Chronic obstructive pulmonary disease (COPD) is one of the leading causes of death worldwide. Main risk factors are tobacco smoking and air pollution.1 With nearly 5% of the population, a significant number of people are affected by this condition.2 Chronic obstructive pulmonary disease usually worsens over time, especially when the risk factors persist, and can ultimately result in death. Before that, relevant disability is caused, leading to impaired life quality as well as to high health care and economical costs. It is estimated that 3% of all disability is related to COPD.2 The disease is induced by a chronic inflammatory process of the airways, causing a degradation of the connective tissue of the lungs. This leads to a destruction of the small airways, impairing the ability to exhale completely, and to a destruction of the alveoli, resulting in pulmonary emphysema.

Chronic obstructive pulmonary disease is diagnosed by spirometry,3,4 as it measures breathing patterns and provides several parameters representing lung function. However, the sensitivity of spirometry for early stages of emphysema is rather low.5,6 X-ray computed tomography (CT) can visualize density variations in lung tissue and is currently the most accurate imaging technique to assess the in vivo presence, pattern, and extent of emphysema.5,7 Hereby, early stages of emphysema can be detected, even in patients without meeting the diagnosis of COPD (corresponding to GOLD 0).6 The visual presence and severity of emphysema is correlated with an increased mortality risk.5 However, CT causes a relevant radiation exposure with approximately 6.2 mSv for standard chest CT examinations.8 Although a significant dose reduction is possible by using low-dose techniques,9 still a significant radiation exposure is applied. For example, the mean effective dose of 1.4 mSv was found for low-dose CT in the national lung screening trial.10 A most recent study examining ultra-low-dose chest CT via deep learning–based image reconstruction reported a mean effective dose of 0.39 mSv.11

Conventional radiography is a commonly used first-line imaging modality of the chest, as it is widely available, relatively cheap, and fast. Only a low radiation dose of 0.02 mSv for a posterior-anterior scan is reported for this modality.12 However, it offers only very limited sensitivity to the presence of emphysema, especially in cases of mild emphysema.13 In summary, neither spirometry, CT, nor chest radiography is viable screening modalities for pulmonary emphysema. Therefore, a new imaging method, which is sensitive for lung diseases, dose-compatible with screening applications, and quickly performable, is highly desirable.

Dark-field chest radiography (dfCXR)14 may present a solution to this problem. It is sensitive to the microstructure of the specimen under examination by measuring the amount of small-angle scattering at material interfaces, in particular at length scales far below the spatial resolution of the imaging system itself. Ever since its transition from synchrotron imaging to laboratory setups,15 the technique has been further refined16,17 and its diagnostic potential has been proven in various small animal models,18–20 pigs,21 and human cadavers,22,23 finally leading to a first clinical prototype.24 Our system uses a conventional x-ray tube and a flat-panel detector combined with 3 gratings, enabling the acquisition of dark-field and attenuation-based radiographs during a single examination, at about twice the dose of conventional chest radiography.25 We have previously used this system for a first proof-of-principle demonstration that dfCXR can detect and quantify emphysema,24 and for quantitative evaluations of the dark-field signal.26,27 Here, we now present a clinical reader study on 88 participants, comparing the diagnostic value of conventional and dfCXR for early detection and quantification of emphysema.

MATERIALS AND METHODS Participants

This prospective study was conducted in accordance with the Declaration of Helsinki (as revised in 2013). Approval of the institutional review board and national radiation protection agency was obtained before this study (Ethics Commission of the Medical Faculty, Technical University of Munich, Germany; reference no. 166/20S).

Patients with a medically indicated chest CT scan were screened for participation. Inclusion criteria were legal age (18 years and older) and the ability to stand upright and hold breath for the duration of the scan. We included participants either without lung impairment or with any stage of pulmonary emphysema, based on their CT. Exclusion criteria were pregnancy and any lung condition other than emphysema, such as fibrosis, cancer, pneumothorax, masses, or infiltrates. Participants gave written informed consent before participation. The screening process is visualized in Figure 1. An estimated 10% of screened patients were eligible for participation; of those, approximately 25% agreed to participate in the study, resulting in approximately 3500 screened patients. The maximum time between the CT examination and the dark-field imaging was 24 hours. Typically, the dark-field imaging was obtained directly after the CT examination.

