Pulmonary Functional Imaging: Part 1—State-of-the-Art Technical and Physiologic Underpinnings

Abstract

Over the past few decades, pulmonary imaging technologies have advanced from chest radiography and nuclear medicine methods to high-spatial-resolution or low-dose chest CT and MRI. It is currently possible to identify and measure pulmonary pathologic changes before these are obvious even to patients or depicted on conventional morphologic images. Here, key technological advances are described, including multiparametric CT image processing methods, inhaled hyperpolarized and fluorinated gas MRI, and four-dimensional free-breathing CT and MRI methods to measure regional ventilation, perfusion, gas exchange, and biomechanics. The basic anatomic and physiologic underpinnings of these pulmonary functional imaging techniques are explained. In addition, advances in image analysis and computational and artificial intelligence (machine learning) methods pertinent to functional lung imaging are discussed. The clinical applications of pulmonary functional imaging, including both the opportunities and challenges for clinical translation and deployment, will be discussed in part 2 of this review. Given the technical advances in these sophisticated imaging methods and the wealth of information they can provide, it is anticipated that pulmonary functional imaging will be increasingly used in the care of patients with lung disease.

© RSNA, 2021

Online supplemental material is available for this article.

Summary

This state-of-the-art review (part 1) examined the field of pulmonary functional imaging and focused on technical requirements and physiologic measurements.

Essentials

• ■ Understanding the relationship between pulmonary structure and function at the alveolar-capillary interface and secondary pulmonary lobule requires imaging tools to interrogate lung disease.

• ■ The techniques and the functional biomarkers representing pulmonary ventilation, perfusion, gas exchange, and biomechanics were reviewed.

• ■ Pulmonary imaging analysis tools including computational and artificial intelligence (machine learning) approaches were also highlighted.

Pulmonary physiology is the basis of the intelligent practice of pulmonary medicine. And it always will be.

—John B. West, MD, PhD

Introduction

This review highlights the current state of the art for pulmonary functional imaging and areas of potential future advancement for radiologists, pulmonologists, basic researchers, trainees, and industry personnel. In the preface of his book (1), Dr West pointed out that “respiratory physiology frequently enjoys a surge of activity in time of war” (1). Our review article was written during the coronavirus disease 2019 (COVID-19) pandemic. Some patients with COVID-19 pneumonia have severe dyspnea and hypoxemia early in the course of the disease. It is known that more than 50% of the alveoli in the lung are located in the outer (peripheral) 30% of the radius of the lung (2–4), which overlaps with the area of diffuse peripheral ground-glass opacities on typical CT scans in patients with COVID-19 pneumonia (Fig 1). We will learn more about the pathophysiologic structure of the lungs through this pandemic. Perhaps this is our generation's opportunity to further the science of pulmonary medicine by highlighting the role of functional imaging in the study of this disease.

Figure 1:

CT scan in lower lung areas in an 83-year-old male patient with coronavirus disease 2019 pneumonia and severe hypoxemia. Bilateral multifocal patchy ground-glass opacities with peripheral distribution are visible (arrows).

The lung is a complex and intriguing organ. The purpose of the lung is simple in concept: to exchange oxygen and carbon dioxide at the alveolar membrane (Fig 2). However, the number of design solutions by evolutionary biology to achieve gas exchange and transport in living organisms belies the unique challenges of maintaining efficiency in varied living environments. The human lung is one of four completely different designs evolutionally developed in vertebrates. These include fish gills, amphibian skin, avian lungs, and mammalian lungs (5). For example, the bird lung works with a cross-current system so that the network of blood capillaries surrounds the ventilatory airflow in the center, a design shared with jet engines. The mammalian lung has evolved during the course of our ancestors, gradually moving from the ocean onto the land.

Figure 2:

Schematic rendering of the functional anatomic unit of gas exchange. The respiratory bronchiole (RespBr) terminating in a group of alveolar sacs (orange) and their associated sac walls (ASW; yellow colored pencil) with pulmonary artery capillaries (red colored pencil) carrying deoxygenated blood to the alveolar walls. Within the alveolar sac plasma carbon dioxide (CO₂; orange arrow) undergoes gas exchange and oxygen (O₂) diffuses (yellow arrow) into the pulmonary capillary venules to be bound by hemoglobin in the red blood cells. The oxygenated blood then flows into the pulmonary veins (dark blue) back to the left atrium.

Since Dr Röntgen's (6) Nobel Prize–winning demonstration of medical radiographic imaging in 1895, multiple imaging technologies have been developed (eg, digital radiography, US, radioisotopes for use in nuclear medicine, CT, PET, and MRI) (Tables 1–3, E1[online]). The rapid evolution of software computational methods, the miniaturization of hardware onto silicon microprocessors, and random-access memory devices have paved the way for the commercialization of the complex imaging devices now available for the noninvasive multidimensional visualization of the living human body. Exogenous administration and inhalation of contrast materials, gases, isotopes, and pharmacologic imaging agents have further aided the understanding of healthy and diseased human lungs.

