European Journal of Radiology 84 (2015) 1383–1391
Contents lists available at ScienceDirect
European Journal of Radiology journal homepage: www.elsevier.com/locate/ejrad
Role of digital tomosynthesis and dual energy subtraction digital radiography in detecting pulmonary nodules Sarvana G. Kumar b , Mandeep Kumar Garg b,∗ , Niranjan Khandelwal b , Pankaj Gupta b , Dheeraj Gupta a , Ashutosh Nath Aggarwal a , Subash Chand Bansal b a b
Department of Pulmonary Medicine, Postgraduate Institute of Medical Education and Research (PGIMER), Sector 12, Chandigarh 160012, India Department of Radiodiagnosis and Imaging, Postgraduate Institute of Medical Education and Research (PGIMER), Sector 12, Chandigarh 160012, India
a r t i c l e
i n f o
Article history: Received 16 December 2014 Received in revised form 11 March 2015 Accepted 14 March 2015 Keywords: Digital radiography Digital tomosynthesis Dual energy subtraction radiography Pulmonary nodule
a b s t r a c t Objective: Digital tomosynthesis (DT) and dual-energy subtraction digital radiography (DES-DR) are known to perform better than conventional radiography in the detection of pulmonary nodules. Yet the comparative diagnostic performances of DT, DES-DR and digital radiography (DR) is not known. The present study compares the diagnostic performances of DT, DES-DR and DR in detecting pulmonary nodules. Subjects and methods: The institutional Review Board approved the study and informed written consent was obtained. Fifty-five patients (30 with pulmonary nodules, 25 with non-nodular focal chest pathology) were included in the study. DT and DES-DR were performed within14 days of MDCT. Composite images acquired at high kVp as part of DES-DR were used as DR images. Images were analyzed for presence of nodules and calcification in nodules. Interpretations were assigned confidence levels from 1 to 5 according to Five-Point rating scale. Areas under the receiver operating characteristic curves were compared using Z test. Results: A total of 110 (88 non-calcified, 22 calcified) nodules were identified on MDCT. For detection of nodules, DR showed cumulative sensitivity and specificity of 25.45% and 67.97%, respectively. DT showed a cumulative sensitivity and specificity of 60.9% and 85.07%, respectively. The performance was significantly better than DR (p < 0.003). DES-DR showed sensitivity and specificity of 27.75% and 82.64%, not statistically different from those of DR (p—0.92). In detection of calcification, there was no statistically significant difference between DT, DES-DR and DR. Conclusions: DT performs significantly better than DES-DR and DR at the cost of moderate increase in radiation dose. © 2015 Published by Elsevier Ireland Ltd.
1. Introduction The detection and characterization of pulmonary nodules has always been one of the most daunting tasks in chest imaging and up to 30% of such lesions are missed even by experienced chest radiologists while interpreting conventional chest radiographs (CXR) [1,2]. Though it seems easier to identify nodules retrospectively while reviewing previous images of patients with known lung nodules; the prospective diagnosis remains elusive. Considering the attenuation characteristics, nodules as small as 3 mm should ideally be identified on a chest radiograph, however in practice, reliable
∗ Corresponding author. Tel.: +91 172 2756380; fax: +91 172 2744401. E-mail address: [email protected] (M.K. Garg). http://dx.doi.org/10.1016/j.ejrad.2015.03.020 0720-048X/© 2015 Published by Elsevier Ireland Ltd.
detection of nodules is often not possible until they reach a diameter of at least 8 mm. This limitation makes the detection of smaller nodules very difficult. That is why chest radiograph is not recommended as a screening tool for lung cancer, and indeed has a very low sensitivity for detecting solitary pulmonary nodules [3]. It is not very sensitive in detecting calcification within a pulmonary nodule either, demonstrating a sensitivity of 50% and a specificity of 87% [4]. The current standard for detection and characterization of pulmonary nodules is MDCT. However, MDCT cannot replace CXR for general thoracic imaging due to its high cost, workflow issues and radiation exposure. Thus, the radiographic technique that can improve the detection of pulmonary lesions without causing considerable increase in the radiation dose or cost could be very useful in day-to-day practice. In this context, the current study was
1384
S.G. Kumar et al. / European Journal of Radiology 84 (2015) 1383–1391
performed to compare the diagnostic performances of DT, DES-DR and DR in detecting pulmonary nodules with MDCT as reference standard.