FIGURE 1:

Visualization of the screening process. Computed tomography (CT) examinations of patients who received a diagnostic CT during were screened for meeting the inclusion criteria. Patients were included when meeting all inclusion criteria and willing to participate.

A total of 40 participants were previously studied by Gassert et al,26 who evaluated the characteristics of dark-field chest radiographs in healthy humans only. A total of 83 participants were also described in a study by Urban et al,27 analyzing the quantitative dark-field signal, without including a reader study and without comparison to conventional radiography. A total of 77 participants were also studied by Willer et al,24 who assessed the diagnostic accuracy of dark-field imaging in emphysema diagnosis based on a reader study of dark-field images only. They did not use the simultaneously acquired attenuation-based chest radiograph, compare the diagnostic value of the dark-field radiographs to conventional radiographs, or perform an receiver operating characteristic (ROC) analysis.

This current study adds the comparison of dfCXR to conventional radiographs for diagnosis and staging of emphysema, in the clinically relevant setting where both dark-field and attenuation-based image from the dfCXR system are used.

Imaging

Participants were imaged at a commercial radiography device (DigitalDiagnost; Philips Medical Systems, Hamburg, Germany) in posterior-anterior orientation, yielding a conventional digital attenuation-based radiograph. The effective dose for the reference person at this device is 0.018 mSv.

The used dfCXR system (see Fig. 2) has previously been described in detail in Willer et al.24 It consists of a conventional radiography system in combination with a 3-grating interferometer, enabling the detection of both attenuation and dark-field signal simultaneously. The effective dose for the reference person for 1 posterior-anterior image acquisition, yielding both attenuation-based and dark-field image, is 0.035 mSv.25 Attenuation radiographs underwent standard postprocessing (UNIQUE; Philips Medical Systems, Hamburg, Germany) adjusted for this specific system.

FIGURE 2:

The dark-field chest radiography prototype. Schematic drawing (A) and photograph (B) of the dark-field radiography system. Three gratings, mounted on the movable interferometer arm, enable the detection of both dark-field and attenuation signal in a single acquisition. The active grating area (dark gray in A) is scanned from bottom to top during 1 image acquisition (7 seconds) to cover the full field of view (light gray in A).

Reader Study

To establish the ground truth on the lung condition of each patient, the CT scans were analyzed and graded independently by 3 radiologists (J.H.B., A.P.S., and A.A.F. with 4, 7, and 13 years of experience in CT imaging) according to the Fleischner scale (absent, trace, mild, moderate, confluent, and advanced destructive emphysema).7 The ground truth was obtained by taking the median of the 3 readers.

Conventional radiographs from the commercial device were read by 3 independent radiologists (F.T.G., A.P.S., and D.P. with 3, 7, and 12 years of experience). Both attenuation-based and dark-field radiographs from the dfCXR system displayed simultaneously were also read by 3 independent radiologists (F.T.G., A.P.S., and A.A.F.), experienced in dark-field image interpretation. Both emphysema detection and severity (no emphysema, mild emphysema, moderate emphysema, severe emphysema) and diagnostic confidence for emphysema severity (0–5) were collected. For directions, the readers discussed together the emphysema severity rating of exemplary patients approximately 4 weeks before the reader study. During the reader study, no further directions were provided. Images were shown in random order, with the same window width and level within the modalities. The readers could adjust window settings to their liking and had the possibility to return to previous images and change their rating.