Table 1:

Uses of Pulmonary Functional Imaging for Study of Respiratory Diseases

Table 3:

Quantitative Imaging Methods for Lung Perfusion with first Report

In this review we will discuss the basic structural and physiologic underpinnings of pulmonary functional imaging; summarize the techniques and derived functional biomarkers for ventilation, perfusion, gas exchange, and biomechanics; and describe advances in image analysis, computational, and artificial intelligence (ie, machine learning) methods pertinent to functional lung imaging. The applications of pulmonary functional imaging including both the opportunities and challenges for clinical translation and deployment will be discussed in Part 2.

Structural and Physiologic Aspects of Pulmonary Functional Imaging

Alveolar-Capillary Interface and Gas Exchange

What is gas exchange?—Gas exchange in the lung is defined by the diffusion of molecular oxygen into the plasma and red blood cell and the exit of carbon dioxide from the plasma to the alveolar space. The body requires a nearly continuous intake of oxygen to maintain aerobic respiration in the mitochondria. Fortunately, oxygen is the second most common molecule in our atmosphere (21%), although oxygen is not easily soluble in water. For aerobic respiration to occur at atmospheric conditions, a dramatic design change evolved to improve the efficiency of oxygen exchange in the lung. The estimated surface area for gas diffusion of a healthy adult human lung is approximately 50–70 m² (about the area of a tennis court) (2,3) and it takes approximately 0.25–0.35 seconds for the necessary amount of oxygen to diffuse from atmospheric air into the blood stream through alveolar surfactant, alveolar epithelial, basement membrane, and capillary wall. The volume of the pulmonary capillary bed is approximately 60–75 mL, which is equal to the normal right ventricular stroke volume (2,3). Consider that this volume of blood must spread evenly over an area the size of a tennis court for 0.8 seconds, the pulmonary capillary circulation time in human lungs, to allow sufficient time for gas diffusion before it transits back to the left atrium (2,3). During the capillary circulation time, oxygen diffuses into the plasma, and carbon dioxide diffuses back out into the alveolus, helping to maintain the pH of the blood by decreasing the amount of (soluble) bicarbonate. The theory of “symmorphosis” (7), as proposed by Dr Weibel (8), predicts that the size of a system's parts should match its functional demand, and that it will also have some reserve to prevent it from failing when stressed beyond normal use. How our lungs effortlessly perform this dynamic exchange every second by using a muscular assist from the chest wall and hemidiaphragms is a fundamental mystery of life.

How does gas exchange occur in human lung?—To breathe, the lung has evolved its miraculous three-dimensional alveolar structure (Fig 3). Dr Ochs' research group in Berlin, also inspired by the research of Dr Weibel, painstakingly performed serial two-dimensional scanning electron microscopy of the alveolus to help build models of its three-dimensional structure, now available for everyone to view as a movie (online) (9). On average, the tracheobronchial tree divides 23 times to create 300 million alveoli with average diameter of 200–300 mm and total surface area equivalent to the aforementioned tennis court. Minute ventilation at rest is approximately 6 L/min (3). The passive expansion and compression of the lungs are facilitated by the muscles of the diaphragm and thoracic cage, with neural control via phrenic nerves that are centrally controlled from our brainstem. During lung development, pulmonary arterial growth is driven by endothelial growth factor receptors under the influence of the tracheobronchial tree endoderm's invagination into the mesoderm of the lung. These pulmonary arteries divide along the airway to form the network of capillary vessels, which surround the alveolar walls to distribute blood along the alveolar surface. Cardiac output from the right ventricle is approximately 5 L/min, which coincides with the volume of minute ventilation of 6 L/min (3). The average thickness of the air-blood barrier, which is made up of the alveolar epithelium and its basal lamina, alveolar wall interstitium, basal lamina of the capillary endothelium, the capillary endothelium, plasma, and the membrane of the red blood cells, is only about 1.5 mm (10). The oxygenated blood is driven back to the left atrium by the negative pressure difference between the left atrium and the pulmonary veins. Consequently, any left heart diseases, leading to increases in left atrial (wedge) pressure, have a proportional increase in pulmonary venous and pulmonary arterial pressures to allow blood to continue to flow back to the left heart.

Figure 3a:

Anatomic basis for understanding pulmonary functional physiologic structure. (a) Magnified view of alveolar ducts and alveoli with orifices in a healthy patient. Alveoli are polygonal in shape and have orifices into alveolar ductal spaces. (b) Histologic sample of the healthy lung with hematoxylin-eosin stain. The divergence from terminal to respiratory bronchioles is shown. (c, d) A radiograph of 1-mm-thick specimen demonstrates the secondary lobules. The pulmonary arteries are located in the center, and pulmonary veins are located in the periphery of the secondary pulmonary lobule. The terminal and respiratory bronchioles are observed as tubular structures accompanying the pulmonary arteries. The primary lobules are indicated as yellow dashed circles in d. (Parts a–d adapted, with permission, from reference 10.) (e) Diagram shows anatomy and dimensions of secondary lobule and pulmonary acinus. Two secondary pulmonary lobules in the lung periphery are illustrated, with approximate dimensions of their components indicated. (Reprinted, with permission, from reference 12.)

