1.

Introduction

Digital tomosynthesis (DTS) of the chest is a relatively modern modality, based on conventional X-ray techniques, which allows superior visualization of anatomical structures and pathological findings in multiple sections that would otherwise be seen in the same projection in a standard posterior-anterior chest X-ray.^1,2 The ability to isolate small structures is ideal for the visualization of lung nodules. A lung nodule is a relatively well-defined approximately rounded opacity measuring up to 3 cm in diameter within the lung parenchyma.³ It is known that nodule size, morphology, nodule location, and nodule growth rate are characteristics associated with the risk of malignancy.⁴ Consequently, nodule size estimate is an important factor in nodule assessment in all major recommended algorithms.^5–7

Lung nodules can be divided into three size categories: small, intermediate, and large. In general, small nodules ($<100{\mathrm{mm}}^3$) can be discharged or followed with longer intervals, intermediate nodules (100 to $250{\mathrm{mm}}^3$) need regular follow-up, and larger nodules ($\ge 300{\mathrm{mm}}^3$) need further assessment.^4,5 The recommended tool for follow-up is computed tomography (CT).^5,6 However, one problem with follow-up CT examinations is that the radiation associated with CT per se is a risk factor for cancer.⁸

DTS has been proposed as an alternative to CT for lung nodule follow-up,⁹ and the technique has been reported to have the potential to become a first-line screening tool for lung cancer.² DTS allows visualization of the lungs in section images with less radiation compared with CT.¹⁰ However, there are some limitations associated with DTS, for instance, limited depth resolution and high sensitivity for artifacts from medical devices. The technique comes with a lower cost when compared with CT, and this could be of importance in centers with limited accessibility to CT scans.¹¹

Lung nodules can be measured with one-dimensional measurements such as diameter, two-dimensional measurements such as area, and three-dimensional measurements such as volume.¹² Measuring the diameter is the most common way to document lung nodule size in a radiological report. The recommendation from the Fleischner Society is to report the mean diameter (mD) based on the longest and its perpendicular diameter of the nodule independently of the plane (axial, coronal, or sagittal).^5,6,13 However, diameter measurements might be unreliable for the assessment of small nodules both on CT¹⁴ and on DTS.⁹

Due to the limitations of one-dimensional measurements, some efforts have been made to incorporate area measurement as a way to provide more information about the lung nodules.¹⁵ However, with the introduction of multiplanar reformations (MPR) of CT images the hot topic in research soon evolved to three-dimensional measurements of lung nodule volume. Some guidelines pointed out volume as the best way to document lung nodules, not only just during the first report but also during follow-up.¹⁶ Two of those are the recommendations from the Fleischner Society and the British Thoracic Society, both recognizing the importance of volumetry for the measurement and management of lung nodules.^5,6 Volumetry is the principal method for measuring solid nodules according to the British Thoracic Society recommendations and the European Lung Cancer Screening position statement.¹⁷ Despite these statements even 3D measurements have multiple limitations. Image acquisition, reconstruction method, CT-slice thickness, and variability between the breath-holds of the patients can influence the precision of volume measurements.¹⁸ A good nodule segmentation is needed to perform volumetry measurements,¹⁶ and the same software should be used when performing automated or semi-automated volumetry both for the initial examination and for the follow-ups.^4,19 Volume cannot be measured on DTS, and consequently, nodule measurements must be performed using diameter or area. To the best of our knowledge, there are no previous studies about area measurement of lung nodules on DTS images of the chest.

If DTS should be considered as a follow-up examination for lung nodules it is important that radiologists are confident in localizing a nodule known from CT in a DTS examination. As far as we are aware, the confidence of radiologists in translating the correspondence of anatomical location of lung nodules on CT to the corresponding nodule on DTS has not been studied before. However, experts in the field have constructed ground truth regarding nodule localization in detection studies.^20–22

The role of CT maximum intensity projections (CT-MIP) has been studied by others,²³ concluding that CT-MIP images allow high diagnostic accuracy in the diagnosis of malignant pulmonary nodules. The best inter-observer agreement has been shown for axial MIP reconstruction.²⁴ However, a comparison between coronal and transaxial CT-MIP regarding their usefulness in the process of localizing a nodule known from CT in a DTS examination has, to the best of our knowledge, not been investigated previously.

