Origins of and lessons from quantitative functional X-ray computed tomography of the lung

Abstract

Functional CT of the lung has emerged from quantitative CT (qCT). Structural details extracted at multiple lung volumes offer indices of function. Additionally, single volumetric images, if acquired at standardized lung volumes and body posture, can be used to model function by employing such engineering techniques as computational fluid dynamics. With the emergence of multispectral CT imaging including dual energy from energy integrating CT scanners and multienergy binning using the newly released photon counting CT technology, function is tagged via use of contrast agents. Lung disease phenotypes have previously been lumped together by the limitations of spirometry and plethysmography. QCT and its functional embodiment have been imbedded into studies seeking to characterize chronic obstructive pulmonary disease, severe asthma, interstitial lung disease and more. Reductions in radiation dose by an order of magnitude or more have been achieved. At the same time, we have seen significant increases in spatial and density resolution along with methodologic validations of extracted metrics. Together, these have allowed attention to turn towards more mild forms of disease and younger populations. In early applications, clinical CT offered anatomic details of the lung. Functional CT offers regional measures of lung mechanics, the assessment of functional small airways disease, as well as regional ventilation-perfusion matching (V/Q) and more. This paper will focus on the use of quantitative/functional CT for the non-invasive exploration of dynamic three-dimensional functioning of the breathing lung and beating heart within the unique negative pressure intrathoracic environment of the closed chest.

Introduction

Quantitative CT (qCT) of the lung has evolved to provide both structural and functional disease phenotypes and subphenotypes previously lumped together (or missed) by the limitations of spirometry and plethysmography. QCT is imbedded into studies seeking to characterize chronic obstructive pulmonary disease (COPD), ^1–4 severe asthma, ⁵ interstitial lung disease ⁶ and more. Reductions in radiation dose by an order of magnitude or more ^7–9 have been achieved. At the same time. we have seen significant increases in spatial and density resolution along with methodologic validations of extracted metrics. Together, these allow attention to turn towards more mild forms of disease and younger populations. In early applications, clinical CT offered anatomic details of the lung. However, linkage of multiple static lung volumes along with dynamic imaging and dual energy CT (DECT)-based extraction of regional perfusion and ventilation ^10,11 are offering regional measures of lung mechanics, ^12,13 the assessment of functional small airways disease (fSAD), ^14–17 as well as regional ventilation-perfusion matching (V/Q). ¹⁸ Advances in micro-CT (µCT) to study pulmonary micro-structure while maintaining ex-vivo lung architecture provides the ability to reconstruct the three-dimensional anatomy of the lung at the acinar level. ^19–23 These advances offer insights into risk factors associated with specific pathologies. Through the tracking of regional structural and functional changes, disease models of increasing complexity are emerging. ^24–27 This paper will provide a brief look at the past, present and future of qCT-based assessment of the cardiopulmonary system designed to gain an understanding of lung function assessed both directly and through modeling. The focus is on the non-invasive exploration of dynamic three-dimensional functioning of the cardiopulmonary system within the unique negative intrathoracic pressure environment of the closed chest.

The scanners

Early CT

Ahead of transmission-based CT, Kuhl in 1963, at the University of Pennsylvania, had already developed the early methodologies for single photon emission CT to explore gravitational effects on pulmonary perfusion distribution. Participants self-injected a technetium compound while riding the gondola of a human centrifuge and were then transferred to a scanner to reconstruct the distribution of the lodged technetium labeled macroaggregated albumin (⁹⁹Tc-MAA) as an index of regional pulmonary perfusion. ²⁸ In that same year, Rogers and colleagues ²⁹ used ⁹⁹Tc-MAA to assess regional pulmonary function prior to lung cancer resections. Meanwhile, Godfrey Hounsfield, at the Electric and Musical Industries (EMI) laboratory, sought to reconstruct a slice of the brain via transmission imaging. The first patient-based brain reconstruction was in October of 1971 and a full report was published in 1973. ³⁰ The first body CT scanner, was sited by EMI in 1975. Sheedy and colleagues ³¹ reported early experiences with suspected masses being the primary scan motivation.

The Dynamic Spatial Reconstructor

In a similar time frame, scientists within the Physiology Department at the Mayo Clinic were seeking to understand cardiac mechanics and myocardial perfusion along with distribution of pulmonary ventilation and perfusion. The team, led by Dr Earl H. Wood and later by Dr Erik L. Ritman, sought to build a scanner [the Dynamic Spatial Reconstructor (DSR)] which would capture cardiopulmonary anatomy and physiology with a targeted spatial resolution of 0.5–1.0 mm and a temporal resolution of 0.01 s. ^32–35 The DSR prototype consisted of a horizontally mounted X-ray source and image intensifier combination within which upright animal models and isolated organs were rotated. The first volumetric heart reconstruction was achieved in 1973. In recognition that volumes, not individual slices, were being generated, the primary picture element was dubbed a “voxel.” Thus, the Mayo team began the task of extracting meaning from the X-ray attenuation values of a temporally sequenced volume of voxels.

The final implementation (depicted in Figure 1) consisted of a continuously rotating, 20 ton gantry spinning at 15 rotations per minute. 14 X-ray guns on one side were aimed at a juxtaposed hemicylindrical fluorescent screen recorded by 14 analog video cameras. Details are provided in early publications. ^32–35 Analysis tools evolved to asses three-dimensional aspects of cardiac chamber anatomy along with the myocardium, lungs, airways, vascular trees, diaphragm surface, rib cage structure and much more. ^37–39 It was the DSR project which set the stage for the evolution of qCT of the heart and lungs.