Figure 3b:

Anatomic basis for understanding pulmonary functional physiologic structure. (a) Magnified view of alveolar ducts and alveoli with orifices in a healthy patient. Alveoli are polygonal in shape and have orifices into alveolar ductal spaces. (b) Histologic sample of the healthy lung with hematoxylin-eosin stain. The divergence from terminal to respiratory bronchioles is shown. (c, d) A radiograph of 1-mm-thick specimen demonstrates the secondary lobules. The pulmonary arteries are located in the center, and pulmonary veins are located in the periphery of the secondary pulmonary lobule. The terminal and respiratory bronchioles are observed as tubular structures accompanying the pulmonary arteries. The primary lobules are indicated as yellow dashed circles in d. (Parts a–d adapted, with permission, from reference 10.) (e) Diagram shows anatomy and dimensions of secondary lobule and pulmonary acinus. Two secondary pulmonary lobules in the lung periphery are illustrated, with approximate dimensions of their components indicated. (Reprinted, with permission, from reference 12.)

Figure 3c:

Anatomic basis for understanding pulmonary functional physiologic structure. (a) Magnified view of alveolar ducts and alveoli with orifices in a healthy patient. Alveoli are polygonal in shape and have orifices into alveolar ductal spaces. (b) Histologic sample of the healthy lung with hematoxylin-eosin stain. The divergence from terminal to respiratory bronchioles is shown. (c, d) A radiograph of 1-mm-thick specimen demonstrates the secondary lobules. The pulmonary arteries are located in the center, and pulmonary veins are located in the periphery of the secondary pulmonary lobule. The terminal and respiratory bronchioles are observed as tubular structures accompanying the pulmonary arteries. The primary lobules are indicated as yellow dashed circles in d. (Parts a–d adapted, with permission, from reference 10.) (e) Diagram shows anatomy and dimensions of secondary lobule and pulmonary acinus. Two secondary pulmonary lobules in the lung periphery are illustrated, with approximate dimensions of their components indicated. (Reprinted, with permission, from reference 12.)

Figure 3d:

Anatomic basis for understanding pulmonary functional physiologic structure. (a) Magnified view of alveolar ducts and alveoli with orifices in a healthy patient. Alveoli are polygonal in shape and have orifices into alveolar ductal spaces. (b) Histologic sample of the healthy lung with hematoxylin-eosin stain. The divergence from terminal to respiratory bronchioles is shown. (c, d) A radiograph of 1-mm-thick specimen demonstrates the secondary lobules. The pulmonary arteries are located in the center, and pulmonary veins are located in the periphery of the secondary pulmonary lobule. The terminal and respiratory bronchioles are observed as tubular structures accompanying the pulmonary arteries. The primary lobules are indicated as yellow dashed circles in d. (Parts a–d adapted, with permission, from reference 10.) (e) Diagram shows anatomy and dimensions of secondary lobule and pulmonary acinus. Two secondary pulmonary lobules in the lung periphery are illustrated, with approximate dimensions of their components indicated. (Reprinted, with permission, from reference 12.)

Figure 3e:

Anatomic basis for understanding pulmonary functional physiologic structure. (a) Magnified view of alveolar ducts and alveoli with orifices in a healthy patient. Alveoli are polygonal in shape and have orifices into alveolar ductal spaces. (b) Histologic sample of the healthy lung with hematoxylin-eosin stain. The divergence from terminal to respiratory bronchioles is shown. (c, d) A radiograph of 1-mm-thick specimen demonstrates the secondary lobules. The pulmonary arteries are located in the center, and pulmonary veins are located in the periphery of the secondary pulmonary lobule. The terminal and respiratory bronchioles are observed as tubular structures accompanying the pulmonary arteries. The primary lobules are indicated as yellow dashed circles in d. (Parts a–d adapted, with permission, from reference 10.) (e) Diagram shows anatomy and dimensions of secondary lobule and pulmonary acinus. Two secondary pulmonary lobules in the lung periphery are illustrated, with approximate dimensions of their components indicated. (Reprinted, with permission, from reference 12.)

Movie:

Download video file ^{(19.7MB, mp4)}

Still image capture from 4D flow MRI movie file showing the right ventricular outflow track (yellow arrow) and the left ventricular outflow track orange arrow) and the accompanying movie file. (Courtesy of Christopher François, MD, University of Wisconsin–Madison).

Secondary Pulmonary Lobule as a Functional Unit of the Lung

The anatomic structure, tissue morphologic structure, and geometry (of the airways, vessels, and parenchyma) and function of the lungs are necessarily related. Hence, a deep understanding of pulmonary anatomy and microstructure is key to interpretation of the measurements that stem from pulmonary functional imaging (11–13). The secondary pulmonary lobules, measuring 1–2.5 cm in diameter and composed of three to 20 acini or primary lobules (14), are polyhedral in shape and are bounded by the interlobular septa (Fig 3) (14,15). The secondary pulmonary lobule is supplied by a lobular bronchiole and a pulmonary arteriole in the center and drained via pulmonary veins located on the periphery. The pulmonary lymphatics distribute both centrally along the bronchovascular bundle and peripherally within the interlobular septa and subpleural pulmonary tissues, but not within the alveolar region.

