Eur Radiol (2008) 18: 1831–1839 DOI 10.1007/s00330-008-0945-6
CHEST
Ricarda Rühl Magdalena M. Wozniak Michael Werk François Laurent Georg Mager Michel Montaudon Andreas Pattermann Antoine Scherrer Jean-Pierre Tasu Maciej Pech Jens Ricke
CsI-detector-based dual-exposure dual energy in chest radiography for lung nodule detection: results of an international multicenter trial
Received: 30 July 2007 Revised: 23 January 2008 Accepted: 25 February 2008 Published online: 19 April 2008 # European Society of Radiology 2008
A. Scherrer Hôpital Foch, Foch, France
R. Rühl (*) Universitätsklinikum Magdeburg, Klinik für Radiologie und Nuklearmedizin, Leipzigerstr. 44, 39120 Magdeburg, Germany e-mail:
[email protected] Tel.: +49-391-6713030 Fax: +49-391-6713029 M. M. Wozniak Dziecięcy Szpital Kliniczny Akademii Medycznej w Lublinie, Lublin, Poland M. Werk Universitätsmedizin Berlin, Charité CVK, Berlin, Germany F. Laurent Hôpital Haut-Lévèque, Le Centre Hospitalier Universitaire de Bordeaux, Hôpitaux de Bordeaux, Pessac, France G. Mager Bundeswehrkrankenhaus, Berlin Berlin, Germany Berlin, Germany M. Montaudon Hôpital Haut-Lévèque, Le Centre Hospitalier Universitaire de Bordeaux, Hôpitaux de Bordeaux, Pessac, France A. Pattermann Wilhelminenspital, Vienna, Austria
J.-P. Tasu Le Centre Hospitalier Universitaire de Poitiers, Poitiers, France M. Pech . J. Ricke Universitätsklinik Magdeburg, Klinik für Radiologie und Nuklearmedizin, Magdeburg, Germany
Abstract To assess both sensitivity and specificity of digital chest radiography alone and in conjunction with dual-exposure dual-energy chest radiography for the detection and classification of pulmonary nodules. One hundred patients with a total of 149 lung nodules (3–45 mm; median, 11 mm) confirmed by CT were included in this study. Dual-exposure dual-energy chest radiographies of each patient were obtained using a CsI detector system. Experienced boardcertified chest radiologists from four different medical centers in Europe reviewed standard chest radiographs alone and in conjunction with dualenergy images blinded and in random order. The reviewers rated the probability of presence, calcification and malignancy of all lung nodules on a five-point rating scale. Lesions detected were identified by applying a specific coordinate system to enable precise verification by the study leader. A receiver-operating characteristic (ROC) analysis was performed. In
addition to the 149 true-positive CT proven lesions, 236 false-positive lung nodules were described in digital chest radiographies in conjunction with dualenergy chest radiographies. The cumulative sensitivity of chest radiography in conjunction with dual energy was 43%, specificity was 55%. For digital radiography alone, sensitivity was 35% and specifity was 83%. For the dual energy system, positive predictive value was 58%, and negative predictive value was 66% compared to the digital radiography with a positive predictive value of 59% and a negative predictive value of 65%. Areas under the curve in a ROC analysis resulted in 0.631 (95% confidence interval=0.61 to 0.65) for radiography with dual energy and 0.602 (95% confidence interval=0.58 to 0.63) for digital radiography alone. This difference was not statistically significant. For the detection of lesion calcification or the determination of malignancy, ROC analysis also failed to show significant differences. CsI-based flat-panel dualexposure dual-energy imaging added to standard chest radiography did not show statistically significant improvement for the detection of pulmonary nodules, nor the identification of calcifications, nor the determination of malignancy. Keywords Thoracic imaging . Dual energy . Pulmonary nodules . Lung . Digital radiography
1832
Introduction Chest radiography has defended its position in the diagnostic workflow despite dramatic advances in the field of computed tomography (CT). The main arguments in favor of chest radiography are broad availability, cost effectiveness as well as the relatively low-dose exposure. However, compared to CT imaging, sensitivity and specificity for the detection of lung nodules are low. In addition, small lung nodules are a frequent finding specifically in a risk population for bronchial carcinoma, such as heavy smokers [1, 2]. Hence, an improved accuracy to determine a lung nodule’s likelihood for malignancy would be desirable, i.e., by a reliable detection of calcifications. Previous studies have suggested potential improvements for the detection of non-calicified as well as calcified lung nodules by cesium-iodide (CsI) detector-based dualexposure dual-energy chest radiography compared to standard chest radiography alone [3]. However, limitations of these studies result from their monocentric study design and the small number of patients and lung nodules involved. The intent of our study was to determine the value of dual-energy images in combination with conventional chest radiography in the detection of lung nodules and their classification as malignant or benign. To enhance the value of the study’s outcome we chose a prospective, multicenter study design including participants from internationally renowned centers.