2. Subjects and methods Institutional review board approval was obtained (Institute Ethics Committee, 8714/PG-2Trg/2011), and all patients gave written in-formed consent. Between January 2012 and June 2013, a total of 55 patients (30 patients with pulmonary nodules and 25 patients without pulmonary nodules) were prospectively and consecutively enrolled in the study by a radiologist not later involved in reading the radiographic studies. Patients who had undergone MDCT within the last 2 weeks and had one or more pulmonary nodules were included as study subjects. Patients who had undergone MDCT within the last 2 weeks and showed normal lung or focal lung pathology other than nodules (such as small patch of consolidation, mild and focal fibrosis or mild pleural effusion) were included as control subjects. Patients less than 18 years of age, pregnant women, patients not able to follow instructions and hold breath for 10 s, MDCT showing features of severe diffuse pulmonary disease (e.g., idiopathic pulmonary fibrosis, chronic pulmonary emphysema, diffuse consolidation, severe cardiac congestion, bronchiectasis etc.) or large pleural effusion masking more than 2/3 of either lung field, Patients who had undergone fine needle aspiration/biopsy or surgical intervention for the lung lesion were excluded. MDCT of chest was taken to be the gold standard in this study. MDCT was done on a 64-detector array CT system (Lightspeed VCT, GE Healthcare). The parameters were: detector collimation 64 × 0.625 mm; helical pitch 1.375; rotation time 0.6 s; tube voltage 120 kVp; AEC controlled tube current; 5 mm section thickness; and continuous reconstruction at 0.625 mm. The images were analysed in a work station (Adwantage Volumeshare 5/AW 4.6, GE Healthcare). The entire volumetric data were analysed in axial plane for nodule detection and then in coronal plane for correlation with DR/DT images. Size of the nodule was determined by the average of its length (longest nodule diameter) and width (widest diameter perpendicular to length) [5] in lung window settings (Level—600 Hounsfield unit (HU), Width 1600 HU). A nodule was considered calcified if calcification was visually detected in mediastinal window settings (Level + 40, Width 400) [4] and/or its attenuation value is greater than 200 HU [6]. DES-DR was performed on a flat panel detector system (Definium 8000, GE Healthcare). Erect postero-anterior chest radiographs were acquired with respiration suspended in deep inspiration using the standard dual shot method (120 kVp and 60 kVp) with automatic exposure control determined tube load. Routine DR images without subtraction and subtracted soft tissue and bone images were generated. DT involves acquisition of projection data over a limited range of X-ray tube movement. This allows reconstruction of data in multiple planes. It offers improved diagnostic performance over conventional radiography by decreasing the overlap of anatomical structures. Compared to CT, DT is cost effective and involved low radiation exposure. However, there is decreased depth resolution compared to CT. All DT examinations were performed immediately after DES-DR on the same system using the volumeRAD option. Erect posteroanterior scout view was taken with 120 kV peak tube potential and automatic exposure control determined tube load. Then this load used for the scout was multiplied by a user-adjustable dose ratio (10:1), and was equally distributed over all tomosynthesis projections (41/51/61 depending on patient thickness). Then for each projection the resulting tube load was rounded off to the
Table 1 study population characteristics. Characteristics
No. of patients
Patients with pulmonary nodules Control
30 25
Age (years) Range Mean
19–80 45.5
Sex Male Female
29 26
Nodule size (total no. of nodules = 110) ≤4 mm 5–6 mm 7–8 mm ≥9 mm
46 25 15 24
Calcification (total no. of nodules = 110) Present Absent
22 88
closest tube load setting possible, keeping a minimum tube load of 0.25 mAs per projection to maintain image quality. This process was done automatically by the VolumeRad software of the system. DT images were acquired with a tube sweep angle of approximately ±15◦ and a stationary detector. The breath hold time was 11 s. In all patients raw data were automatically processed to give a stack of 53 images with 5 mm interval. The DR and subtracted images of DES-DR and the images of DT were analysed independently by two radiologists (with 30 and 15 years’ experience, respectively, in thoracic radiology and 1 year experience in interpreting DT images) using the same commercial viewer (RadiAnt-64 bit—1.8.8). The manipulation of windowing, zoom/pan parameters was allowed. Number of pulmonary nodules in each patient with size and calcification status of each nodule was noted. Nodules were tagged with number and stored for final confirmation on CT. Confidence level from 1 to 5, according to fivepoint rating scale of ROC studies was assigned to detection of each nodule and to calcification in each nodule. DR, DES-DR and DT image interpretations in that order were separated by a time interval of one month and cases were presented for reading in different random order. MDCT images were interpreted for confirmation of nodules by both readers by consensus. 3. Statistical analysis Keeping MDCT as the reference standard, sensitivity, specificity and predictive values of DR, DES-DR and DT in detecting pulmonary nodules and calcification in pulmonary nodules were calculated separately for the two observers. ROC curves were generated using confidence level according to five-point rating scale of ROC studies and area under ROC curves (AUC) were calculated separately for the two observers. Kappa test of agreement was used to assess the inter-observer agreement for detection of nodules. All calculations were performed using SPSS® version 17 (Statistical Packages for the Social Sciences). AUC for different modalities were compared using z test. All statistical tests were two-sided and performed at a significance level of ˛ = 0.05. 4. Radiation dose The average dose area product (DAP) for DT was 0.18 Gy cm2 . Using a conversion factor [7] of 0.26 mSv Gy−1 cm−2 , the average effective dose was calculated to be 0.05 mSv. The DAP for DES-DR 0.07 Gy cm2 and the calculated average effective dose was 0.01 mSv. The average effective dose for MDCT was 6.9 mSv.