It is important for radiologists to think of the secondary pulmonary lobule as the basic macroscopically visible functional unit of the lung when we interpret ventilation and perfusion abnormalities viewed by using various pulmonary imaging methods. It is also important to remember that there will be hypoxemia when there is a mismatch between ventilation and perfusion, and conversely a matched defect will not result in hypoxemia, which is part of the basic grand design of the human lung to maintain or expand gas exchange at stress or increased demand.

A curious feature of COVID-19 lung infection is silent hypoxia. In some of these patients it has been observed that carbon dioxide is driven off faster than normal whereas oxygenation is diminished. These patients are hypoxic but do not have a proportional increase in their respiratory rate. This observation suggests that there is a mismatch between ventilation and perfusion that may be causing intrapulmonary shunting of blood through pulmonary lobules that are compromised by this infection (ie, mismatch between ventilation and perfusion). The exact cause of this hypoxia without an increase in the respiratory rate remains to be more carefully studied.

Regional Distribution of Ventilation and Perfusion

The important role of the gravitational dependence of perfusion and ventilation was first published by West et al (1), who showed that in the upright position the lung bases are more perfused and less well ventilated, whereas the lung apex is less perfused and better ventilated. In the supine position the dorsal (posterior) lung is better perfused and less well ventilated, whereas the anterior lungs are less well perfused and better ventilated. An easy way to think about this is that water follows gravity to that region closest to the ground. Lung per-fusion is gravitationally dependent. We can use this principle to hypothesize that in zero gravity the astronauts of the space station will have an equal distribution of their ventilation and perfusion to all lung segments; the gravitational dependence of pulmonary ventilation and perfusion is “lost in space,” so to speak (16).

Summary of Pulmonary Functional Imaging Methods and Measurements

Pulmonary functional imaging encompasses an array of techniques to view and measure gas distribution, ventilation, perfusion, the exchange of gas across the alveolar gas-tissue barrier, and biomechanical characteristics of the lung (Table 1, Fig 4). As shown in Table E1 (online), in theory there are a number of advantages of pulmonary functional imaging compared with pulmonary function testing measurements that favor pulmonary functional imaging. Unfortunately, in part because of the relative costs and complexity of imaging and despite decades of technical research advances, data showing that pulmonary functional imaging measures improve patient treatments and outcomes have lagged. In this light, different approaches to pulmonary functional imaging are treated systematically in the following sections, with a focus on their quantitative capabilities, clinical potential, and research status.

Figure 4:

Summary schematic for the available pulmonary functional imaging methods. This schematic does not include molecular imaging and its major contributions to lung cancer imaging and treatment response. ADC = apparent diffusion coefficient, ASL = arterial spin labeling, DCE = dynamic contrast enhancement, DECT = dual-energy CT, Exp = expiratory, FD = Fourier decomposition imaging, FDG = fluorodeoxyglucose, HP = hyperpolarized, Insp = inspiratory, NM = nuclear medicine, RBC = red blood cell, VDP = ventilation defect percentage.

Ventilation Imaging

Pulmonary inhaled gas distribution and ventilation can be viewed by using nuclear medicine methods with gaseous radionuclides such as xenon 133 (¹³³Xe), krypton 81 (⁸¹Kr), aerosolized technetium 99m (⁹⁹Tc) diethylenetriamine penta-acetic acid, or ⁹⁹Tc-labeled carbon nanoparticles (Technegas). The resultant γ emissions of the inhaled gases result in planar (two-dimensional) or three-dimensional tomographic images by using SPECT and PET (with gallium 68 [⁶⁸Ga]-labeled carbon nanoparticles or Galligas). These may be used alone to measure ventilation or in combination with injected contrast agents to measure perfusion and then to estimate the ratio of ventilation to perfusion.

A number of CT inhaled gas methods may be exploited to provide functional or ventilation information. For example, free-breathing or four-dimensional (4D) CT has been used in lung cancer radiation treatment planning and can be manipulated to generate ventilation maps (17,18). Inhaled krypton (19,20) and Xe dual-energy CT (21–23) have also been pioneered and used as research tools (Fig 5). Another indirect measure of ventilation stems from co-registration of inspiratory and expiratory CT to generate parametric response maps (24). In images on which voxels have normal lung attenuation at inspiration but are abnormal at expiration, parametric response maps are believed to reflect gas trapping and functional small airway disease. In preliminary studies, parametric response maps in patients with chronic obstructive pulmonary disease showed some correlations with inhaled gas hyperpolarized MRI ventilation defects (25).

Figure 5:

Xenon (Xe) ventilation dual-energy CT in patients with, A, B, chronic obstructive pulmonary disease and mild emphysema and with, C, D, moderate to severe emphysema. A, Dual-energy CT image shows minimal centrilobular emphysema in right lower lobe on axial weighted average image and, B, homogeneous Xe enhancement on Xe ventilation map in both lungs. C, Axial weighted average image shows severe emphysema with bronchial wall thickening in both upper lobes. D, On Xe ventilation map, the inhomogeneously decreased Xe enhancement is identified in both upper lobes, whereas Xe ventilation is relatively preserved in the central areas of both upper lobes (arrows).