Figure 1.

Mayo Clinic’s Dynamic Spatial Reconstructor. This nearly two storey high, 20 ton scanner consisted of 14 X-ray tubes (A) juxtaposed by 14 television cameras (B) aimed at a continuous hemicylindrical fluorescent screen and controlled from position (E). Fitted with a high voltage slip-ring assembly (C) the cantilevered scanner rotated over an open pit at 15 rotations per minute. With American television consisting of 240 usable television lines, every 1/60 of a second projection images were recorded which allowed for the reconstruction of a stack of 240 cross-sectional images. Through continuous rotation and the gathering of each set of projection images from 14 imaging chains onto a bank of analog video disc recorders, dynamic images of the heart and lungs were achieved, providing functional volumetric images of the heart and lungs. This one of a kind scanner came online in 1979 following approximately 5 years of development. Modified from Kaufmann et al. ³⁶

Electron Beam CT

Motivated by dynamic imaging, Dr Douglas Boyd and team developed Electron Beam CT (EBCT), ⁴⁰ introduced by Imatron in 1985. Unlike the DSR, EBCT ‘s X-ray source and detectors remained stationary. As Figure 2 depicts, X-ray was generated by electromagnetically sweeping the electron beam along each of four pairs of semi-circular tungsten target rings covering 210° of circumference about the patient. One sweep of the tungsten target occurred in 50 ms with an 8 ms reset time. For the high temporal resolution mode, with four tungsten targets, eight slices (8 mm thick each) of the heart (or lungs) were capture in 224 ms spanning the z-axis extent of the cardiac ventricles. For perfusion, imaging was gated relative to each QRS complex of the electrocardiogram over many cardiac cycles to capture the inflow and outflow of contrast agent. For any pair of slices, timing was always at the same point of each cardiac cycle. For the four pairs of slices, the time within the cardiac cycle was shifted. Via this approach, Wolfkiel et al ⁴² established the validity of regional myocardial perfusion. There were good correlations under conditions of modest coronary flow rates. However, under high flow rate conditions, EBCT underestimated myocardial perfusion which was addressed by Lerman et al. ⁴³ On the pulmonary side, Stern et al ⁴⁴ scanned patients with COPD as they performed forced expiratory maneuvers. A slab of the lungs was captured with the EBCT set to acquire 8 slices every 224 ms. The dynamic image sets demonstrated regions of air trapping and, in some cases, regions of paradoxical expansion during expiration. It is recognized that, during this maneuver, the lungs are moving through the z-axis.

Figure 2.

Imatron’s EBCT scanner. It went through several iterations: C-100; C-150 and C-300. The final version was marketed by General Electric before being abandoned as spiral CT capabilities advanced. Modified from Webster. ⁴¹ EBCT, Electron beam CT.

One contribution of EBCT was the development of a coronary calcium score for use as a marker of coronary artery disease ⁴⁵ and served as an integral part of the multiethnic study on atherosclerosis (MESA) ^46,47 seeking to understand population-based risk factors for coronary artery disease. These scans have fortuitously provided a rich pulmonary image data set, allowing for the exploration of risk factors leading to COPD. ⁴⁸ As with the DSR, a key role of EBCT (no longer in production) was to provide a tool by which the methodologies and utility of dynamic volumetric CT could be explored and established.

High resolution CT (HRCT)

With an interest in evaluating interstitial lung disease and to assess texture of lung nodules, it was observed that parenchymal structure was better depicted by thin (1.5 mm) slices sharpened via use of high frequency reconstruction kernels ^49–51 and optimizing the acquisition protocols. ⁵² A visual score reflecting the burden of emphysema evolved using just three slices. It was shown that, in patients with normal pulmonary function tests, the HRCT-derived emphysema scores were strongly correlated with the patient’s lung diffusing capacity (DLco). ⁵³

Spiral CT

The lessons from the DSR and EBCT scanners slowly gained ground as the limitations of single slice, step and shoot became clear. By simultaneously preserving the slice-to-slice relationships and the advantages of thin slices to avoid partial volume effects, Kalender and colleagues introduced the concept of spiral CT in 1990. ⁵⁴ Spiral scanning with a single row of detectors could take upwards of 40 s, too long for a heartbeat or even a breath-hold. To solve this problem, multidetector-row CT (MDCT) was introduced in 1998 with a rapid expansion in the number of detector rows across all manufacturers. This has brought imaging the full lung down to a few seconds or less.