Materials and methods
patients remained unconfirmed. Among patients with neoplasms the majority had lung cancer (26 patients, 36.6%) and colorectal carcinoma (18 patients, 25.4%). Other malignancies included: breast carcinoma, renal carcinoma, lymphoma, cervix carcinoma, endometrial carcinoma, stomach carcinoma, esophagus carcinoma, pancreas carcinoma, hepatocellular carcinoma, cholangiocellular carcinoma and others. In the study group four patients presented with two different primary malignancies and two other patients with three different primary tumors. Of 100 patients included, 74 displayed pulmonary nodules. Thirty-seven patients displayed 1 nodule, 19 patients displayed 2 nodules, 7 patients 3 nodules, 4 patients 4 nodules, 5 patients 5 nodules and 2 patients 6 nodules. Twenty-six patients did not have any pulmonary nodule. The median size of all pulmonary nodules was 11 mm (range 3–45 mm). The median size of 121 non-calcified nodules was 12 mm (range 3–45 mm), and the median size of 28 calcified nodules was 10 mm (range 4–23 mm) with a median attenuation of 239 HU (range 177–456 HU). The exact distribution of nodule sizes is shown in Fig. 1. Ninety-two nodules were located within the right lung, and 57 within the left lung. The majority of nodules was found within the segments 6 (n=31), 3 (n=29) and 2 (n=26). Other nodules were diagnosed within all the remaining segments with an exception of segment 7, where none of the nodules was localized. The status of all lung nodules was assessed either histologically or by follow-up CT. With respect to CT follow-up, we proposed malignancy if a diameter growth of a nodule of more than 25% was confirmed within 6 months follow-up.
We included 100 patients from four European medical centers with a total of 149 lung nodules confirmed by CT between April 2003 and April 2005 consecutively. The study group comprised 47 male and 53 female patients with a median age of 64 years (age range 22–92 years). Written informed consent was obtained from all patients. The study was approved by the institutional ethics committees.
14 12
Eligibility criteria Inclusion was limited to patients with no more than six lung nodules with a size of 3 to 45 mm confirmed by CT. The chest radiographs and thoracic CT were performed within a maximum interval of 4 weeks. Exclusion criteria were any severe diffuse pulmonary diseases (e.g., idiopathic pulmonary fibrosis, chronic pulmonary emphysema, pneumonia, cardiac congestion or diffuse bronchiectasis). Calcification of nodules was defined by a density of those nodules above 175 HU (Hounsfield units) in CT [4]. Out of all 100 patients included in this study, 71 patients had known malignant disease, 18 patients had no evidence of malignancy. The origin of pulmonary nodules in 11
Frequency (n)
10 8 6 4 2 0 3
5 7
9 11 13 15 17 20 22 24 26 33 35 41 45
Diameter (mm) Fig. 1 Distribution of nodule sizes for all types of nodules
1833
Out of these 47, 43 nodules were confirmed to be malignant and 4 nodules were benign. Image acquisition and verification of nodule location Chest radiographs of all patients were obtained using a flatpanel digital chest system (Revolution XQ/I, General Electric Healthcare, UK), based on a CsI scintillator and an amorphous silicon photodiode-transistor array.The detector has an image size of 41×41 cm and a pixel dimension of 0.2×0.2 mm. The dual-energy examination consisted of standard posteroanterior (PA) radiographs as well as subtracted soft tissue and bone images (Fig. 2). Dualenergy images were acquired applying a dual-exposure technique within 200 ms. The imaging parameters included a 120 kV and 2 mAs image at a speed equivalent to approximately 400 and a 60 kVand 8 mAs image at a speed equivalent to approximately 1,000. In addition to subtraction techniques generating the isolated soft tissue as well as the isolated bone image, post-processing algorithms with pixel shifting and noise reduction were applied [5]. In a trial performed at