6. Discussion The detection of pulmonary nodules is hampered by summation, quantum mottle and anatomical noise on conventional chest radiography. It has been shown by Samei and colleagues [8] that
28.18% 80.49% 79.49% 29.46% 0.521 (0.420–0.622)
DES-DR DR
26.36% 70.73% 70.73% 26.36% 0.515 (0.410–0.619) 60% 84.78% 90.41% 46.99% 0.738 (0.658–0.818)
DT DES-DR
27.27% 84.78% 81.08% 32.77% 0.554 (0.457–0.650) 24.54% 65.22% 62.79% 26.55% 0.546 (0.449–0.644)
DR
Observer 1
Table 2 Comparison of performance in detection of pulmonary nodules.
A total of 55 patients were enrolled in the study. Thirty patients had one or more pulmonary nodules on MDCT, while 25 patients were controls. The mean age of patients was 45.56 years with age range of 19–80 years and standard deviation of 16.43 years. There was an equal sex distribution. Of the total 55 patients 29 were male and 26 were female. Among patients with nodules, 2 had pathologically proven bronchogenic carcinoma and 16 had pathologically proven malignant tumor elsewhere. Among patients without nodules 12 had a pathologically proven malignant tumor elsewhere. A total of 110 nodules were identified in 30 patients. All were solid nodules. Number of nodules in each patient ranged from 1 to 14. Patient characteristics are shown in Table 1. The nodules were classified according to their size into 4 groups (≤4 mm, 5–6 mm. 7–8 mm, ≥9 mm). Grouping into such categories (Graph 1) was followed in accordance with the recent Fleischner society guidelines for follow up of small pulmonary nodules detected on CT [5]. Distribution of nodules into size categories is shown in Fig. 1. Of the 110 nodules, 22 were calcified). There were 3 non-nodular calcifications (sclerotic focal lesions in ribs). Diagnosis of true nodules with confidence ≥3 was considered true positive. Diagnosis of nodule with confidence ≥3 which was not confirmed in MDCT was considered false positive. Such false positive nodules in DES-DR and DT and the number of truly nodule free cases were counted as true negatives from which sensitivity, specificity and predictive values were calculated. ROC curves were plotted separately for the two observers. Similarly detection of calcification with confidence ≥3 in a calcified nodule was considered true positive. Detection of calcification with confidence ≥3 in a non-calcified nodule was considered false positive. Non-calcified nodules, false positive nodule localisations in DES-DR and DT and the control group (patients without nodules) served as true negatives from which sensitivity, specificity and predictive values were calculated (Tables 2 and 3). ROC curves (Graphs 2 and 3) were plotted separately for the two observers (Graph 4). . Combined Sensitivities for detection of nodules for both observers according to size of nodules are given in Table 4. Diagnostic utility of DT compared to DES-DR and DR is shown in Figs. 1–3.
Observer 2
5. Results
Sensitivity Specificity Positive Predictive Value Negative Predictive Value ROC-AUC (95% CI in parenthesis)
Graph 1. Distribution of nodules into size categories.
61.82% 85.37% 90.89% 45.45% 0.736 (0.650–0.821)
1385
DT
S.G. Kumar et al. / European Journal of Radiology 84 (2015) 1383–1391
1386
S.G. Kumar et al. / European Journal of Radiology 84 (2015) 1383–1391
Fig. 1. A case of bronchogenic carcinoma in right lung (arrows in A–D). DT and the corresponding CT images (E, F) show multiple additional nodules (short arrows) in ipsilateral lung parenchyma. This information has a bearing on staging of bronchogenic carcinoma.