Consequently, the aims of the present study were to investigate (1) the preferred orientation of CT-MIP for facilitating localization of lung nodules in DTS, (2) the visibility of nodules in DTS in direct comparison to CT, and (3) which nodule size measurement technique (diameter or area) results in the smallest measurement variability in DTS.

2.

Materials and Methods

2.2.

Study Design and Statistical Analysis

The present study entails three parts. In the first part, the preferred orientation of CT-MIP when localizing a nodule in DTS was evaluated. The second part assessed nodule localization and visibility in DTS compared with CT, and the third part concerned nodule size estimated by diameter and area measurements. Five radiologists with different levels of experience participated as observers in the study. Observers 1 and 2 were senior cardiothoracic consultant radiologists each with over 25 years of experience. Observer 3 was a senior cardiothoracic consultant radiologist with over 10 years of experience. Observers 4 and 5 were radiologists each with $\sim 5$ years of experience. A medical doctor with a few months of training in radiology (observer 6) participated in the study, but the measurements of this observer were not compared with the other observers. The image viewer ViewDEX^30–33 was used for image visualization and data collection, and all cases were evaluated in a randomized order, unique for each observer. An illustration of the image display used in the study is presented in Fig. 1. The study consisted of two review sessions, described in detail below. For each review session, the observers reviewed the cases using two monitors. The first monitor was divided into four canvases displaying transverse CT images, coronal CT images, CAD results (showing the location and size of the largest nodule in the case), and CT-MIP images (transaxial or coronal in review session 1 and only coronal in review session 2, for details, see below). The second monitor displayed the corresponding DTS.

Fig. 1

Illustration of the image display used in the study. The observers were instructed to select the largest nodule in the CAD series (c) and to mark and localize the nodule in all CT series and DTS (e) images using the transaxial (a), the coronal MPR (b), and the CT-MIP (d) series as reference. In the first review session, a coronal or transaxial MIP series were shown to the observers in a unique randomized order. Abbreviations; CT = computed tomography, CAD = computer-aided detection, MIP = maximum intensity projection, and DTS = digital tomosynthesis.

3.

Results

3.2.

Part 2, Nodule Localization and Visibility of Nodules

In most cases (79%), the observers were confident or somewhat confident that their marked nodule on DTS corresponded to their nodule marked on CT, as shown in Fig. 2. The markings were confirmed by coordinates in ViewDEX, and only one nodule was wrongly marked on DTS by one of the observers. One observer could not identify the nodule on DTS in two cases. Hence, only 21 of the 24 nodules were included in the further analysis.

Fig. 2

Results of the question “confident that marking in DTS corresponds to the marking CT.” 1 = Confident that the statement is true. 2 = Somewhat confident that the statement is true. 3 = I do not know if the statement is true or not. 4 = Somewhat confident that the statement is not true. 5 = Confident that the statement is not true. Abbreviations; DTS = digital tomosynthesis and CT = computed tomography.

The vast majority of the observers considered that the nodules were better visualized on coronal MPR CT images than in DTS (Fig. 3). However, one of the experienced observers considered five nodules to be somewhat better visualized on DTS in comparison to CT.

Fig. 3

Results of the question “The visibility of the nodule in MPR COR and DTS.” 1 = Definitely better in MPR COR. 2 = Somewhat better in MPR COR. 3 = Equally good in MPR COR and DTS. 4 = Somewhat better in DTS. 5 = Definitely better in DTS. Abbreviations; MPR = multiplanar reformation, COR= coronal, and DTS = digital tomosynthesis.

3.3.

Part 3, Nodule Measurement

Three nodules out of 24 were not considered measurable by all the observers. Consequently, diameter and area measurements of 21 nodules performed by five observers were included in the analysis. An illustration of the measurement of a representative nodule is given in Fig. 4.