our institution before, Fischbach et al. [6] found a slight increase in radiation dose by applying dual-energy images. In a phantom, the entrance dose for standard chest radiographs was 110 μGy for the posteroanterior shot and 680 μGy for the lateral shot using an automatic exposure control at a speed equivalent of 400. Applying dual energy, the entrance dose for the highkilovolt image at a speed equivalent to 400 was 110 μGy as well as for the low-kilovoltage shot at a speed equivalent of 1,000. The high-kilo voltage image was used as the standard chest image for the dual-energy examination so that the only additional image was the low-kilovoltage image. Hence, the total dose for the chest examination
including posteroanterior and lateral shots increased 14% by applying the dual-energy mode versus the standard chest radiography examination. Chest CT examinations were performed using multislice CT modalities of various vendors. Parameter settings applied at all centers were done with an increment of 5 mm and window/center of 1,600/-600. The entire lung was scanned continuously in single breath-hold technique. Chest CT images served as the gold standard to determine the exact sizes, locations and calcification of the lesions. The CT images obtained from the 100 patients were reviewed by two experienced radiologists (R.R. and M.W.; no participation in the observer performance study). The decision on location, size and presence of calcification was made in consensus fashion. Image review Image review was performed by five radiologists from four different large academic medical centers independently. All readers were general radiologists experienced in chest radiography (reader 1: 19 years of experience, reader 2: 12 years of experience, reader 3: 15 years of experience, reader 4: 27 years of experience and reader 5: 15 years of experience). Image review was done on PA and lateral view images on soft copies on flat-panel screens (SIEMENS SMD 21500-D and DOME Mds/PCX and BARCO 3MP2FH), in the same light conditions for each particular reader. Two groups of data sets were established, the first set containing only the chest radiographs of one half of the patients and the full set including the dual-energy images (soft tissue and bone images) for the other half of the patients. The second set contained the complementary images (full set
Fig. 2 a–c Chest radiograph and dual energy soft tissue and bone images of a 61-year-old woman with lung metastases from breast cancer. Images show multiple nodules on both lungs
1834
Table 2 AUC (Az) values for the confidence of nodule existence, lesion calcification and malignancy per reviewer Reader Reader Reader Reader Reader 1 2 3 4 5
Fig. 3 Identification of a pulmonary nodule. First, the reviewer indicates the lung compromised. Second, two accessory lines are drawn: line A, horizontal–linking both apecies of the lungs; line B– perpendicular to line A and centered at the trachea. Line A-Point D serves as the starting point for line C, which defines the distance and the angle to the center of each nodule
for first half of patients and chest radiographs only for second half). In order to avoid a learning effect, the reviews of the second set of cases were performed after a minimum of 4 weeks after the first set. Chest images were blinded and viewed in random order. Readers were allowed to change windowing and to use magnification when reading the images. All readers had the same training in reading dual-energy chest radiographs before starting the trial by reading the same training CD with test pattern while knowing the findings. Furthermore, this training CD was read every time before starting a new reading session. To correlate a nodule location from chest radiography to the according CT, an identification system was established by the study coordinator. The reviewers indicated each nodule by size and location. With respect to location, they determined the chest side as well as specific coordinates.