MRI also provides several free-breathing and static breath-hold methods that can be used to generate maps of gas distribution and ventilation. Early studies that used inhaled hyperpolarized gas MRI (26) showed relatively homogeneous gas signal across healthy tissue, whereas signal voids or ventilation defects were noted in patients with lung disease (27). In lung cancer, large, homogeneous ventilation defects were observed that directly related to tumor burden, whereas in patients with asthma or chronic obstructive pulmonary disease there were both large homogeneous and smaller patchy defects, including wedge-shaped defects that follow the segmental and subsegmental airways. Wedge-shaped defects in asthma (28) were observed to be either persistent or intermittent over time, but typically these remained in the same regional locations over long periods (29). Like 4D CT, free-breathing proton 4D MRI provides a way to generate maps of specific ventilation (30) and these maps correlate well with ³He MRI ventilation defect maps. Free-breathing proton MRI acquisitions can also be manipulated by using Fourier decomposition (31) and other phase-resolved methods (32) to generate ventilation and perfusion maps.

In a similar manner, oxygen-enhanced MRI exploits the paramagnetic properties of inhaled oxygen as a T1 contrast agent by using conventional proton-based MRI methods and receiver coils. In addition, oxygen transfer abnormalities may be quantified in patients with different pulmonary diseases (33–43) (Fig 6).

Figure 6:

Images in a 43-year-old male patient with pulmonary emphysema and bullae. A, Thin-section coronal image at lung window setting (left) and quantitatively assessed thin-section multiplanar reconstruction image with density-masked CT technique (right). Emphysematous lung in right upper lung (small arrow) and giant bulla in the left upper lung (large arrow) are clearly demonstrated. On quantitatively assessed thin-section multiplanar reconstruction image, healthy lung appears as green, and emphysematous lung or bullae appears as red with applying threshold value 2950 HU. B, Oxygen-enhanced MRI scan shown as a relative-enhancement map (gray scale, 0% [black] and 50% [white]). Emphysematous lung in right upper lung (small arrow) and giant bulla in the left upper lung (large arrow) are clearly demonstrated as black areas. In addition, the remainder of both lungs are heterogeneously enhanced because of emphysematous lung and airflow limitations.

Fluorine 19 (¹⁹F) ventilation MRI has recently been pioneered to provide an alternative to hyperpolarized noble gas MRI (44–46). Inert fluorinated gases are nontoxic, abundant, relatively inexpensive, and may be undertaken without hyperpolarization equipment (44–46). The ¹⁹F gas MRI potentially provides functional information that is similar to images obtained from hyperpolarized noble gas MRI (44–46), although multiple breaths of gas are typically required to improve signal-to-noise ratios. In addition, the pattern of ventilation defects for the steady state of ventilation imaged with ¹⁹F MRI is not the same as what is found with hyperpolarized ³He or ¹²⁹Xe methods (47). This is thought to be related to the fact that with steady-state imaging, collateral ventilation allows poorly ventilated areas to be ventilated by the imaging gas. This occurs with ¹⁹F ventilation MRI because there are multiple breaths used for the image acquisition, whereas hyperpolarized noble gas imaging relies on one single breath hold (inspiration, total lung capacity, and expiration) for all of its data. Various quantitative methods for ventilation imaging are summarized in Table 2.

Table 2:

Quantitative Imaging Methods for Ventilation with First Report

Perfusion Imaging

Pulmonary perfusion measurements are clinically important in obstructive airways disease and when ventilation and perfusion mismatch are suspected. The current methods for perfusion imaging use technetium 99m (^99mTc) macroaggregated albumin, SPECT, and SPECT fused with CT (ie, SPECT/CT). These methods have been clinically used for years to image diseases of perfusion (eg, chronic thromboembolic pulmonary hypertension). However, the applicability of perfusion scanning, SPECT, and SPECT/CT with ^99mTc macroaggregated albumin has remained problematic compared with the higher spatial resolution of CT and MRI. Compared with SPECT or SPECT/CT, ¹³N₂-saline bolus injection method, intravenous administration of ¹⁵O-labeled water and albumin microspheres radiolabeled with ⁶⁸Ga (ie, ⁶⁸Ga-labeled microspheres) have been accepted as PET tracers for lung perfusion assessment (48,49). The current PET/MRI instruments are useful because the MRI can be performed simultaneously with PET. However, PET with these tracers is primarily used for research and is not covered by health insurance reimbursement. There is a concerted effort on the part of functional pulmonary imagers to move from ventilation and perfusion scanning to modalities that have better spatial and temporal resolution of the component ventilation and perfusion images.