anatomical noise becomes the bigger limiting factor than random noise in the detection of small pulmonary nodules. However DESDR and DT can efficiently reduce or eliminate this noise caused by the ribs projecting over the lungs. Yet, the detection of pulmonary nodules by DES-DR has not shown to be uniformly better (as in the present study). This is partly due to the fact that as the lesion size becomes smaller, the advantage of anatomical noise reduction is negated. This is inferred from studies by Kashani et al. [9], and Ruhl et al. [10], who found no statistically significant difference in detection of pulmonary nodules (based on their size) by addition of DES-DR. Similarly, DT images are affected by the inherent tomographic noise that is related to the limited angle reconstruction involved in this technique. However, technological advancements
[11,12] have eliminated this effect on the image quality of DT. Thus, despite few limitations, both DES-DR and DT is a subject of considerable interest in detection of pulmonary nodules. DES-DR and DT can be performed in the same DR suite, provided it has the required modifications in the hardware set-up and additional image processing software. They impart radiation doses comparable to or slightly higher than DR [7–11]. There are many published studies on dual energy techniques [12–17] which have given discrepant and conflicting results regarding the role of these techniques in routine clinical practice. There are some published studies on tomosynthesis technique [18–20] and they have all shown that DT performs significantly better than DR in detecting pulmonary nodules. One phantom study
S.G. Kumar et al. / European Journal of Radiology 84 (2015) 1383–1391
1387
Graph 2. ROC curves for detection of nodules. (a, b, c—detection of nodules on DR, DES-SR, DT, respectively, by Observer 1. d, e, f—detection of nodules on DR, DES-SR, DT, respectively, by Observer 2).
1388
S.G. Kumar et al. / European Journal of Radiology 84 (2015) 1383–1391
Graph 3. ROC curves for detection of calcification. (a, b, c—detection of calcification on DR, DES-SR, DT, respectively, by Observer 1. d, e, f—detection of calcification on DR, DES-SR, DT, respectively, by Observer 2).
S.G. Kumar et al. / European Journal of Radiology 84 (2015) 1383–1391
1389
Fig. 2. DT (A) and CT (B) showed a large nodule (arrow) in right lung base which was missed on both DR (C) and DES-DR (D) prospectively. Also note the better depiction of ribs (thin arrows) and the vessels by DT (arrowhead).
[21] has been performed to compare DES-DR with DT, in which DT performed significantly better. In our study the overall cumulative sensitivities of DR, DES-DR and DT in detection of pulmonary nodules were 25.45%, 27.75% and 60.9%, respectively, with specificities of 68.9%, 82.64% and 85.07%,
respectively. The low overall sensitivity can be explained by the overwhelming number of small pulmonary nodules (≤8 mm—86 out of a total of 110). For nodules ≥ 9 mm which, according to the recent guidelines for management of small pulmonary nodules detected on MDCT scans from Fleischner society, require
Fig. 3. DT (A) detected tiny nodule (arrow) in right lung that was missed by DES-DR (B).
DT
52% 99.21% 92.86% 91.24% 0.651 (0.502–0.800) 32% 99.21% 88.89% 88.03% 0.613 (0.474–0.751)
DR
48% 98.47% 85.71% 90.84% 0.648 (0.498–0.798)
DT DES-DR
36% 99.24% 90% 89.04% 0.696 (0.563–0.829) 25% 97.7% 72.72% 88.28% 0.590 (0.457–0.724)
DR
Table 3 Comparison of performance in detection of calcification in pulmonary nodules.
Sensitivity Specificity Positive Predictive Value Negative Predictive Value ROC-AUC (95% CI in parenthesis)
Observer 2 Observer 1
36% 99.24% 90% 88.65% 0.696 (0.562–0.829)
S.G. Kumar et al. / European Journal of Radiology 84 (2015) 1383–1391
DES-DR
1390
Graph 4. Combined Sensitivities for detection of nodules for both observers according to size of nodules.
