their correlation with different lung aeration conditions allows for bedside monitoring of injured lungs, and daily lung ultrasound can reduce utilization of chest radiograph and computed tomography in the critically ill.²²⁷  Lung ultrasound predicted the distribution of lung aeration measured by computed tomography in ARDS.²²⁸  It can monitor lung reaeration after management changes such as treatment of pneumonia,⁶⁴  mechanical ventilation,²²⁹  prone positioning,^230–232  recruitment maneuvers or changes in PEEP,^{180,233–235}  or reaeration during extracorporeal membrane oxygenation²³⁶  and can identify tidal recruitment.²³⁷  Indeed, aeration changes during the first hour of prone positioning may be a good predictor of successful response,²³⁸  and formal assessment of lung recruitment yields similar results as with pressure–volume curves.²²⁹  In addition, lung ultrasound can detect loss of aeration despite passing a spontaneous breathing trial, conferring a substantial risk of respiratory failure after extubation.^239,240 

B lines are a sensitive marker of injury, appearing early with loss of lung aeration; in fact, they appear before gas exchange deterioration in oleic-acid injury and correlate with worsening of lung compliance.²⁴¹  They are an early and sensitive finding in lung contusion,²²⁴  infection,^242,243  and inflammation or fibrosis.²⁴⁴  Lung ultrasound examination of patients undergoing whole lung lavage for alveolar proteinosis demonstrated real-time increasing numbers of B lines followed by development of consolidation; after lavage, the alveolar and interstitial syndromes resolved, reflecting reaeration.²⁴⁵  An analogous report documented real-time resolution of B lines during hemodialysis.¹⁶⁹  The number of B lines correlates with the amount of lost aeration²⁴⁶  and with extravascular lung water, where the interstitial syndrome is caused by fluid accumulation.^{210, 247–249} 

Several methods have been used to quantify the severity of alveolar–interstitial syndrome with lung ultrasound.^250–253  In critically ill patients, a semiquantitative score reflecting loss of aeration (normally aerated lung: 0 point; moderate loss – B1: 1 point; severe loss – B2: 2 points; lung consolidation: 3 points) is commonly used.^229,239,240 

Detection of diffuse bilateral interstitial syndrome suggests the diagnosis of heart failure and performs well compared with chest radiograph and natriuretic peptides.^253–257  In addition, severe loss of lung aeration (i.e., alveolar syndrome) indicates consolidation (e.g., pneumonia) or atelectasis (table 2). Finally, the presence (and quantity) of interstitial syndrome may be prognostic in heart failure and in end-stage renal disease.^258,259 

Integrated approaches that utilize lung ultrasound (in the setting of multiorgan assessment) can increase the diagnostic yield for diagnosis of pulmonary embolism,²¹⁶  which may be useful when computed tomography is contraindicated, and may clarify diagnosis in acute hypoxemic respiratory failure.²⁶⁰  Addition of lung ultrasound findings to a modified version of the ARDS criteria suggested that it is feasible in resource-limited settings,²⁶¹  although this is context-sensitive.²⁶²  Finally, lung ultrasound most accurately detects pleural conditions such as pneumothorax and pleural effusion (table 2).^213,226 

Although lung ultrasound is highly sensitive for several conditions (table 2), in general the specificity is low, and this may be overcome by appreciation of advanced characteristics²⁶³  and by integration with the clinical context²⁵⁶  and other imaging (e.g., radiography, echocardiography).²⁶⁴  Operator dependency is a concern, especially as the technique becomes more commonly used; although this may increase standards, it may also lead to inconsistent use. Adequate training^265,266  and use of standardized imaging protocols^25,267  will limit diagnostic error. It seems that 25 supervised lung ultrasound examinations may represent an adequate number to achieve minimal competence.²⁶⁵  Although readily repeatable, lung ultrasound cannot be used continuously, nor can it be automated; this limits its use as a monitoring tool. Finally, although lung ultrasound accurately detects and quantifies lung recruitment, it cannot identify overdistension.