CT methods to evaluate perfusion include contrast-enhanced multienergy CT that shows iodine distribution maps in patients with pulmonary vascular diseases (Fig 7). Contrast-enhanced multienergy CT can be performed by using a number of different acquisition methods such as the dual-source and single-source CT systems with ultrafast tube voltage switching or the dual-layer detector system (50,51). Another approach uses quantitative dynamic first-pass contrast-enhanced perfusion CT by means of area-detector CT systems (52–60). The area-detector CT refers to a 320-detector-row CT system, initially developed for whole-heart coverage. This is designed with a large detector that is able to obtain isotropic volume data of the chest for analysis of the lung parenchyma, lung nodules, or masses in a single rotation (Fig 8). This hardware also enables dynamic first-pass contrast-enhanced perfusion by means of continuous dynamic scanning, allowing for qualitative and quantitative evaluation of lung perfusion (52–60). Although multienergy CT techniques offer advantages over conventional CT regarding the ease of obtaining qualitative assessment of pulmonary perfusion in the clinical setting, dynamic first-pass contrast-enhanced perfusion area-detector CT has the potential to become one of the most practicable pulmonary functional imaging methods and to provide morphologic and functional information with high spatial and temporal resolution. However, dose reduction by using iterative reconstruction needs to be carefully paired with maintenance of the accuracy of all perfusion parameter measurements (57). Therefore, radiation dose reduction may be one of the important factors in the clinical adoption of this technique.

Figure 7:

Example of dual-energy CT angiography in a patient with acute pulmonary embolism. A, Axial CT pulmonary arteriographic image shows saddle emboli in the main pulmonary artery and right and left pulmonary arteries (arrowheads) and small clots in a segmental pulmonary artery in the right lower lobe (arrow). B, Axial fusion image of pulmonary blood volume map of dual-energy CT shows large wedge-shaped perfusion defects in the right lower lobe (arrowheads).

Figure 8:

Images in a 73-year-old male patient with invasive adenocarcinoma (arrows) in the left lower lobe. Thin-section coronal multiplanar reconstruction image and quantitatively assessed whole-lung perfusion CT image (left) displayed as total perfusion map (right; combination of pulmonary arterial blood supply and systemic arterial blood supply from bronchial artery). Invasive adenocarcinoma (arrows) appears as part-solid nodule and decreased perfusion area compared with normal lungs. In addition, solid component shows relatively higher perfusion than ground-glass opacity. Whole-lung perfusion map was generated from two dynamic contrast-enhanced area-detector CT data sets by using our proprietary software.

Since the 1990s, noncontrast-enhanced and contrast-enhanced MR angiography and perfusion MRI have been suggested as having potential for viewing pulmonary vasculature and blood flow by using a variety of methods (61,62). In addition, contrast-enhanced MR angiography with parallel imaging techniques, known as time-resolved contrast-enhanced MR angiography or 4D contrast-enhanced MR angiography, has been used clinically in patients with pulmonary diseases over the last few decades (63–66). Moreover, dynamic first-pass contrast-enhanced perfusion MRI has been used for the semiquantitative and qualitative assessment of the pulmonary circulation in patients with various pulmonary diseases (60–62,67–72). Currently, all MRI scanners from any vendor with magnetic field strength equal to or greater than 1.5 T can support time-resolved contrast-enhanced MR angiography and dynamic first-pass contrast-enhanced perfusion MRI (Fig 9). Whereas the hardware and software for this has been available for more than 10 years, few centers have adopted MRI lung perfusion for clinical use. A better understanding of the strengths and weaknesses of MRI perfusion compared with other clinical modalities is needed (Table 3).

Figure 9:

Raw perfusion coronal MRI scan from a movie by using the differential subsampling with Cartesian ordering technique (water images) after the bolus administration of contrast material (Multihance; Bracco) shows the dextrophase and levophase of cardiovascular, pulmonary, and great vessel contrast enhancement (arrows).

Imaging of Gas Exchange

Opportunities for alveolar tissue and blood imaging are offered by ¹²⁹Xe MRI because unlike ³He or ¹⁹F, Xe has modest tissue solubility, which allows for transmembrane alveolar tissue translocation and binding to red blood cell hemoglobin (73–75). Therefore, whole-lung studies (76,77) have demonstrated that inhaled ¹²⁹Xe exhibits distinctly different resonance frequencies related to its presence in the gas, tissue, and red blood cell environments (73,76,78). By tuning the MRI scanner to these frequencies, spectral maps of alveolar tissue density and red blood cell distribution may be acquired in addition to the maps of gas distribution (Fig 10). By using such approaches, ¹²⁹Xe MRI provides an opportunity to noninvasively estimate alveolar thickness, previously only accessible through invasive biopsy and histologic analysis. In this approach, the ¹²⁹Xe signal can be manipulated by repeatedly saturating the tissue compartment and acquiring images at different delay times. Mathematical models of alveolar geometry can then be used to estimate alveolar thickness on the basis of the measured diffusion rate and amplitude of ¹²⁹Xe gas diffusion into the tissue (79–82). Such calculations determined an idiopathic pulmonary fibrosis septal thickness that was 7.2 mm thicker (72%) than in healthy participants, and which can also be used to estimate alveolar surface-to-volume ratios, a crucial physiologic measure (82). Therefore, with the application of this phenomenon in academic practices, Xe polarization transfer contrast at hyperpolarized ¹²⁹Xe MRI has been suggested as useful for gas transfer imaging in the last few decades (83–86).