DSCT/DECT

With a move towards emphasizing functional measurements along with increased speed of scanning (for cardiac imaging), multispectral imaging was introduced with GE using rapid kV switching and a single X-ray tube and Philips using a sandwich detector (capturing lower energy photons and then high energy photons). As depicted in Figure 3, the early dual energy scanner by Siemens consisted of two X-ray tubes to provide simultaneous high and low kV images for a dual energy mode while offering a dual source mode to achieve the scanning speeds needed for cardiac imaging. With a pair of images obtained at two different kV settings, there is a reconstructed attenuation offset between certain materials such as calcium, iodine, or xenon gas. Using material decomposition methods ³⁴ one can extract, from the image, iodine in the pulmonary vasculature to assess regional perfused blood volume as an index of perfusion, ¹⁰ xenon in the lung air space as an index of regional ventilation, ¹¹ or calcium in the contrast enhanced coronary vasculature to allow for lumen characterization. ⁵⁶ In 2006, Flohr et al ⁵⁷ published the first performance evaluation of a dual source (dual energy) CT scanner targeted largely at the heart since it’s second detector ring had a maximum diameter field of view (dFOV) of 26 cm. Following a series of incremental advances, in December 2013, Siemens introduced the SOMATOM FORCE which incorporated 96 detector rows (capable of generating 192 slices) associated with each X-ray gun covering 52.5 mm along the z-axis. The VECTRON tube expanded the kVp range to 70–150 kVp with a 20–1,300 mA range. A more powerful X-ray tube allowed for more photons when using a tin filter to narrow the energy spectrum, allowing for improved material separation. In addition, by having two X-ray tubes, mAs is adjusted separately to achieve similar noise characteristics between the two energies. In 2013, Philips Healthcare announced their IQon Spectral CT system with the sandwich detectors obtaining a CE Mark in 2015. GE’s rapid kV switching approach to DECT was implemented on their Revolution CT. Expansion of z-axis coverage has continued with Toshiba’s (now Cannon) Aquillion One consisting of 320 detector rows covering 16 cm along the z-axis and rapid kV switching has recently been added. GE’s Revolution CT supports 256 detector rows capable of generating 512 slices spanning 16 cm along the z-axis. To support lower radiation dose, ⁷ all manufacturers have implemented various versions of iterative reconstruction and most recently, deep learning algorithms have been introduced to improve the reconstructed images further with additional noise reduction. ^58–61

Figure 3.

MDCT/DSCT/DECT. The graphic inset in the center of the image depicts a dual source configuration whereby the kV of each X-ray gun can be set separately at, for instance, 140 and 80 kV to allow for material decomposition. Speed provided by the use of two X-ray guns eliminates motion blurring in the single energy mode of scanning, as represented by detailed artery/vein anatomy shown for the right lung (A) and material decomposition is provided through dual energy reconstruction pairs. (here, extracting iodine from parenchymal regions yields regional perfused blood volume (heated object scale revealed in the right lower lobe cutaway). (B) Modified from Iyer et al ⁵⁵ with permission of the American Thoracic Society. Copyright ^© 2022 American Thoracic Society. All rights reserved. DECT, dual energy CT; DSCT, dual source CT; MDCT, multidetector row CT.

Vertical CT

It is well recognized that gravitational effects alter the shape of the chest wall and diaphragm, the intrathoracic position of the heart, the distribution of ventilation and perfusion, the mechanical strain on the lung parenchyma and much more. Yet, our CT imaging is accomplished in the recumbent body postures, mostly supine. Thus, we seek to understand the links between form, function, disease distribution and disease susceptibility with our window into the thorax failing to represent the body posture in which our cardiopulmonary system undergoes the greatest challenges: upright or sitting. To address this, Cannon has introduced a one-of-a-kind vertical CT scanner ⁶² by reorienting their 320 detector-row Aquilion One CT scanner 90° as depicted in Figure 4. The vertical scanner provides Isotropic 0.5 mm voxels, a 0.275 s/rotation speed and a maximum vertical translation of 100 mm/s. Similar to the standard, clinical Aquiion ONE, the 320 detector-rows cover 16 cm of the z-axis (cephalon-caudal). In the spiral/helical scan mode, 80 detector-rows are used to avoid issues with wide cone beam geometry. Early reports demonstrate a significant difference in lung volumes and a redistribution of lobar volumes comparing supine to sitting and upright postures. ^63,64

Figure 4.

Vertical CT scanner by Cannon. On the left, the scanning hardware is located at the top of the gantry. On the right, the scanning hardware has traveled to the bottom of the gantry. In the center of the figure, the person being scanned is depicted in the standing position. Modified from Jinzaki et al. ⁶²

Photon counting CT

The scanners discussed above, all utilize detector technologies relying on the integration of photon energy to produce light which is then converted to a digital signal. The conversion introduces inherent electronic noise into the system. In addition, the integration of photons loses information regarding the relative attenuation of photons across the energy spectrum, thus sacrificing information which can be used in multispectral-derived material decomposition. Having to rely on the integrated attenuation across a broad energy spectrum leads to further image artifacts including beam hardening. ⁶⁵ A photon counting CT scanner provides the capability of counting each photon and provides the ability to bin the photons, post-scanning, according to desired energy ranges. The first clinical dual source photon counting scanners ⁶⁶ have shipped (Siemens Naeotom Alpha). By binning of photons according to their energies, one can reconstruct images from a narrower energy range, improving material separation and largely eliminating beam hardening artifacts. This new generation of scanner improves spatial resolution, further reduces radiation exposures and allows for multispectral imaging with every scan, significantly expanding the possibilities for functional imaging of the lungs.

µCT

By using small animal models and isolated tissue samples, µCT has opened the doors to the non-destructive exploration of lung micro-structure in relation to genetic modifications. Using respirator control of the anesthetized mouse, Namatti et al ⁶⁷ have been able to capture images of the in-vivo mouse lung at multiple levels of inflation with a voxel size of approximately 20 microns. Using interior tomography methods with optical magnification, the Zeiss Xradia Versa 520 scanner has been used to achieve voxel sizes on the order of 1–2 microns (Figure 5) allowing for the reconstruction of alveolar level anatomy and the interrogation of ductal branching within the acinus. ^19–22 To animate the acinar anatomy, a high-resolution image of the mouse acinus is aligned with the full inspiratory in-vivo lung reconstruction of the same mouse. Then, the warping function, linking the multiple in-vivo images (obtained at multiple airway pressures) to each other can be used to animate the ex-vivo derived acinus. This then allows, or instance, for more informed estimations of particle deposition. ⁶⁸ Similarly, µCT has provided details of destruction associated with various forms of COPD, ^69–71 supporting the notion that disappearance of the peripheral airways precede alveolar destruction. ⁷² Using an ultrabright beam line at the European Synchrotron Radiation Facility and hierarchical phase-contrast tomography (HiP-CT) to achieve 1 micron voxel level imaging within the intact lobe of a human lung, Walsh and colleagues have begun to explore the vascular and small airways pathologies associated with SARS-CoV-2. ²³ Radiation dose needed is much higher for this level of spatial resolution. Thus, µCT is limited to isolated organ specimens or small animals.