further investigations (PET/biopsy) to rule out malignancy, the cumulative sensitivities were 54%, 62.5% and 95.8%, respectively. However, none of the three modalities tested here can be used in par with MDCT (gold standard) in for screening or confirmation of pulmonary nodules in view of overall low sensitivity and only moderate specificity. Though DES-DR showed a slight improvement over DR in terms of sensitivity (particularly for nodules larger than 8 mm), there was no statistically significant improvement in overall diagnostic accuracy (p1 = 0.91, p2 = 0.93). This result is in agreement with the previously published studies by Kashani et al. [9] and Ruhl et al. [10], both of which had used dual-shot DES-DR technique as in this study. The former was a single centre study with 129 patients with 158 nodules and there was no significant difference in ROC performance between DE and DR (AUC DES-DR = 0.795 AUC DR = 0.789; p = 0.696). The latter was an international multicentric trial with 149 nodules and there was no significant difference in ROC performance between DE and DR (AUC DES-DR = 0.602 & AUC DR = 0.631; p = 0.4). On the contrary, Szucs et al. [16], have reported a statistically significant improvement with single shot DES-DR technique. However, they had noted that detection of nodules
Contents lists available at ScienceDirect
European Journal of Radiology journal homepage: www.elsevier.com/locate/ejrad
Role of digital tomosynthesis and dual energy subtraction digital radiography in detecting pulmonary nodules Sarvana G. Kumar b , Mandeep Kumar Garg b,∗ , Niranjan Khandelwal b , Pankaj Gupta b , Dheeraj Gupta a , Ashutosh Nath Aggarwal a , Subash Chand Bansal b a b
Department of Pulmonary Medicine, Postgraduate Institute of Medical Education and Research (PGIMER), Sector 12, Chandigarh 160012, India Department of Radiodiagnosis and Imaging, Postgraduate Institute of Medical Education and Research (PGIMER), Sector 12, Chandigarh 160012, India
a r t i c l e
i n f o
Article history: Received 16 December 2014 Received in revised form 11 March 2015 Accepted 14 March 2015 Keywords: Digital radiography Digital tomosynthesis Dual energy subtraction radiography Pulmonary nodule
a b s t r a c t Objective: Digital tomosynthesis (DT) and dual-energy subtraction digital radiography (DES-DR) are known to perform better than conventional radiography in the detection of pulmonary nodules. Yet the comparative diagnostic performances of DT, DES-DR and digital radiography (DR) is not known. The present study compares the diagnostic performances of DT, DES-DR and DR in detecting pulmonary nodules. Subjects and methods: The institutional Review Board approved the study and informed written consent was obtained. Fifty-five patients (30 with pulmonary nodules, 25 with non-nodular focal chest pathology) were included in the study. DT and DES-DR were performed within14 days of MDCT. Composite images acquired at high kVp as part of DES-DR were used as DR images. Images were analyzed for presence of nodules and calcification in nodules. Interpretations were assigned confidence levels from 1 to 5 according to Five-Point rating scale. Areas under the receiver operating characteristic curves were compared using Z test. Results: A total of 110 (88 non-calcified, 22 calcified) nodules were identified on MDCT. For detection of nodules, DR showed cumulative sensitivity and specificity of 25.45% and 67.97%, respectively. DT showed a cumulative sensitivity and specificity of 60.9% and 85.07%, respectively. The performance was significantly better than DR (p < 0.003). DES-DR showed sensitivity and specificity of 27.75% and 82.64%, not statistically different from those of DR (p—0.92). In detection of calcification, there was no statistically significant difference between DT, DES-DR and DR. Conclusions: DT performs significantly better than DES-DR and DR at the cost of moderate increase in radiation dose. © 2015 Published by Elsevier Ireland Ltd.
1. Introduction The detection and characterization of pulmonary nodules has always been one of the most daunting tasks in chest imaging and up to 30% of such lesions are missed even by experienced chest radiologists while interpreting conventional chest radiographs (CXR) [1,2]. Though it seems easier to identify nodules retrospectively while reviewing previous images of patients with known lung nodules; the prospective diagnosis remains elusive. Considering the attenuation characteristics, nodules as small as 3 mm should ideally be identified on a chest radiograph, however in practice, reliable
∗ Corresponding author. Tel.: +91 172 2756380; fax: +91 172 2744401. E-mail address: [email protected] (M.K. Garg). http://dx.doi.org/10.1016/j.ejrad.2015.03.020 0720-048X/© 2015 Published by Elsevier Ireland Ltd.
detection of nodules is often not possible until they reach a diameter of at least 8 mm. This limitation makes the detection of smaller nodules very difficult. That is why chest radiograph is not recommended as a screening tool for lung cancer, and indeed has a very low sensitivity for detecting solitary pulmonary nodules [3]. It is not very sensitive in detecting calcification within a pulmonary nodule either, demonstrating a sensitivity of 50% and a specificity of 87% [4]. The current standard for detection and characterization of pulmonary nodules is MDCT. However, MDCT cannot replace CXR for general thoracic imaging due to its high cost, workflow issues and radiation exposure. Thus, the radiographic technique that can improve the detection of pulmonary lesions without causing considerable increase in the radiation dose or cost could be very useful in day-to-day practice. In this context, the current study was
1384
S.G. Kumar et al. / European Journal of Radiology 84 (2015) 1383–1391
performed to compare the diagnostic performances of DT, DES-DR and DR in detecting pulmonary nodules with MDCT as reference standard.