Figure 10:

Demonstration of gas diffusion block in interstitial lung disease. Xenon 129 (¹²⁹Xe) gas is soluble in the tissue and plasma and red blood cells (RBCs) from the alveolar airspace into the tissue-capillary interface. This results in measurable chemical shifts for the tissue and plasma and red blood cell compartments approximately 200 ppm downfield from the gas resonance (upper right). Spectroscopic imaging can be fused with anatomic ultrashort echo time (UTE) images (lower left) in fibrotic lung disease to demonstrate diffusion block in which the tissues and plasma take up the ¹²⁹Xe homogeneously (lower middle), but red blood cell uptake is heterogeneous (lower right) or not present (lower row arrows) in a region of apparent honeycombing. (Upper left image used with permission from reference 111).

Pulmonary Biomechanics

Functional imaging measurements have been pioneered to estimate lung compliance by tracking voxel motion at full inspiration (total lung capacity) and at full expiration (residual volume). Lung compliance estimates derived from free-breathing and static volume 4D CT and 4D MRI can be obtained by using deformable image registration. Regional compliance can be computed by using the ratio of volumetric variation and the associated stress in each voxel, representing lung elasticity, and computed by using a finite element model. The mathematics involved in this transformation are nontrivial and the interested reader is encouraged to study other publications that deal with this issue in more detail (87). The method for MRI borrows from CT and uses ultrashort echo time images to track the change in position for each lung voxel between the total lung capacity and residual volume scans (88).

Multisystem Functional Imaging

All of the organs have some degree of interconnectivity with the lungs. This has been clearly demonstrated during the current COVID-19 pandemic. The musculoskeletal system, cognition (cortical function), brain stem function, cranial nerve function, cardiovascular system, hematologic system, immune system, hepatic function, renal function, types of particulates in the air, and the gaseous environment (molecular composition and concentration) the human is exposed to critically affects longevity. The multiple ways in which this occurs is beyond the scope of this article. It would be accurate to state that the rate-limiting step for mammalian life and aerobic metabolism is lung function.

For example, the metabolic syndrome can be used to help illustrate the interconnected nature of the lung with the rest of the body. With an excess of caloric consumption leading to the metabolic syndrome, there are a number of changes the body goes through (89). In the liver, nonalcoholic steatohepatitis may occur leading to hepatitis and cirrhosis. Hepatopulmo-nary syndrome and portopulmonary hypertension may occur in the advanced stages of cirrhosis. Insulin resistance with the development of type II diabetes can occur, increasing the risk of infections because of neutrophilic dysfunction in the setting of elevated glucose. Systemic hypertension may also occur, causing small arteriolar sclerosis in multiple organs. The function of the renal tubules decreases with systemic hypertension, activating the renin-angiotensin system to further increase the systolic pressure helping to worsen end-organ damage. The increased amount of intra-abdominal fat exerts pressure on the hemidiaphragms and makes them less efficient for the depth of inspiration. The subsequent development of obstructive sleep apnea adds insult to injury as pulmonary hypertension develops with the repeated bouts of nocturnal hypoxia that occurs in this over-weight population.

Advances in Dose Reduction, Image Analysis, Computational Methods, and Machine Learning

Radiation Dose Reduction

There has been extensive interest in reducing medical radiation from CT imaging. This has resulted in the use of multiple strategies to mitigate the radiation dose and the effective radiation dose: use of nationally vetted and standardized low-dose CT protocols optimized to age and body habitus; acceptance of more noise in diagnostic imaging for screening examinations; use of iterative reconstruction; combining deep learning and machine learning with iterative reconstruction methods; and, for younger individuals, simply using other imaging tests that do not involve x-rays or CT (eg, US and MRI) (90–93).

Imaging Analysis for Pulmonary Functional Imaging

To extract functional information in a quantitative way, images should be processed by dedicated software. Image processing methods vary according to imaging methods and the specific pathologic structure that must be evaluated. Nevertheless, it generally includes several typical processes: segmentation of lung, lobes, pulmonary vessels, and airways; registration of sequential images to match exact anatomically corresponding locations; normalization of images to minimize variation of measured signals between sequential images or cases; enhancement of signals or suppression of noise; and model fitting to extract quantitative functional measures from dynamic images.

Many image processing methods have been proposed and used for ventilation and perfusion and other functional assessment of lung physiologic structure (Fig 11). However, there are several fundamental issues to be overcome in applying these methods in routine clinical practice. First, variation of the measured signal and noise pattern originating from different imaging sequences and imaging machines prevent the evolution of functional measures to imaging biomarkers. Second, there is a lack of consensus for imaging processing methods. Different segmentation methods, registration methods, and model fitting are used in various software so that the results with these different methods cannot be compared. Last, most of the image processing methods are imperfect, particularly when they are applied to diseased lung images. They are often semiautomatic and require human interactions. This may result in significant measurement variations. Moreover, the requirement of human correction or confirmation for the processing seriously hampers the use of this technique in routine clinical workflow because it raises issues of seamless and timely implementation.

Figure 11:

Still image capture from four-dimensional flow MRI movie file shows the right ventricular outflow track (yellow arrow) and the left ventricular outflow track (orange arrow).