Statistical Analysis

Statistical analysis was performed using Python, specifically NumPy,28 SciPy,29 and Pandas,30 as well as R.31 A significance level of 0.05 was chosen for all tests. Interreader agreement was assessed with the multirater Fleiss κ. Diagnostic confidence ratings of the 2 imaging methods were tested for significant differences using the Student t test. To analyze the capability to detect at least mild emphysema, subjects were grouped by presence of at least mild emphysema (Fleischner scale “absent” and “trace” versus “mild” to “advanced destructive”). Data from all readers were pooled. The utility of images from either device for prediction of emphysema was assessed by the use of ROC curves and the respective area under the curve (AUC). The AUC values were tested for significant differences using the Obuchowski method32 for correlated and clustered AUC data. In addition, a more detailed analysis was performed regarding the staging of emphysema. Participants were grouped by their ground truth classification, and AUC values for differentiation of adjacent groups were calculated. A z test based on these AUC values was used to determine whether there was a significant difference in the ratings of the 2 groups.

RESULTS Participants

A total of 88 participants were included in this study, 54 men and 34 women, with a mean age of 64 ± 12 years (standard deviation). Further details are listed in Table 1. In the course of this study, according to the determined CT-based visual emphysema grades, we included 36 subjects with absent, 21 with trace, 11 with mild, 9 with moderate, 7 with confluent, and 4 with advanced destructive emphysema.

TABLE 1 - Subject Demographics Emphysema Severity No. Participants Men/Women Age, y Weight, kg All 88 54/34 64 ± 12 74 ± 15 Absent 36 21/15 62 ± 11 79 ± 16 Trace 21 13/8 63 ± 12 74 ± 15 Mild 11 5/6 67 ± 13 67 ± 11 Moderate 9 6/3 65 ± 10 74 ± 14 Confluent 7 5/2 72 ± 8 65 ± 10 Advanced destructive 4 4/0 67 ± 10 69 ± 13 Values are given as mean ± standard deviation. Emphysema severity was assigned according to the Fleischner Society grading system.7
Conventional and Dark-Field Radiographs

Figure 3 shows exemplary conventional radiographs from the commercial system and attenuation-based and dark-field images acquired with our prototype dfCXR system of 5 participants with increasing emphysema severity. The different contrast generating mechanisms between the 2 modalities are clearly visible: the attenuation chest x-rays highlight attenuating structures such as bones and soft tissue. The contrast of ventilated lungs is inherently low, since the lung consists mainly of weakly attenuating air and is overlaid by other tissue. Therefore, emphysema can mainly be diagnosed based on secondary symptoms such as the flattened diaphragm or the barrel shape of the lung in more severe emphysema stages (Figs. 3D, E, I, J). A diagnosis based on a decreased density is only possible in advanced stages of COPD in which the mentioned secondary signs are also present. The dark-field images on the other hand gain their contrast from the alveolar microstructure itself and are therefore sensitive to the destruction of the alveoli in emphysema. Thus, while the healthy lungs with intact alveolar structure exhibit a strong dark-field signal (Fig. 3 F), lungs affected by—even mild—emphysema show a much lower dark-field signal (Figs. 3H–J), where the signal decrease is a measure for emphysema severity.

FIGURE 3:

Example radiographs. Example conventional attenuation-based chest radiographs from a commercial system (A–E) and attenuation-based and dark-field chest radiographs from our prototype dark-field radiography system (F–J) of participants with increasing CT-based visual emphysema scores on the Fleischner scale. A and F, A 64-year-old woman without emphysema. B and G, A 57-year-old man with trace emphysema. C and H, A 54-year-old woman with mild emphysema. D and I, A 71-year-old man with moderate emphysema. E and J, A 73-year-old man with confluent emphysema. Same window and level were applied within all images of each modality. With attenuation-based radiographs emphysema can mainly be detected due to secondary symptoms such as the flattened diaphragm and the barrel-shape of the lung (D, E, I, J). In contrast, the dark-field radiographs are directly sensitive to the lung's alveolar structure. Therefore, the lung with an intact alveolar structure (F) shows a strong dark-field signal, whereas the dark-field signal strength is decreased in emphysematous lungs (H, I, J).