Fig. 4

Illustration of the same nodule on CT and DTS (one case out of 24). (a) The selected nodule on a CAD axial view. (b) The selected nodule on DTS (2 diameter measurements used to calculate the mD in red). (c) The selected nodule on a coronal MIP. (d) The selected nodule on DTS (drawn area in red). (e) The selected nodule on DTS. Abbreviations; CT = computed tomography, DTS = digital tomosynthesis, CAD = computer-aided detection, mD = mean diameter, and MIP = maximum intensity projection.

Concerning linear measurement, there were no statistically significant differences between measuring the mD ($p=0.40$) and the area ($p=0.68$) on DTS or CT, respectively.

Scatter plots showing the nodule volumes based on mD measurements and area measurements compared with the nodule volumes obtained by CT-CAD are presented in Fig. 5.

Fig. 5

Scatter plots of the lung nodule volume estimate on DTS of the chest based on mD-based nodule volume (a) and area-based nodule volume (b) of the five observers in comparison to CAD-derived nodule volumes. Scatter plots of the lung nodule volume estimate on DTS of the chest based on mD-based nodule volume (average of the five observers) with a SEE 42,7 (c), and area-based nodule volume (average of the five observers) with a SEE 38,7 (d) of the five observers in comparison to CAD-derived nodule volumes and respective linear regressions. Abbreviations; DTS = digital tomosynthesis, mD = mean diameter, CAD = computer-aided detection, SEE = standard error of the estimate, and $R^2$ = coefficient of determination.

When assessing the inter-observer agreement for volumes based on mD measurements, the mean difference ranged from $-23.7$ ($\mathrm{CI}-40.2$ to $-7.2$; $\mathrm{LoA}-101.0$ to 53.6) ${\mathrm{mm}}^3$ for the best inter-observer agreement to 65.1 ($\mathrm{CI}-4.4$ to 134.6; $\mathrm{LoA}-259.4$ to 389.7) ${\mathrm{mm}}^3$ for the worst inter-observer agreement. When assessing the corresponding inter-observer agreement for volumes based on area measurements, the mean difference ranged from $-12.7$ ($\mathrm{CI}-28.2$ to 2.8; LoA $-85.2$ to 59.8) ${\mathrm{mm}}^3$ for the best inter-observer agreement to 38.1 ($\mathrm{CI}-12.1$ to 88.2; LoA $-196.4$ to 272.5) ${\mathrm{mm}}^3$ for the worst inter-observer agreement. Bland-Altman plots showing the agreement are presented in Fig. 6.

Fig. 6

Bland-Altman plots of the best (a) and (b) and worst (c) and (d) inter-observer agreement when assessing measurements of mD-based nodule volume and area-based nodule volume on DTS of the chest. Abbreviations: mD = mean diameter and DTS = digital tomosynthesis.

The performed Wilcoxon signed-rank test showed that the measurement variability was significantly larger when calculating nodule volume based on mD measurements in comparison to nodule volume based on area measurements ($p<0.05$).

Three observers (observers 1, 4, and 5) measured all nodules twice (review session 1 and review session 2). These data were used to assess the intra-observer agreement of the measurements. For nodule volumes based on mD measurements, the mean difference in intra-observer agreement ranged from 4.6 ($\mathrm{CI}-11.2$ to 20.4; $\mathrm{LoA}-69.5$ to 78.7) ${\mathrm{mm}}^3$ for the best intra-observer agreement to $-38.0$ ($\mathrm{CI}-67.7$ to $-8.2$; $\mathrm{LoA}-177$.1 to 101.2) ${\mathrm{mm}}^3$ for the worst intra-observer agreement. When assessing the corresponding intra-observer agreement for nodule volumes based on area measurements, the mean difference ranged from $-1.3$ ($\mathrm{CI}-10.4$ to 7.7; $\mathrm{LoA}-43.8$ to 41.1) ${\mathrm{mm}}^3$ for the best intra-observer agreement to $-4.0$ ($\mathrm{CI}-35.2$ to 27.2; LoA $-150.0$ to $142.0{\mathrm{mm}}^3$ for the worst intra-observer agreement. Bland-Altman plots showing the agreement are presented in Fig. 7.