2. Subjects and methods Institutional review board approval was obtained (Institute Ethics Committee, 8714/PG-2Trg/2011), and all patients gave written in-formed consent. Between January 2012 and June 2013, a total of 55 patients (30 patients with pulmonary nodules and 25 patients without pulmonary nodules) were prospectively and consecutively enrolled in the study by a radiologist not later involved in reading the radiographic studies. Patients who had undergone MDCT within the last 2 weeks and had one or more pulmonary nodules were included as study subjects. Patients who had undergone MDCT within the last 2 weeks and showed normal lung or focal lung pathology other than nodules (such as small patch of consolidation, mild and focal fibrosis or mild pleural effusion) were included as control subjects. Patients less than 18 years of age, pregnant women, patients not able to follow instructions and hold breath for 10 s, MDCT showing features of severe diffuse pulmonary disease (e.g., idiopathic pulmonary fibrosis, chronic pulmonary emphysema, diffuse consolidation, severe cardiac congestion, bronchiectasis etc.) or large pleural effusion masking more than 2/3 of either lung field, Patients who had undergone fine needle aspiration/biopsy or surgical intervention for the lung lesion were excluded. MDCT of chest was taken to be the gold standard in this study. MDCT was done on a 64-detector array CT system (Lightspeed VCT, GE Healthcare). The parameters were: detector collimation 64 × 0.625 mm; helical pitch 1.375; rotation time 0.6 s; tube voltage 120 kVp; AEC controlled tube current; 5 mm section thickness; and continuous reconstruction at 0.625 mm. The images were analysed in a work station (Adwantage Volumeshare 5/AW 4.6, GE Healthcare). The entire volumetric data were analysed in axial plane for nodule detection and then in coronal plane for correlation with DR/DT images. Size of the nodule was determined by the average of its length (longest nodule diameter) and width (widest diameter perpendicular to length) [5] in lung window settings (Level—600 Hounsfield unit (HU), Width 1600 HU). A nodule was considered calcified if calcification was visually detected in mediastinal window settings (Level + 40, Width 400) [4] and/or its attenuation value is greater than 200 HU [6]. DES-DR was performed on a flat panel detector system (Definium 8000, GE Healthcare). Erect postero-anterior chest radiographs were acquired with respiration suspended in deep inspiration using the standard dual shot method (120 kVp and 60 kVp) with automatic exposure control determined tube load. Routine DR images without subtraction and subtracted soft tissue and bone images were generated. DT involves acquisition of projection data over a limited range of X-ray tube movement. This allows reconstruction of data in multiple planes. It offers improved diagnostic performance over conventional radiography by decreasing the overlap of anatomical structures. Compared to CT, DT is cost effective and involved low radiation exposure. However, there is decreased depth resolution compared to CT. All DT examinations were performed immediately after DES-DR on the same system using the volumeRAD option. Erect posteroanterior scout view was taken with 120 kV peak tube potential and automatic exposure control determined tube load. Then this load used for the scout was multiplied by a user-adjustable dose ratio (10:1), and was equally distributed over all tomosynthesis projections (41/51/61 depending on patient thickness). Then for each projection the resulting tube load was rounded off to the
Table 1 study population characteristics. Characteristics
No. of patients
Patients with pulmonary nodules Control
30 25
Age (years) Range Mean
19–80 45.5
Sex Male Female
29 26
Nodule size (total no. of nodules = 110) ≤4 mm 5–6 mm 7–8 mm ≥9 mm
46 25 15 24
Calcification (total no. of nodules = 110) Present Absent
22 88
closest tube load setting possible, keeping a minimum tube load of 0.25 mAs per projection to maintain image quality. This process was done automatically by the VolumeRad software of the system. DT images were acquired with a tube sweep angle of approximately ±15◦ and a stationary detector. The breath hold time was 11 s. In all patients raw data were automatically processed to give a stack of 53 images with 5 mm interval. The DR and subtracted images of DES-DR and the images of DT were analysed independently by two radiologists (with 30 and 15 years’ experience, respectively, in thoracic radiology and 1 year experience in interpreting DT images) using the same commercial viewer (RadiAnt-64 bit—1.8.8). The manipulation of windowing, zoom/pan parameters was allowed. Number of pulmonary nodules in each patient with size and calcification status of each nodule was noted. Nodules were tagged with number and stored for final confirmation on CT. Confidence level from 1 to 5, according to fivepoint rating scale of ROC studies was assigned to detection of each nodule and to calcification in each nodule. DR, DES-DR and DT image interpretations in that order were separated by a time interval of one month and cases were presented for reading in different random order. MDCT images were interpreted for confirmation of nodules by both readers by consensus. 3. Statistical analysis Keeping MDCT as the reference standard, sensitivity, specificity and predictive values of DR, DES-DR and DT in detecting pulmonary nodules and calcification in pulmonary nodules were calculated separately for the two observers. ROC curves were generated using confidence level according to five-point rating scale of ROC studies and area under ROC curves (AUC) were calculated separately for the two observers. Kappa test of agreement was used to assess the inter-observer agreement for detection of nodules. All calculations were performed using SPSS® version 17 (Statistical Packages for the Social Sciences). AUC for different modalities were compared using z test. All statistical tests were two-sided and performed at a significance level of ˛ = 0.05. 4. Radiation dose The average dose area product (DAP) for DT was 0.18 Gy cm2 . Using a conversion factor [7] of 0.26 mSv Gy−1 cm−2 , the average effective dose was calculated to be 0.05 mSv. The DAP for DES-DR 0.07 Gy cm2 and the calculated average effective dose was 0.01 mSv. The average effective dose for MDCT was 6.9 mSv.