Subjective Image Analysis

Values are given as mean ± standard deviation. The diagnostic confidence of all readers increased significantly (P < 0.001) from 2.78 ± 0.96 for conventional radiographs to 4.30 ± 0.71 for dark-field and attenuation radiographs from the clinical dfCXR system. The interreader agreement for emphysema severity increased from a slight agreement for conventional images (κ = 0.12) to a fair agreement for assessment of images from the clinical dfCXR system (κ = 0.39).

Figure 4A shows the ROC curves for the performance of both imaging methods regarding the detection of at least mild emphysema. The increase in emphysema detection capability from conventional radiography (AUC = 0.73) to dfCXR (AUC = 0.85) was significant (P = 0.005).

FIGURE 4:

Statistical analysis of reader study. The legend applies to all panels. Abbreviations: mod., moderate; confl., confluent; adv. destr., advanced destructive. A, ROC curves and corresponding AUC values for the detection of at least mild emphysema for images from both devices. B and C, A more detailed analysis regarding the staging of emphysema. AUC values indicating significant differentiation of the groups are printed in bold. Although readers were only able to differentiate between trace and mild emphysema based on attenuation images, with use of the dark-field images a differentiation was possible between trace, mild, moderate, and confluent emphysema.

Figure 4B and C and Table 2 provide a more detailed analysis of the reader study regarding the staging of the emphysema. For images from both devices, the readings were not significantly different between absent and trace emphysema. Based on the conventional radiographs (Fig. 4B), readings were different between trace and mild emphysema (AUC = 0.68, P < 0.001), but were not significantly different between mild and moderate emphysema, moderate and confluent emphysema, as well as between confluent and advanced destructive emphysema. When readers had attenuation-based and dark-field radiographs from the clinical dfCXR system (Fig. 4C) available, however, readings were significantly different between trace and mild emphysema (AUC = 0.72, P = 0.002), mild and moderate emphysema (AUC = 0.72, P = 0.011), and moderate and confluent emphysema (AUC = 0.75, P < 0.001). Readings reached the maximum score (“severe”) already for nearly all images of participants with confluent emphysema, so no further differentiation against advanced destructive emphysema was possible.

TABLE 2 - Differentiation of Emphysema Stages Emphysema Severity Absent/Trace Trace/Mild Mild/Moderate Moderate/Confluent Confluent/Adv. Destr. No. participants 36/21 21/11 11/9 9/7 7/4 Conventional radiography AUC = 0.47, P = 0.85 AUC = 0.68, P < 0.001 AUC = 0.62, P = 0.10 AUC = 0.62, P = 0.12 AUC = 0.66, P = 0.13 Dark-field radiography AUC = 0.49, P = 0.55 AUC = 0.72, P = 0.002 AUC = 0.72, P = 0.011 AUC = 0.75, P < 0.001 AUC = 0.54, P = 0.25

Significant entries are in bold font.

AUC, area under the receiver operating characteristic curve; Adv. Destr., advanced destructive.

DISCUSSION

In this study, we compared dfCXR and conventional chest radiography regarding their performance to detect and stage emphysematous changes. Computed tomography examinations were used as the imaging criterion standard. Previous works could demonstrate the capability of dfCXR to visualize emphysematous changes in murine lungs19 and the first application to human COPD patients.24 As shown in a previous work,26 the dark-field signal does not depend on demographic or constitutional patient parameters and is a direct measure for the alveolar density. Other prior studies showed that the quantitative value of the dfCXR signal can also be used to differentiate between emphysematous and healthy lungs27 and found correlations of dfCXR signal with lung diffusion capacity24 and CT-based emphysema quantification.27 However, these studies did not perform a comparison to conventional chest radiography. Here, we present the first comparison of conventional radiographs from a commercial device and dark-field and attenuation radiographs from a clinical dfCXR system for the detection and quantification of emphysema, based on a reader study with CT-based visual emphysema scores as the underlying imaging criterion standard.