Figure 5.

Adjacent acini (yellow and green), terminal bronchus (gray) and supplying vasculature (red and blue). Imaged via Zeiss Xradia Versa 520 micro-CT using interior tomography of a mouse lung. Modified from Vasilescu et al. ²⁰

Image evaluation

Early qCT studies

Studies associated with the DSR project served to set the stage for a non-invasive assessment of pulmonary structure-to-function relationships. As a first step, it was demonstrated that volumetric CT was able to extract accurate measures of lung shape and volume. Anesthetized dogs were imaged, exsanguinated and the air from the lungs were subsequently captured via a Super Syringe (Hamilton, Reno, NV). The lungs were then water displaced and all measures were corrected for pressure and temperature. Additionally, known volumes of air were introduced in vivo, and volume change was assessed via imaging. An agreement to within 3% of the gold-standard was demonstrated (Figure 6). ⁷³

Figure 6.

Demonstration of lung volume measurement accuracy via volumetric CT (DSR). Left panel compares the lung volume via the DSR on the y-axis with lung volume assessed via excision and water displacement. Right panel compares the lung volume change via the DSR vs known inflation steps via a Super Syringe. With permission from Hoffman et al. ⁷³ DSR, Dynamic Spatial Reconstructor.

A next step was to calibrate the X-ray attenuation coefficients for each voxel using regional sampling of pure air or pure “water” (blood). (Figure 7) With the assumption that lungs are composed of only air or “water,” voxel were converted to a percent air content. This was then used to assess the vertical air content gradients in the supine and prone body posture of anesthetized dogs with the lungs held at multiple lung volumes. The steep gradient in air content at functional residual capacity and air content change (ventilation) with increased airway pressure in the supine posture along with the more uniform inflation along the gravitational axis in the prone posture were demonstrated. ^74,75 (Figure 8) It is well recognized today that by placing a patient in the prone posture, there is less compression of the dependent lung, and thus less positive pressure needed to ventilate the lung while keeping the dependent lung open. The first human study via the DSR was of a young girl with pulmonary atresia. It was of interest to assess how much lung perfusion came from the right vs left heart. This study demonstrated the ability of dynamic volumetric CT imaging to follow a sharp bolus of contrast agent through the pulmonary circulation on a regional basis for quantitation of pulmonary parenchymal perfusion. ⁷⁶ An additional example of the power of dynamic volumetric CT came from a simple question as to whether or not the total volume of the heart remains constant through the cardiac cycle. With the DSR, it was demonstrated that indeed the total heart volume (contents of the pericardial sac) remains essentially constant (relative to the combined stroke volumes of the left and right ventricles) throughout the cardiac cycle under normal conditions. ⁷⁷

Figure 7.

With the assumption that the lung has only two densities (air and water) and by sampling pure air within the trachea and pure “water” in the heart chamber, one can then convert each voxel to a percent water and percent air value. By tracking differences in and changes in regional air content, one begins to understand the regional distribution of ventilation and lung mechanics. With permission from Hoffman et al. ⁷⁴

Figure 8.

Vertical lung air gradients and lung expansion. Supine (A) and Prone (B) anesthetized dog. With permission from Hoffman et al. ⁷⁴

Lung segmentation

A first challenge in evaluating the lung via qCT is to accurately identify the lungs and lobes even in the presence of fibrosis, emphysema and consolidation. The early work by Hu and colleagues ⁷⁸ have more recently been significantly improved upon through the use of deep learning algorithms to find the lungs ⁷⁹ and fissures. ⁸⁰ Accurate identification of fissure completeness has proven to be important in determining the suitability of a candidate for endobronchial valve-based lung volume reduction as an intervention for severe COPD. With the goal being to reduce the volume of a lobe, communication across lobes serves to limit volume reduction.

Lung density

A key set of metrics used in qCT is based upon the scaled Hounsfield Unit (HU: scaled X-ray attenuation coefficient) of a voxel, whereby air should be −1000 HU and water 0 HU. To provide objectivity to emphysema quantitation, Müller and colleagues ^81,82 introduced the concept of the lung density mask. Within the lung field, voxels falling at or below −950 HU (density mask) were empirically defined as being emphysema-like. The value of −950 was confirmed to be the optimal threshold through comparisons with postmortem lung samples. ⁸³ The mask for air trapping on expiration, ⁸⁴ voxels < −856HU, is set at a value found to be the lower density of normal lung at total lung capacity (TLC). Both functional residual capacity (FRC) and residual volume (RV) have been used as the expiratory volume. However, RV is in agreement with the volume used in a pulmonary function laboratory to assess air trapping. Coxson and colleagues ⁸⁵ provided a transformation taking HUs to a measure of the surface area for gas exchange. Because the transformation is linear, HU and the surface area index are readily interchangeable. However, this gave the lung community a physiologic-based measure they were better able to understand and interpret. Accurately scaled density allows for assessment of changes in regional air content and thus an index of regional ventilation. This implies that it is critical for HUs to accurately represent the continuum of densities between air and water.