Fig. 7

Bland-Altman plots of the best (a) and (b) and worst (c) and (d) intra-observer agreement when assessing measurements of mD-based nodule volume and area-based nodule volume on DTS of the chest. Abbreviations: mD = mean diameter and DTS = digital tomosynthesis.

4.

Discussion

The findings in the present study indicate that coronal CT-MIPs are the preferred orientation when localizing lung nodules known from CT in DTS images, that most nodules detected with CT are visible in DTS, and that measurements of nodule area on DTS result in less measurement variability than measurements of nodule diameter.

In previous studies, CT-MIP have been shown to be a useful tool to help radiologists detect lung nodules on CT.^44–47 However, further research on the preferred orientation of the CT-MIP was needed to specifically address the usefulness of CT-MIP in the process of identifying nodules on DTS. In this study, coronal CT-MIP images were considered to facilitate the localization of the nodule in DTS and were rated by the observers as superior to transaxial CT-MIP images, at a strong statistical significance level.

Concerning the localization and visibility of lung nodules on DTS, the observers were highly confident in their ability to localize the same nodules on DTS and on CT (see Fig. 2), even though the visibility of the nodules was rated higher in the CT images than in the DTS images (see Fig. 3). Most of the lung nodules could be identified on DTS regardless of observer experience. These results are in line with the results from other studies. For example, in a study by Lee et al.,⁴⁸ focusing on lung nodule visibility on DTS and learning curves associated with DTS reading times, it was shown that most malignant nodules $>5\mathrm{mm}$ were visible on DTS.

The fact that five nodules were considered somewhat better visualized on DTS when compared with coronal CT could possibly be explained by their more central location in the coronal plane on DTS. It is known that DTS presents better spatial resolution in the $x-y$ plane (coronal plane) than CT.⁴⁹ This benefit may however be obscured by blurring from out-of-plane objects caused by the limited depth resolution in reconstructed DTS images. According to Machida et al.,⁵⁰ the blurring effect increases if the structures are denser and located perpendicular to the sweep angle and decreases with increasing distance from the dense object. Consequently, the visualization of nodules with a more central location in the reconstructed volume may be less affected by disturbing blurring from, e.g., ribs compared with nodules located close to the rib cage and therefore benefit more from the higher in-plane resolution in DTS compared with CT.

It has previously been suggested that DTS might have the potential to be a modality that could be used in lung cancer screening⁵¹ as well as for lung nodule follow-up.⁹ The benefits of DTS in terms of lower radiation and lower cost when compared with CT could promote the use of DTS as an alternative screening method.⁵² The recent European policy statement encourages countries to implement lung cancer screening,⁵³ and low-dose CT is the primary diagnostic method for lung cancer screening.⁵⁴ However, Bertolaccini et al. suggested that DTS could be used as the first-line lung cancer screening modality,⁵² which could be an alternative if CT resources are scarce.

Determination of nodule size is crucial in lung cancer screening. Previous studies have shown that measurement variability is a concern in DTS.⁹ Nodule volume is the preferred size estimate in lung cancer screening as well as for lung nodule follow-up.¹⁶ Unfortunately, volume cannot be measured on DTS. Measurements of diameter and or area should therefore be considered. In a study by Söderman et al.,⁵⁵ it was shown that there is a risk for underestimation of lung nodule size when lung nodule diameter is manually measured, especially if in-plane artifacts are present in the images. In-plane artifacts are perceived as a darker rim around for example a lung nodule, related to the direction of the scan. As the artifact is more prominent in the scan direction, it can be anticipated that the effect of the artifact on nodule size determination is larger for diameter measurements than for area measurements. Consequently, area-based measurements may be a way to decrease the measurement variability. This hypothesis is supported by the results of the current study. It should be noted that in the case of nodule follow-up, it is recommended by RECIST guidelines⁵⁶ that the same observer measures the nodule using the same modality and measurement technique. This is a well-known method to avoid inter-modality and inter-observer variability.