The differentiation of the CT-based visual emphysema scores may be of high relevance as it is an independent predictor of subsequent progression of emphysema in participants who are current or former tobacco smokers with and without COPD.33 For early stages, the score is based mainly on small, localized changes. Because of the projection in radiography, these changes in lung density are overlaid with the signal from other tissues in the beam path, such as bones and soft tissue. In contrast to attenuation, the dark-field signal originates from the lung's alveolar structure itself. Other tissues do not generate any signal, and the assessment of structural changes in the lung parenchyma is possible without impeding overlaying signals. This leads to a significantly higher capability for detection of at least mild emphysema with dfCXR (AUC = 0.85) than with conventional radiography (AUC = 0.73, P = 0.005). This is not only of great interest for the early detection of pulmonary emphysema and its predominant corresponding disease COPD, it also highlights the potential of dfCXR for lung imaging in general and for possible screening applications in particular. As recent studies could show,5,6 mortality in smokers is increased when an emphysema is found in CT, even if there is no obstruction in the spirometry (GOLD 0). Thus, an early detection of these patients is crucial to be able to start preventive interventions, including smoking cessation and other risk factor modifications. Another study34 found that mortality is also increased in a general population without a significant smoking history or a clinical disease when emphysema is found in CT. This suggests that early emphysema is not a benign incidental finding. Thus, an early detection of emphysema could also be beneficial in the general population. However, radiation exposure in CT is several times as high as for dark-field imaging. Even for recently reported ultra-low-dose CT using deep learning–based image reconstruction,11 radiation exposure was more than 10 times higher than the reference dose for the imaging at the dark-field scanner (0.39 mSv vs 0.035 mSv). In addition, we assume that radiation dose will be lower in commercially available dark-field scanners compared with our prototype. Given the result of our study that dark-field imaging can detect emphysema in early stages, emphysema could be detected in more patients if a dark-field scanner is used for clinical routine chest x-rays (eg, obtained for preoperative screening, trauma, or pneumonia), making interventions or further diagnosis possible. In addition, follow-up examinations for emphysema could be possible given the low radiation exposure of dark-field imaging.

When the readers had only conventional attenuation-based chest radiographs available, they were able to distinguish between trace and mild emphysema—other adjoining stages could not be differentiated. More severe stages (moderate, confluent, advanced destructive) could be differentiated from trace emphysema. However, patients in advanced stages often already have other symptoms such as shortness of breath or fatigue, limiting the value of attenuation-based radiography to excluding other diagnoses responsible for the symptoms.

As a dark-field imaging system always yields both attenuation-based and dark-field image from a single acquisition, the imaging information from both modalities can be used for image evaluation. For this combined analysis, readers were able to distinguish between the adjoining stages trace/mild, mild/moderate, and moderate/confluent emphysema with the largest effect sizes, expressed as AUCs, respectively. The only stages that could not be differentiated were absent/trace as well as confluent/advanced destructive. As CT-detected trace emphysema could not be differentiated statistically significantly from absent emphysema using dark-field imaging, the sensitivity of dark-field imaging seems to be slightly lower than for CT-based emphysema detection. However, the current study is based on the first patients ever receiving dark-field imaging for emphysema detection. Thus, an increasing sensitivity can be assumed as the radiologists' experience in dark-field imaging improves and upcoming dark-field scanners provide higher image quality.

The current study has some limitations that have to be addressed. The main limitation is the small number of participants, especially in the higher emphysema stages. The results of this study will have to be confirmed with larger cohorts in the future. Another limitation is the restriction to emphysema as the only lung pathology present in the study. Only different stages of emphysema had to be differentiated by the readers. Therefore, the specificity reported in this work is only valid for the comparison of healthy and emphysematous lungs. However, a decreased dark-field signal can also be caused by other pathologies such as pneumonia35 or masses. In these cases, the attenuation-based images can be used to differentiate between emphysema and other causes, as the first leads to reduced density and the latter lead to opacities. Further studies including other pathologies of the lung, such as pneumonia, fibrosis, or pneumothorax, will help to investigate the specificity of dfCXR for differentiation between different lung diseases or conditions. Finally, very early stages of emphysema could be missed using CT. Thus, an evaluation of these stages was not possible in the current study as patients without evidence of emphysema in the CT examination were rated as healthy.