Functional small airways disease and regional lung mechanics

Because the lung is often imaged at two lung volumes, TLC and FRC or TLC and RV, image matching methods have served to link lung regions between the two volumes and to use the warping field to generate regional measures related to lung mechanics. ^12,13,86,87 The warping function provides a set of regional “Jacobians” which represents a regional unit volume change, indexing regional ventilation. ^88–90 With a hypothesis that small airways disease is the predecessor to emphysema, Galban and colleagues ¹⁴ developed the Parametric Response Map (PRM). After matching the inspiratory to expiratory images, all voxels <-856 HU but linked to an inspiratory emphysema voxel (<-950 HU) are eliminated from the pool of air trapped voxels (dubbed PRM-functional small airways disease or PRM-fSAD). By tracking the PRM-fSAD voxels longitudinally, one can determine if PRM-fSAD voxels preferentially emerge as emphysema. An alternative to the PRM approach is the Ventilation Defect Probability Measure (vDPM). ¹⁶ vDPM also matches the inspiration and expiration scan. However, instead of defining fSAD by a fixed −856 threshold, vDPM evaluates the regional change in HU between inspiration and expiration. Based upon the HU change, a probability of fSAD or “vDPM-fSAD” is assigned. Along with regional fSAD assessment, the lung is (as with PRM) labeled as normal or emphysema. The emphysema can, in both PRM and vDPM, be labeled as non-emptying or emptying based upon whether or not the emphysema voxel at TLC links with an air trapped voxel on expiration. In addition to the Jacobian other regional metrics of lung mechanics are derived including the Anisotropic Deformation Index (ADI), ^12,13 which provides an index of shape change uniformity. (As a note of caution, because it is easier to match a high to a low lung volume, in both PRM and vDPM, percent emphysema has been indexed to the low rather than the high lung volume.)

Texture

While radiologists are good at recognizing presence of texture differences, it is difficult for them to assign percent of the lung involved or to label a specific small volume-of-interest. To solve this problem, Uppaluri et al. ^91–93 developed what has become known as the AMFM or adaptive multiple feature method. Texture is characterized via 25–30 mathematical formulations. An expert observer or group of expert observers label a set of regions of interest with a consensus opinion (in the case of multiple observers) regarding the appropriate descriptors (ground glass opacity, honeycomb, etc.). Using a Bayesian classifier, the AMFM identifies the small set of mathematical formulations which best separate one tissue type from another. Xu and colleagues have extended the mathematical formulations from 2D to 3D. ^94,95 Salsbury et al ⁹⁶ have demonstrated the ability of the AMFM to identify patients with interstitial pulmonary fibrosis who are most likely to rapidly progress. Others texture approaches have shown good success in characterizing fibrotic lung. ^97–100 Texture, of course, is influenced not only by structure, but also function, indexing peripheral air trapping, heterogeneous vasoconstriction, and more.

Pulmonary airways

Through early airway segmentation, Wood and colleagues ¹⁰¹ demonstrated that detectability of the airway segments were influenced both by the size of the airway, slice thickness and airway orientation to the scanning plane. Tschirren together with Palagyi ^102–104 provided methodology to extract the bronchial tree, identify the centerline and branch points, and to automatically label the bronchial segments out to the segmental bronchi. By reslicing the airway segments perpendicular to the local long axis defined by the centerline, measurements of the lumen and wall were provided. There have been numerous iterations on the process for finding the airways including more recent deep learning methodologies. ^105–110 With an interest in airway remodeling in asthma, Aysola et al ¹¹¹ demonstrated significant correlation between CT measures of airway wall area percent (wall area divided by the cross-sectional area defined by the outer wall) vs endobronchial biopsy evidence of wall remodeling. To avoid the need for extracting the full airway tree, Nakano et al ¹¹² developed a method known as Pi10. Initially, random airway segments were sampled approximately perpendicular the segmental long axes (judged by circularity) and assessed. The lumen perimeters were plotted (x-axis) vs the square root of wall areas (y-axis). From a linear fit to the data points, a point on the regression line corresponding to a 10 mm lumen perimeter was used to identify the square root of wall area for this virtual airway segment. Pi10 does not distinguish wall remodeling from other processes altering lumen area (airway/parenchymal interdependence etc.).

Utilizing CT and biopsy-based metrics, McDonough et al ⁷² observed that terminal bronchi tend to disappear prior to the disruption of the distal alveolar walls. A µCT study by Tanabe et al ⁷⁰ provided additional evidence of alterations to the terminal bronchi and bronchi immediately proximal in both centrilobular and pan-lobular emphysema but with greater wall thickening in centrilobular emphysema. Because µCT allowed for the assessment of wall volume in addition to thickness, it was observed that wall volume was reduced despite apparent thickening. The thickening was thought to be explained by the observed decrease in alveolar attachments. ⁷¹ Because the airway tree extracted from CT can be labeled consistently, Smith et al ¹¹³ observed that when the same anatomic locations are compared, individuals with COPD, on average, had thinner airway walls compared to participants without COPD and with participants who didn’t smoke. Additionally, Kirby et al found that, with a count of total airway segments, the number of segments (detectable on CT) served as a prediction for progression of COPD in at-risk smokers. ¹¹⁴ Oelsner et al ¹¹⁵ demonstrated that a higher genetic risk score (GRS) was associated with and increased risk of COPD along with qCT evidence of lower lung density, smaller bronchial lumens and fewer small bronchial segments independent of a smoking affect modification. Additionally, aberrant airway branching patterns ¹¹⁶ and airway dysanapsis ^117,118 have been shown to impart a greater risk of COPD even in individuals who never smoked. Examples of lungs with a range of airway dysanapsis ¹¹⁸ are displayed in Figure 9.