Radiomics in Cardiopulmonary Imaging

Radiomics is a field of image postprocessing that aims to use computer algorithms to analyze image data in ways the human eye and brain are not able to process intuitively (94). There has been a great interest in developing imaging-based radiomic signatures that help to predict survival for non–small cell lung cancer (95–97). Some authors have used this method to help predict recurrence-free survival in stage 1 non–small cell lung cancer (98). The concept of radiomics, which is generation of high-throughput quantitative data from the image, is also applicable to other pulmonary functional imaging because most of pulmonary functional imaging produce multiple quantitative results and their integration by using a radiomics approach is necessary. Although interesting, the radiomics method is being rapidly overtaken by the use of artificial intelligence and deep learning methods that are able to automatically process large data bases for training and validation, and also use external test sets to demonstrate their accuracy in predicting outcomes in patients with lung cancer and other chronic diseases.

Artificial Intelligence for Pulmonary Functional Imaging

Artificial intelligence, particularly deep learning technology with convolutional neural networks, has recently been introduced for medical imaging (99,100). The usefulness of these methods in the detection and characterization of chest pathologic abnormalities by using imaging has been substantiated in many studies (101). Artificial intelligence will significantly accelerate the clinical use of pulmonary functional imaging. First of all, use of deep learning methods at image acquisition will significantly improve the quality of functional imaging. In most of pulmonary functional imaging, dynamic scanning of the lung to acquire multiple time series of the region of interest is essential. To obtain multiple CT series within acceptable radiation exposure, low-dose scanning of the region is essential, with resulting increased image noise and decreased signal-to-noise ratio and contrast-to-noise ratio. By using deep learning methods, the noise on the low-dose CT images can be significantly reduced, providing improved information for the functional assessment (99). For MRI, these deep learning approaches can be used to decrease acquisition time, thereby improving the temporal resolution. Deep learning is also useful to reduce artifacts related to patient motion and rapid acquisition (102).

The variation of signal intensity caused by using different imaging protocols, different imaging machines, and patient factors is one of the critical issues to be addressed in the routine clinical use of pulmonary functional imaging. Many researchers have tried to minimize this variation by standardizing imaging protocols and quality control for multicenter studies. However, this level of quality control is largely impractical in a busy clinical setting. Recent studies have shown that deep learning methods are also useful in minimizing the signal variation between images generated with different acquisition protocols and by using different machines. Several studies showed that this technology can decrease the measurement variation in emphysema and radiomic assessment (103,104).

Artificial intelligence will also be helpful in every step of image analysis in pulmonary functional imaging, which will serve to fully automate the analysis and totally integrate the output to current clinical picture archiving and communication systems. Defining the region of interest, whether it is the whole lung or a specific region (eg, the pulmonary artery for dynamic perfusion analysis), is the first step in the image analysis. However, this segmentation step is not always easy, particularly when the lung or target area has an atypical shape or location caused by disease or anatomic variation, which often requires correction from a human expert. Recently, studies have shown that deep learning methods significantly improve the performance of image segmentation (105–107). For dynamic assessment, aligning the anatomically corresponding locations in sequential images is the next critical step. However, exact registration is challenging, particularly when considering the complex motion of the lung caused by respiration and cardiac pulsation. Imperfect image registration is a critical problem for dynamic analysis by using multiple images because this often results in measurement errors. Another issue in image registration is that it often requires long image processing times, particularly for CT data. To address this limitation, deep learning methods are also useful in enhancing the accuracy and speed of imaging registration (108). In addition, deep learning methods are useful in normalizing signals and decreasing noise in dynamic series images.

In the near future, it is likely that all of these artificial intelligence methods will be used in pulmonary functional image analysis and will help to accelerate the clinical adoption of functional lung imaging. Moreover, pattern recognition and clustering of pooled imaging data will provide insights into our understanding of the pathophysiologic causes of lung diseases (Fig 12). Data mining by using machine learning technology will be helpful in building models of diseases which may further act to improve image acquisition times.

Figure 12:

Artificial intelligence–based automatic quantification of regional disease pattern on CT images. A, The original image and, B, parametric map of thin-section CT of interstitial lung disease show that deep learning–based automatic quantification of local disease pattern is possible. Red on the parametric map represents the areas of honeycombing; yellow, reticulation; and green, ground-glass opacity.

Future Directions

A Fleischner Society position paper by Hatabu et al was published in the November 2020 issue of Radiology. This white paper discussed the recent advancements in functional MRI lung imaging, and three-category recommendations for the clinical use of pulmonary MRI were provided (Appendix E1 [online]) (109,110).

Conclusions

We reviewed the structural and physiologic foundations of pulmonary functional imaging. The various techniques and functional biomarkers for ventilation, perfusion, gas exchange, and biomechanics, and the relevant advances in image analysis, computational methods, machine learning, and artificial intelligence that have evolved over the past 3 decades were highlighted. Building on this summary of the basic underpinnings of pulmonary functional imaging and the broad array of physiologic biomarkers it can provide, part 2 of this state-of-the-art review will summarize current and emerging clinical applications, including both the opportunities and challenges for the clinical translation and adoption of functional lung imaging.