Overall, we present the results from a clinical reader study evaluating dfCXR in comparison to conventional radiography for the imaging of pulmonary emphysema in living humans. We show that dark-field chest radiographs are superior to conventional radiographs for emphysema diagnosis and quantification. Thus, dfCXR could be used as a drop-in replacement for chest radiography in the future, enhancing its diagnostic value. As the effective patient dose is comparable to conventional radiography and much lower than in CT, dfCXR may also be suitable for early detection of emphysema and could be used for long-term treatment or disease progression monitoring.

ACKNOWLEDGMENTS

The authors acknowledge Andre Braunagel, Thomas Pralow, Hanns-Ingo Maack, Hendrik van der Heijden, Klaus-Jürgen Engel, Bernd Lundt, Sven Prevrhal, Karsten Rindt, Roland Proksa, Michael Heider, Pascal Meyer, and Jürgen Mohr for assistance during the hardware and software development for the dfCXR system. Furthermore, Bernhard Haller kindly provided statistical advice for this manuscript.

REFERENCES 1. Decramer M, Janssens W, Miravitlles M. Chronic obstructive pulmonary disease. Lancet. 2012;379:1341–1351. 2. Murray CJL, Vos T, Lozano R, et al. Disability-adjusted life years (DALYs) for 291 diseases and injuries in 21 regions, 1990-2010: a systematic analysis for the global burden of disease study 2010. Lancet. 2012;380:2197–2223. 3. Qaseem A, Wilt TJ, Weinberger SE, et al. Diagnosis and management of stable chronic obstructive pulmonary disease: a clinical practice guideline update from the American College of Physicians, American College of Chest Physicians, American Thoracic Society, and European Respiratory Society. Ann Intern Med. 2011;155:179–191. 4. Vestbo J, Hurd SS, Agustí AG, et al. Global strategy for the diagnosis, management, and prevention of chronic obstructive pulmonary disease. Am J Respir Crit Care Med. 2013;187:347–365. 5. Lynch DA, Moore CM, Wilson C, et al. CT-based visual classification of emphysema: association with mortality in the COPDGene study. Radiology. 2018;288:859–866. 6. Oh AS, Strand M, Pratte K, et al. Visual emphysema at chest CT in GOLD stage 0 cigarette smokers predicts disease progression: results from the COPDGene study. Radiology. 2020;296:641–649. 7. Lynch DA, Austin JHM, Hogg JC, et al. CT-definable subtypes of chronic obstructive pulmonary disease: a statement of the Fleischner Society. Radiology. 2015;277:192–205. 8. Mahesh M, Ansari AJ, Mettler FA Jr. Patient exposure from radiologic and nuclear medicine procedures in the United States and Worldwide: 2009–2018. Radiology. 2023;307:e221263. 9. Gierada DS, Pilgram TK, Whiting BR, et al. Comparison of standard- and low-radiation-dose CT for quantification of emphysema. AJR Am J Roentgenol. 2007;188:42–47. 10. Larke FJ, Kruger RL, Cagnon CH, et al. Estimated radiation dose associated with low-dose chest CT of average-size participants in the National Lung Screening Trial. Am J Roentgenol. 2011;197:1165–1169. 11. Yeom J-A, Kim K-U, Hwang M, et al. Emphysema quantification using ultra-low-dose chest CT: efficacy of deep learning–based image reconstruction. Medicina. 2022;58:939. 12. Mettler FA Jr., Huda W, Yoshizumi TT, et al. Effective doses in radiology and diagnostic nuclear medicine: a catalog. Radiology. 2008;248:254–263. 13. Miniati M, Monti S, Stolk J, et al. Value of chest radiography in phenotyping chronic obstructive pulmonary disease. Eur Respir J. 2008;31:509–515. 