Figure 9.

Coronal CT images with their segmented central airway trees. The airway to lung volume ratio provides a measure of dysanapsis. Panels A–D represent participants in the first, 50th, 75th and 99th percentile of dysanapsis assessed in the MESA Lung population. Modified from Vameghestahbanati et al. ¹¹⁸ MESA, multiethnic study on atherosclerosis.

Computational fluid dynamics (CFD)

With the capture of the lung’s three-dimensional structure as well as the structural changes occurring with static or dynamic inflation and deflation, engineering methods associated with CFD have allowed for the subject-specific prediction of gas and particulate distributions ^25,119 in addition, one can provide regional estimates of flow resistance ¹²⁰ and surface shear. ¹²¹ Such modeling provides the basis for the generation of new hypotheses related to lung function and disease etiology. This has been reviewed in detail. ¹²²

Xenon CT

While structure has been shown to strongly infer function, there remains the need to directly provide those links. Early studies ¹²³ found that xenon and krypton gas were sufficiently radiodense as to be detectible within the lungs when inhaled. Initial studies focused on subtraction of images with and without xenon. While showing some success, ^124,125 image matching left considerable misalignment artifacts. Via EBCT, Tajik was able to acquired images at the same point in the ventilator cycle over multiple breaths during wash in of xenon gas with an anesthetized animal model. In agreement with early CT observations regarding differences in supine and prone distributions of inspired air volumes, ^74,75 xenon imaging demonstrated preferential ventilation to the dependent lung regions and increased as inspiratory flow rates increased.

With an interest in the utility of single breath, high concentration xenon, Tajik et al ¹²⁶ demonstrated that scanning should commence shortly after the onset of a breath-hold since preferential gas absorption in the dependent region occurs rapidly, complicating the assessment of ventilation. Chon et al ¹²⁷ demonstrated that wash-in and wash-out curves are not mirror images and wash-out has considerably longer time constants. Because xenon is anesthetic and can interfere with the drive to breath, concentrations used for multiple breath methods have been limited to concentrations of 40% or lower. In a second study, ¹²⁸ krypton gas (less radiodense than xenon) was shown to be useful in enhancing signal-to-noise when used in combination with Xenon allowing improved detectability of regional characteristics of ventilation heterogeneity.

Fuld et al ¹²⁹ presented a dual piston device for use in human CT imaging of xenon wash-in or wash-out kinetics, keeping inspiratory and expiratory volumes constant across a multibreath protocol. However, the anesthetic nature of xenon gas and the increased inspiratory resistance proved distracting for some participants, significantly limiting clinical utility. With the introduction of DECT, it became possible to return to the concept of imaging a ventilation index with a single scan volume. Earlier xenon studies ^124,125 were hindered by a mismatched non-contrast and contrast volumes needed to generate a xenon only image. However, with DECT it is possible to use material decomposition to extract regional xenon concentrations from a simultaneously obtained high and low kV images. Fuld and colleagues ¹¹ outlined methodology and demonstrated pitfalls of DECT xenon imaging. Using a CT-derived hollow bronchial tree cast, it was demonstrated that xenon delivery to the non-dependent portions of the tree were affected by flow rate with preferential gravity based delivery of xenon occurring at slower flow rates, likely accentuated by the monopodial airway branching pattern of the pig. The interaction of airway branching geometry and carrier gas properties in determining particle delivery to the lung has been modeled by Miyawaki et al, ¹¹⁹ and the modeling is in agreement with the observations of Fuld et al. ¹¹ Combining xenon with helium in combination with 20% oxygen was found to avoid gravitational dominance on gas flow distribution. There have been a number of studies utilizing DECT and xenon gas to assess regional ventilation associated with bronchial atresia, asthma and COPD. ^130–132 With the improved contrast resolution of DECT, there has been a renewed interest in the use of krypton as compared with xenon. Human ¹³³ and rabbit ¹³⁴ studies have demonstrated that krypton can indeed provide a signal. However, the strength of the signal is such that distinguishing regional differences in ventilation (as opposed to an all or none signal associated with more central airway obstructions) is problematic. When using a single DECT volume to capture regional differences after the lung is equilibrated with the inhaled xenon gas concentration, differences (except for regions of no ventilation) are eliminated.