6. Discussion The detection of pulmonary nodules is hampered by summation, quantum mottle and anatomical noise on conventional chest radiography. It has been shown by Samei and colleagues [8] that
28.18% 80.49% 79.49% 29.46% 0.521 (0.420–0.622)
DES-DR DR
26.36% 70.73% 70.73% 26.36% 0.515 (0.410–0.619) 60% 84.78% 90.41% 46.99% 0.738 (0.658–0.818)
DT DES-DR
27.27% 84.78% 81.08% 32.77% 0.554 (0.457–0.650) 24.54% 65.22% 62.79% 26.55% 0.546 (0.449–0.644)
DR
Observer 1
Table 2 Comparison of performance in detection of pulmonary nodules.
A total of 55 patients were enrolled in the study. Thirty patients had one or more pulmonary nodules on MDCT, while 25 patients were controls. The mean age of patients was 45.56 years with age range of 19–80 years and standard deviation of 16.43 years. There was an equal sex distribution. Of the total 55 patients 29 were male and 26 were female. Among patients with nodules, 2 had pathologically proven bronchogenic carcinoma and 16 had pathologically proven malignant tumor elsewhere. Among patients without nodules 12 had a pathologically proven malignant tumor elsewhere. A total of 110 nodules were identified in 30 patients. All were solid nodules. Number of nodules in each patient ranged from 1 to 14. Patient characteristics are shown in Table 1. The nodules were classified according to their size into 4 groups (≤4 mm, 5–6 mm. 7–8 mm, ≥9 mm). Grouping into such categories (Graph 1) was followed in accordance with the recent Fleischner society guidelines for follow up of small pulmonary nodules detected on CT [5]. Distribution of nodules into size categories is shown in Fig. 1. Of the 110 nodules, 22 were calcified). There were 3 non-nodular calcifications (sclerotic focal lesions in ribs). Diagnosis of true nodules with confidence ≥3 was considered true positive. Diagnosis of nodule with confidence ≥3 which was not confirmed in MDCT was considered false positive. Such false positive nodules in DES-DR and DT and the number of truly nodule free cases were counted as true negatives from which sensitivity, specificity and predictive values were calculated. ROC curves were plotted separately for the two observers. Similarly detection of calcification with confidence ≥3 in a calcified nodule was considered true positive. Detection of calcification with confidence ≥3 in a non-calcified nodule was considered false positive. Non-calcified nodules, false positive nodule localisations in DES-DR and DT and the control group (patients without nodules) served as true negatives from which sensitivity, specificity and predictive values were calculated (Tables 2 and 3). ROC curves (Graphs 2 and 3) were plotted separately for the two observers (Graph 4). . Combined Sensitivities for detection of nodules for both observers according to size of nodules are given in Table 4. Diagnostic utility of DT compared to DES-DR and DR is shown in Figs. 1–3.
Observer 2
5. Results
Sensitivity Specificity Positive Predictive Value Negative Predictive Value ROC-AUC (95% CI in parenthesis)
Graph 1. Distribution of nodules into size categories.
61.82% 85.37% 90.89% 45.45% 0.736 (0.650–0.821)
1385
DT
S.G. Kumar et al. / European Journal of Radiology 84 (2015) 1383–1391
1386
S.G. Kumar et al. / European Journal of Radiology 84 (2015) 1383–1391
Fig. 1. A case of bronchogenic carcinoma in right lung (arrows in A–D). DT and the corresponding CT images (E, F) show multiple additional nodules (short arrows) in ipsilateral lung parenchyma. This information has a bearing on staging of bronchogenic carcinoma.