Even though measurements of nodule size in CT and DTS cannot be used interchangeably,⁵⁷ it might still be possible to use DTS for follow-up of nodules detected on DTS in a screening scenario. It has been shown that manual measurements on DTS have comparable repeatability when compared with manual measurements on CT and that no bias existed between the two modalities.⁵⁷ Most nodules in the present study were rated as visible in DTS in contrast to the study by Lee et al.⁴⁸ where only 53.3% of the nodules were reported as visible on DTS. The results in the present study probably reflect the inclusion of larger nodules in comparison to the previously mentioned study⁴⁸ and the access to CT images used as a reference for nodule location. Even if all the observers considered that the nodules were better visualized on coronal MPR CT images compared with DTS, some nodules were at least equally visible on DTS. Regarding the three nodules that were not measured by all the observers, one observer measured another nodule positioned close to the nodule measured by the other four observers and one out of five observers considered two nodules non-measurable, which might be due to lesser contrast in the DTS images in comparison to CT. However, the reason is at the discretion of this observer.

It should also be noted that factors such as different patient positionings, nodule sizes, and dose levels during different scan occasions (baseline and follow-ups) could interfere with the precision of the measurements performed on DTS, as described by Söderman et al.⁵⁸

The present study aims at investigating nodule size estimates in a situation that resembles clinical praxis. Due to the limited depth resolution in DTS, there is little relevance in reconstructing section images using an interval smaller than 5 mm, whereas CT images frequently are reconstructed with a slice thickness of 3 mm with a 2-mm interval. Nevertheless, this limited depth resolution on DTS will affect the size measurements.

The present study had the strength of having a decent number of observers, and the examinations were carried out on the same day and on the same equipment. Furthermore, the present study benefited from having observers with different experiences in the field of radiology, with a range of experience extending from just some months of training in radiology to many years of experience in thoracic radiology. Our results indicate that even an observer with minor experience in the field of radiology can perform measurements on DTS comparable to radiologists. In addition, this study benefited from the fact the images were visualized with anonymization of the cases and in a unique randomized order for each observer. However, there were also several limitations in the present study. The numbers regarding both cases and nodules were low, and consequently, the number of nodules in each size category was limited. Only solid lung nodules were included, and the measuring tool used in the present study was new to the observers. Also, even if our study shows that most intermediate or large-size nodules (nodules greater than or equal to $100{\mathrm{mm}}^3$) that were found on CT could be visualized, accurately localized, and correctly measured in DTS, the assumption made in the present study that lung nodules are sphere-like structures will introduce an uncertainty in the resulting estimated nodule volumes. Using this assumption, even small differences in measured diameters and areas might result in considerable differences in resulting nodule volume. For example, if one observer measures the diameter of a specific nodule in DTS to be 6 mm and a different observer measures the same nodule diameter to be 7 mm, the assumption that the lung nodules are sphere-like structures results in estimated nodule volumes of 113 and $179.6{\mathrm{mm}}^3$ for the 6-mm-diameter nodule and the 7-mm-diameter nodule, respectively. Hence, there will be a large difference in resulting nodule volume despite only a 1-mm difference in measured nodule diameter. Using area-based measurements instead of diameter-based measurements for the estimation of the nodule volumes in DTS improves the agreement of the estimated nodule volumes between DTS and CT, but even with area-based volume estimations on DTS, the method might still introduce considerable differences in nodule volumes between DTS and CT.

To summarize, assessing lung nodules with DTS would be a way to reduce patient exposure to radiation and the costs associated with lung nodule assessment. Estimating lung nodule size through area measurements at screening and follow-up could be a key to reduce inter-observer variation on DTS. The results of the present study will be further explored in a larger cohort of participants with lung nodules in SCAPIS.

5.

Conclusions

The present study shows that coronal CT-MIP images are superior to transaxial CT-MIP images in facilitating lung nodule localization in DTS. Furthermore, most nodules with a size larger than $100{\mathrm{mm}}^3$ found on CT can be visualized, correctly localized, and measured in DTS. Finally, when assessing nodule size in DTS, area measurements might result in less variability in comparison to diameter measurements.