Pulmonary parenchymal perfusion and vascular volumes

The earliest CT pulmonary perfusion studies ^135,136 were accomplished by injecting a sharp bolus of contrast near the right heart chambers, so that one came close to achieving an impulse function within the pulmonary trunk. Imaging was carried out via axial scanning with a breath-hold at FRC and imaging gated to the same location within the cardiac cycle over approximately 12 cycles. This allowed for a capture of both the input function from the main pulmonary artery as well as the output function for peripheral parenchymal lung regions. The method assumed that all of the contrast arrives prior to leaving a region, and it assumes contrast stays within the blood pool. This is violated, for instance, when inflammation promotes capillary leak. Despite these shortcomings, CT perfusion showed good agreement to radioactive microsphere assessments under modest flow rates. ⁴² Using this methodology in conjunction with the 50 ms scan aperture of the EBCT scanner, Won et al demonstrated ¹³⁷ the use of deconvolution/reconvolution of regional time–intensity curves to isolate microvascular time–intensity curves, providing a regional map of microvascular mean transit times (MTTs). The method demonstrated that the dependent vs non-dependent microvascular MTTs reflected the earlier observed prone-supine differences in regional air volume changes with lung inflation (index of ventilation), thus demonstrating a ventilation/perfusion match. While many advantages have come with the evolution of CT, a scan aperture of 50 ms remains lost with EBCT’s discontinuation. (Scan aperture is the time required to obtain the projection images used to produce a reconstruction) With an 8 ms beam reset time, the scan interval is 58 ms.

With DECT, it became possible to capture regional measures of pulmonary PBV. A slow (4 ml/s) infusion of contrast agent delivered via an antecubital vein prior to scanning such that contrast arrives on the left side prior to scan onset and continues throughout scanning using 2 kVs, The goal is to approximate a homogeneous dilution of iodine throughout the pulmonary circulation with the iodine concentration reflected in the concentration sampled within the main pulmonary trunk. By using material decomposition, one can extract the amount of iodine in each voxel, creating a virtual non-contrast (VNC) image. By use of standard lung segmentation software, iodine in each lung voxel is divided by the iodine concentration in pure blood sampled within the pulmonary trunk. To the extent that a voxel is not pure blood, the ratio of the voxel’s iodine content to that in pure blood provides a measure of regional percent blood volume which in turn provides a measure of regional blood volume when taking into account the voxel volume. This, of course, only works if the voxel is perfused at the time of imaging, thus the term “perfused blood volume.” Fuld et al ¹⁰ demonstrated, in an animal model, that pulmonary perfused blood volume, under normal flow conditions, is a surrogate for perfusion. As perfusion increases (up to a limit) capillaries are recruited and expand. Thus, blood volume goes up with increased perfusion.

Using methods to assess true perfusion ¹³⁸ and PBV ⁵⁵ Alford et al and Iyer et al respectively, demonstrated that normal smokers (normal pulmonary function tests) with early visual signs of centriacinar emphysema (emphysema susceptible) have double the heterogeneity (coefficient of variation) of regional perfusion or perfused blood volume compared with normal smokers without signs of emphysema susceptibility. Furthermore, the higher heterogeneity in emphysema susceptible normal smokers is eliminated by a single dose of 20 mg of sildenafil. ⁵⁵ The increased heterogeneity of PBV is accompanied by an increase in cross-sectional area of the segmental pulmonary arteries relative to the cross-sectional area of the associated segmental bronchi, suggesting increased peripheral vascular resistance. It is hypothesized that, in the general population, hypoxic pulmonary vasoconstriction (HPV) is inhibited when regional hypoxia is in association with regional inflammation. If a portion of the population has an inability to block HPV in the presence of inflammation, the cascade of events serving to eliminate the cause of the inflammation and to repair tissue damage is limited or eliminated locally possibly leading to emphysema. Others have also implicated vascular pathologies associated with COPD. Synn et al ¹³⁹ demonstrated that in the Framingham Heart Study, a decrease in the CT-derived small vessel fraction of total pulmonary vessel volume (TPVV) was correlated with a greater association with COPD. Barr and colleagues ¹⁴⁰ demonstrated, via CT, a correlation between impaired left ventricular filling and CT-derived percent emphysema down to low amounts of emphysema. This observation has suggested an endothelial dysfunction association with the regional development of emphysema. Aaron et al ¹⁴¹ found direct correlation between pulmonary vascular volumes on CT and impaired LV filling and dyspnea even in subjects without impaired lung function. The ability to separate the total pulmonary vascular volume into arteries and veins is shown in the upper panels of Figure 10, Pistenmaa et al ¹⁴² have further supported a vascular role in COPD. They show that there is a faster progression of emphysema and a greater decline in pulmonary function (FEV1/FVC) associated with a greater loss of smaller pulmonary arteries. The lower panels of Figure 10 show Professor Weibel’s ¹⁴³ cast of human airway and vascular trees for comparison with the CT-derived vascular segmentation.

Figure 10.

Upper panels: segmentation of pulmonary arteries (blue) and veins (red) from non-contrast CT. modified from Pistenmaa. ¹⁴² Lower panels: Professor Ewald Weibel’s cast of the pulmonary arteries (red), veins (blue) and airways (white). Modified from Ochs et al. ¹⁴³

Role for vascular function CT assessment in the global pandemic

Using total pulmonary vascular volumes and dividing the volumes into large and small cross-sectional area vessels, Slerno et al ¹⁴⁴ demonstrated a negative correlation between CT-derived larger vessel volume and the diffusing capacity of the lung (DL_CO), while inflammatory biomarkers were not correlated with DL_CO, suggesting a role for vascular measures when following subjects post COVID-19. The presence of PBV abnormalities via DECT have also been instrumental in demonstrating vascular dysfunction associated with COVID-19. ^18,145–147 As depicted in Figure 11, DECT-PBV combined with a non-contrast TLC scan was used to demonstrate significant disruption of regional V/Q relationships in the lung of an asymptomatic athlete with COVID-19 with an improvement V/Q at 30 and 90 days. ^18,145–147

Figure 11.