Increased portability and lower cost have facilitated the use of lung ultrasound in nontraditional scenarios, such as in prehospital and resource-limited settings. Lung ultrasound is already in use in underresourced settings,²⁶¹  where conventional diagnostics are less available; this may promote a standardized approach to assessment of lung injury around the world. Development of teleultrasound will augment expansion of lung ultrasound,²⁶⁸  and newer approaches to quantitation and automation²⁶⁹  will likely increase use and acceptability. In ARDS, lung ultrasound could help categorize lung morphology at bedside and to assign treatment based on imaging phenotypes.¹²² 

The major challenges to lung ultrasound may be the requirement for clinicians to learn the necessary additional skills, and this may be especially the case among clinicians who graduated and trained in the preultrasound era. In addition, the capital equipment and transducers are still relatively expensive (total acquisition costs for versatile equipment may be approximately $30,000 to 50,000), meaning that its availability may be limited. However, more affordable hand-held devices ($8,000 to 10,000) are now available and may further facilitate widespread use of this technology. In addition, some scenarios (e.g., obesity, surgical dressing) are not amenable to accurate imaging, and unavailable windows (e.g., behind the sternum or scapulae) represent significant limitations. Ultrasound-generated tissue damage is possible as a result of thermal and mechanical energy transfer,²⁷⁰  and lung hemorrhage has been reported in animal models. However, these effects are likely of limited clinical relevance within mandated limits of exposure and energy.²⁷⁰ 

Lung ultrasound may evolve into a powerful therapeutic tool. Commercially available microbubbles, which are routinely used as intravascular and tissue contrast for echocardiographic studies, undergo cavitation when exposed to higher energy ultrasound. This causes transient formation of pores and enhanced endocytosis through biologic membranes and can induce local uptake of drugs or genes.^271,272  In a murine experimental model of Gram-negative pneumonia, lung ultrasound with microbubbles enhanced gentamicin delivery to injured lungs and increased bacterial killing.²⁷³  This application of lung ultrasound could provide new therapeutic options for treatment of injured lungs. Finally, increased automation and sophistication of signal analysis may enable lung ultrasound to be eventually for continuous monitoring.

Further research will reveal how conventional and advanced imaging techniques can be integrated in the care of patients with ARDS. Plain radiography, although nonspecific for lung injury, identifies patients with hypoxemic respiratory failure who are at risk to be further damaged by mechanical ventilation and is useful as a screening tool. Diagnostic computed tomography may be confirmatory, clarifies the distribution of injury, and may rule out alternative diagnoses. Lung ultrasound is increasingly used in bedside diagnosis; it is invaluable for guiding interventional procedures. In addition, both electrical impedance tomography and lung ultrasound are promising in the determination of ventilation patterns and responses to respiratory maneuvers.

Positron emission tomography and magnetic resonance imaging have improved our understanding of pulmonary function and biology. Although they cannot yet be considered routine in the management of patients with ARDS, ongoing technologic improvements will increase their accessibility and appeal.

Quantitative computed tomography also shows great promise. As image processing techniques become more rapid and sophisticated, it is conceivable that computed tomography characterization of potential for lung recruitment, strain, and micro heterogeneity of inflation could yield viable targets for individualized treatment. This is an important addition because clinical criteria do not anticipate the clinical trajectory and treatment response of an individual patient. With careful transport, dose-reducing protocols, and portable scanners, future studies will establish whether or not the risk and cost of quantitative computed tomography will be compensated for by the ability to obtain substantial information and improve management.

This work was supported in part by National Institutes of Health grant Nos. R01HL137389 (to Dr. Cereda), RO1HL139066 (to Dr. Rizi, Dr. Cereda, and Mr. Xin), R01-HL112986, and R01-HL126838 (to Dr. Kaczka); the University of Iowa, Department of Anesthesia (to Dr. Kaczka and Dr. Herrmann); the Office of the Assistant Secretary of Defense for Health Affairs through the Peer Reviewed Medical Research Program (award No. W81XWH-16-1-0434; to Dr. Kaczka); and the Dr. Geoffrey Barker Chair in Critical Care Research (to Dr. Kavanagh).

Dr. Kaczka and Dr. Herrmann are coinventors on a pending patent involving multifrequency oscillatory ventilation. In addition, they are cofounders and shareholders of OscillaVent, Inc. Dr. Kavanagh and Dr. Yoshida have a patent pending on a device for mechanical ventilation.