anatomical noise becomes the bigger limiting factor than random noise in the detection of small pulmonary nodules. However DESDR and DT can efficiently reduce or eliminate this noise caused by the ribs projecting over the lungs. Yet, the detection of pulmonary nodules by DES-DR has not shown to be uniformly better (as in the present study). This is partly due to the fact that as the lesion size becomes smaller, the advantage of anatomical noise reduction is negated. This is inferred from studies by Kashani et al. [9], and Ruhl et al. [10], who found no statistically significant difference in detection of pulmonary nodules (based on their size) by addition of DES-DR. Similarly, DT images are affected by the inherent tomographic noise that is related to the limited angle reconstruction involved in this technique. However, technological advancements
[11,12] have eliminated this effect on the image quality of DT. Thus, despite few limitations, both DES-DR and DT is a subject of considerable interest in detection of pulmonary nodules. DES-DR and DT can be performed in the same DR suite, provided it has the required modifications in the hardware set-up and additional image processing software. They impart radiation doses comparable to or slightly higher than DR [7–11]. There are many published studies on dual energy techniques [12–17] which have given discrepant and conflicting results regarding the role of these techniques in routine clinical practice. There are some published studies on tomosynthesis technique [18–20] and they have all shown that DT performs significantly better than DR in detecting pulmonary nodules. One phantom study
S.G. Kumar et al. / European Journal of Radiology 84 (2015) 1383–1391
1387
Graph 2. ROC curves for detection of nodules. (a, b, c—detection of nodules on DR, DES-SR, DT, respectively, by Observer 1. d, e, f—detection of nodules on DR, DES-SR, DT, respectively, by Observer 2).
1388
S.G. Kumar et al. / European Journal of Radiology 84 (2015) 1383–1391
Graph 3. ROC curves for detection of calcification. (a, b, c—detection of calcification on DR, DES-SR, DT, respectively, by Observer 1. d, e, f—detection of calcification on DR, DES-SR, DT, respectively, by Observer 2).
S.G. Kumar et al. / European Journal of Radiology 84 (2015) 1383–1391
1389
Fig. 2. DT (A) and CT (B) showed a large nodule (arrow) in right lung base which was missed on both DR (C) and DES-DR (D) prospectively. Also note the better depiction of ribs (thin arrows) and the vessels by DT (arrowhead).
[21] has been performed to compare DES-DR with DT, in which DT performed significantly better. In our study the overall cumulative sensitivities of DR, DES-DR and DT in detection of pulmonary nodules were 25.45%, 27.75% and 60.9%, respectively, with specificities of 68.9%, 82.64% and 85.07%,
respectively. The low overall sensitivity can be explained by the overwhelming number of small pulmonary nodules (≤8 mm—86 out of a total of 110). For nodules ≥ 9 mm which, according to the recent guidelines for management of small pulmonary nodules detected on MDCT scans from Fleischner society, require
Fig. 3. DT (A) detected tiny nodule (arrow) in right lung that was missed by DES-DR (B).
DT
52% 99.21% 92.86% 91.24% 0.651 (0.502–0.800) 32% 99.21% 88.89% 88.03% 0.613 (0.474–0.751)
DR
48% 98.47% 85.71% 90.84% 0.648 (0.498–0.798)
DT DES-DR
36% 99.24% 90% 89.04% 0.696 (0.563–0.829) 25% 97.7% 72.72% 88.28% 0.590 (0.457–0.724)
DR
Table 3 Comparison of performance in detection of calcification in pulmonary nodules.
Sensitivity Specificity Positive Predictive Value Negative Predictive Value ROC-AUC (95% CI in parenthesis)
Observer 2 Observer 1
36% 99.24% 90% 88.65% 0.696 (0.562–0.829)
S.G. Kumar et al. / European Journal of Radiology 84 (2015) 1383–1391
DES-DR
1390
Graph 4. Combined Sensitivities for detection of nodules for both observers according to size of nodules.
further investigations (PET/biopsy) to rule out malignancy, the cumulative sensitivities were 54%, 62.5% and 95.8%, respectively. However, none of the three modalities tested here can be used in par with MDCT (gold standard) in for screening or confirmation of pulmonary nodules in view of overall low sensitivity and only moderate specificity. Though DES-DR showed a slight improvement over DR in terms of sensitivity (particularly for nodules larger than 8 mm), there was no statistically significant improvement in overall diagnostic accuracy (p1 = 0.91, p2 = 0.93). This result is in agreement with the previously published studies by Kashani et al. [9] and Ruhl et al. [10], both of which had used dual-shot DES-DR technique as in this study. The former was a single centre study with 129 patients with 158 nodules and there was no significant difference in ROC performance between DE and DR (AUC DES-DR = 0.795 AUC DR = 0.789; p = 0.696). The latter was an international multicentric trial with 149 nodules and there was no significant difference in ROC performance between DE and DR (AUC DES-DR = 0.602 & AUC DR = 0.631; p = 0.4). On the contrary, Szucs et al. [16], have reported a statistically significant improvement with single shot DES-DR technique. However, they had noted that detection of nodules