Histogram of regional measures of pulmonary an index of regional ventilation/perfusion matching (V/Q). Here, we show this application of DECT to assess regional lung function. At baseline (BL) a male athlete without respiratory symptoms was found to have a lesion in the left lower lobe (Upper Insert, Red Arrow) consistent with and later confirmed positive for COVID-19. A non-contrast scan was obtained at coached TLC and a contrast/DECT scan was obtained at FRC at a time point 17 sec into a slow delivery of half concentration iodinated contrast agent. As represented by the V and Q insets, the TLC image was matched to a DECT-derived virtual non-contrast FRC scan to assess regional volume change as an index of ventilation. Note that this ventilation map is, by design, aligned with the regional PBV results whereby PBV served as an index of perfusion. The baseline, 30 and 90 day PBV scans are shown in the “Q” inset with PBV superimposed onto a mid-transverse section from each time point and demonstrated by volume renderings placed below each transverse image. By representing regional ventilation and regional PBV as a percent of total volume change or total PBV respectively, the ratio of the two regional percentages provides an index of regional V/Q. The bars to the right represent the percentage of the lung with a V/Q of 4 or higher. Note that at baseline there was a considerable shift of the V/Q histogram away from a ratio of 1 with greater numbers of lung regions at both high and low V/Q. At 30 and 90 days the V/Q distributions moved closer to a ratio of 1 and the bars representing portions of lung above a V/Q of 4 were reduced. Modified from Nagpal et al. ¹⁸ BL, baseline; DECT, dual energy CT; FRC, functional residual capacity; PBV, perfused blood volume; TLC, total lung capacity.

Summary

As CT has evolved, so have tools for objective image evaluation. In Figure 12, key metrics are demonstrated. A key to qCT success is to image the lung with an up-front intent to objectively interrogate the images for the purpose of gaining an understanding of the underlying structure–function relationships. It has been clear from the onset of lung imaging that structure and function of the lung are highly dependent upon body posture and the state of lung inflation. Standardized protocols have been proposed seeking to harmonize data across scanners and scanning sites. ¹⁵⁰ This standardization has been adopted by numerous other multicenter studies and has served as a starting basis for standards set by the Quantitative Imaging Biomarker Alliance of the Radiologic Society of North America. ¹⁵¹ These standards should not just be used in research, but also in daily clinical practice. For qCT to have a clinical impact, one must be able to assess an individual in the context of normal. Key features of standardization for imaging the lungs include imaging at full inspiration (total lung capacity or TLC) to assess lung texture, search for nodules, assess airway structure and more. With deviations from TLC comes changes in texture, density, airway and vascular caliber and more. Similar to the pulmonary function laboratory, the patient must be coached to take a full breath in and hold it for the duration of imaging. Additionally, to assess regional fSAD one must scan at coached full expiration. Perfusion or PBV should not be assessed at high lung volumes and preferably at a FRC. CT technologists are the new front line in quantitative lung assessment, a role which has resided in the pulmonary function laboratory. The reconstruction dFOV should be set to maximize and standardize the spatial resolution and the dFOV should be kept the same upon return visits. Reconstruction kernels which do not artificially sharpen the edges of structures are desirable for quantitative evaluation of lung density. Radiation dose should be set as low as possible without interfering with quantification. Shortly after introducing spiral CT, Kalender and colleagues ¹⁵² introduced use of a hand-held spirometer to standardize lung volumes but it’s use may have just been ahead of the scanning technology. Reliable and repeatable achievement of RV and TLC remain one of the ongoing challenges of qCT still seeking a solution. Until then, coaching by the CT tech is the first line of defense.

Figure 12.

Composite of primary methods using CT to visualize and assess lung structure and function. (A) Segmentation of the whole lung, left and right lungs and individual lobes. (B) Airway segmentation with automated naming down to the sublobar segment. (C) Computational fluid dynamics used to estimate particle deposition (N/mm²) within a severe asthmatic upstream of an airway narrowing. Inset shows isosurfaces for 2.5 m/s (green) and 5 m/S(brown). ¹⁴⁸ (D) Voxels <-856 at RV have been designated as emphysema-like or air trapped respectively. Spheres show regions of air trapping concentrations and are color coded for lobar location. (E) One-dimensional CFD model predictions of pressure distributions for peak expiration of a subject with asthma. ¹⁴⁹ (F) Through matched inspiratory and expiratory images, the vDPM demonstrates regions of functional small airway disease (yellow), emphysema (purple), and normal lung (green) mapped onto a tMPR image where the airways and their associated parenchyma are projected onto a flat surface. ¹⁶ (G) AMFM-defined texture map ^91–95 of ground-glass opacities (green), ground-glass reticular (dark blue), honeycombing (yellow), emphysema (light blue), broncho-vascular bundles (pink) and normal (gray) (patient with interstitial pulmonary fibrosis). (H) Segmentation of the “total pulmonary vascular volume” or “TPVV.” Vessels above a radius of 0.75 mm are in blue and below 0.75 mm are in pink. (courtesy of Dr Sarah Gerard, PhD) I: (Courtesy of Drs. Amin Motahari, PhD and Sarah Gerard, PhD) 3D representation of the regional distribution of PBV assessed via DECT. ¹⁰ AMFM, adaptive multiple feature method; DECT, dual energy CT; PBV, perfused blood volume; TPVV, total pulmonary vascular volume; VDPM, ventilation defect probability method.

In this review, we have demonstrated that, with carefully controlled protocols, the relationships between lung structure, function and disease pathologies can provide critical insights needed for advancing our understanding of the normal and diseased lung.