14. Pfeiffer F, Bech M, Bunk O, et al. Hard-x-ray dark-field imaging using a grating interferometer. Nat Mater. 2008;7:134–137. 15. Pfeiffer F, Weitkamp T, Bunk O, et al. Phase retrieval and differential phase-contrast imaging with low-brilliance x-ray sources. Nat Phys. 2006;2:258–261. 16. Velroyen A, Yaroshenko A, Hahn D, et al. Grating-based x-ray dark-field computed tomography of living mice. EBioMedicine. 2015;2:1500–1506. 17. Bech M, Tapfer A, Velroyen A, et al. In-vivo dark-field and phase-contrast x-ray imaging. Sci Rep. 2013;3:3209. 18. Hellbach K, Yaroshenko A, Willer K, et al. Facilitated diagnosis of pneumothoraces in newborn mice using x-ray dark-field radiography. Invest Radiol. 2016;51:597–601. 19. Hellbach K, Yaroshenko A, Meinel FG, et al. In vivo dark-field radiography for early diagnosis and staging of pulmonary emphysema. Invest Radiol. 2015;50:430–435. 20. Yaroshenko A, Hellbach K, Yildirim AO, et al. Improved in vivo assessment of pulmonary fibrosis in mice using x-ray dark-field radiography. Sci Rep. 2015;5:17492. 21. Gromann LB, De Marco F, Willer K, et al. In-vivo x-ray dark-field chest radiography of a pig. Sci Rep. 2017;7:4807. 22. Sauter AP, Andrejewski J, De Marco F, et al. Optimization of tube voltage in x-ray dark-field chest radiography. Sci Rep. 2019;9:8699. 23. Fingerle AA, De Marco F, Andrejewski J, et al. Imaging features in post-mortem x-ray dark-field chest radiographs and correlation with conventional x-ray and CT. Eur Radiol Exp. 2019;3:25. 24. Willer K, Fingerle AA, Noichl W, et al. X-ray dark-field chest imaging for detection and quantification of emphysema in patients with chronic obstructive pulmonary disease: a diagnostic accuracy study. Lancet Digit Health. 2021;3:e733–e744. 25. Frank M, Urban T, Willer K, et al. Dosimetry on first clinical dark-field chest radiography. Med Phys. 2021;48:6152–6159. 26. Gassert FT, Urban T, Frank M, et al. X-ray dark-field chest imaging: qualitative and quantitative results in healthy humans. Radiology. 2021;301:389–395. 27. Urban T, Gassert FT, Frank M, et al. Qualitative and quantitative assessment of emphysema using dark-field chest radiography. Radiology. 2022;303:119–127. 28. Harris CR, Millman KJ, van der Walt SJ, et al. Array programming with NumPy. Nature. 2020;585:357–362. 29. Virtanen P, Gommers R, Oliphant TE, et al. SciPy 1.0: fundamental algorithms for scientific computing in Python. Nat Methods. 2020;17:261–272. 30. McKinney W. Data structures for statistical computing in Python. In: van der Walt S, Millman J, eds. Proceedings of the 9th Python in Science Conference. 2010:56–61. 31. Pilson D, Decker KL. Compensation for herbivory in wild sunflower: response to simulated damage by the head-clipping weevil. Ecology. 2002;83:3097–3107. 32. Obuchowski NA. Nonparametric analysis of clustered ROC curve data. Biometrics. 1997;53:567–578. 33. El Kaddouri B, Strand MJ, Baraghoshi D, et al. Fleischner Society visual emphysema CT patterns help predict progression of emphysema in current and former smokers: results from the COPDGene study. Radiology. 2021;298:441–449. 34. Oelsner EC, Carr JJ, Enright PL, et al. Per cent emphysema is associated with respiratory and lung cancer mortality in the general population: a cohort study. Thorax. 2016;71:624–632. 35. Frank M, Gassert FT, Urban T, et al. Dark-field chest x-ray imaging for the assessment of COVID-19-pneumonia. Commun Med. 2022;2:147.