Table 1 True-positive and false-positive findings as indicated by each of the reviewers by both techniques Reader 1 True-positive findings Dual energy 71 Digital radiography 47 False-positive findings Dual energy 38 Digital 31 radiography
Reader 2
Reader 3
Reader 4
Reader 5
82 66
60 58
65 62
42 28
99 82
35 30
26 19
38 20
Area under the ROC curve values for the confidence of nodule existence 0.61 0.63 0.68 0.58 Dual-energy AUC 0.66 (Az) Confidence interval 0.60– 0.55– 0.57– 0.62– 0.52– (95%) 0.72 0.67 0.69 0.74 0.64 0.57 0.63 0.67 0.55 Digital radiography 0.60 AUC (Az) Confidence interval 0.54– 0.51– 0.58– 0.61– 0.49– (95%) 0.66 0.64 0.70 0.73 0.61 Significance level P P=0.06 P=0.23 P=0.96 p=0.72 p=0.34 Area under the ROC curve values for the confidence of lesion calcification 0.63 0.60 0.54 0.54 Dual-energy AUC 0.61 (Az) Confidence Interval 0.46– 0.45– 0.49– 0.44– 0.46– (95%) 0.58 0.57 0.61 0.57 0.58 0.54 0.52 0.57 0.54 Digital radiography 0.59 AUC (Az) Confidence interval 0.48– 0.43– 0.49– 0.46– 0.46– (95%) 0.60 0.55 0.61 0.59 0.59 Significance level P p=0.85 p=0.16 p=0.25 p=0.67 p=0.99 Area under the ROC curve values for the confidence of malignancy 0.64 0.62 0.68 0.61 Dual-energy AUC 0.68 (Az) Confidence interval 0.59– 0.55– 0.56– 0.61– 0.52– (95%) 0.71 0.67 0.68 0.73 0.64 Digital radiography 0.63 0.62 0.62 0.68 0.57 AUC (Az) Confidence interval 0.54– 0.51– 0.57– 0.60– 0.50– (95%) 0.66 0.64 0.69 0.72 0.62 Significance level P P=0.293 P=0.656 P=0.940 P=0.960 P=0.446
The side, distance and angle were determined on PA conventional digital radiographs and on soft tissue PA images for each lesion (Fig. 3). The right- or left-side had to be selected as the first step to locate the nodule. Two accessory lines were drawn to define the distance: line A linking apexes of both lungs and line B perpendicular to line A and parallel to the trachea. The point where lines A and B met was the starting point to determine the distance. The distance was measured in millimeters to the middle of each nodule. Subsequently the angle was defined as the angle between line B and the distance line (Fig. 3). The nodule was recognized as such, if all of the values described and additionally the size of the nodule did not
1835
ROC Curve - Reader 3 1,0
0,8
0,8
Sensitivity
Sensitivity
ROC Curve - Reader 1 1,0
0,6
0,4
0,2
0,6
0,4
0,2
0,0
0,0 0,0
0,2
0,4
0,6
0,8
1,0
0,0
0,2
0,4
1 - Specificity Digital radiography
Dual-energy
0,6
0,8
1,0
1 - Specificity Digital radiography
Dual-energy
Fig. 4 ROC curve reader 1 for lesion detection
Fig. 6 ROC curve reader 3 for lesion detection
differ more than 10% to values given by the study leader. Otherwise, the nodule was noted as a different lesion. In case any of the reviewers reported nodules that were not confirmed with CT, such lesions were categorized as false positive and still recorded.
ROC analysis utilized a five-point scale to estimate the existence of a lesion, the presence of calcification or malignancy (1=no evidence, 5=strongest evidence). In radiology a five-point rating scale has proved its worth to categorize radiographs [7]. ROC Curve - Reader 4 1,0
0,8
0,8
Sensitivity
Sensitivity
ROC Curve - Reader 2 1,0
0,6
0,4
0,2
0,0 0,0
0,6
0,4
0,2
0,0
0,2
0,4
0,6
0,8
1 - Specificity Dual-energy
Fig. 5 ROC curve reader 2 for lesion detection
1,0
0,0
0,2
0,4
0,6
0,8
1 - Specificity Digital radiography
Dual-energy
Fig. 7 ROC curve reader 4 for lesion detection
Digital radiography
1,0
1836
ROC Curve - Reader 5
radiologist, the area under the ROC curve as well as 95% confidence limits was calculated for digital chest radiography alone or in conjunction with dual energy. We applied Student’s t-test to prove significance with respect to the results for the probability of nodule existence (detection), of nodule calcification and of nodule malignancy, respectively. Furthermore, we introduced five subgroups comprising lung nodules: smaller than 1 cm, 1 to 1.9 cm, 2 to 2.9 cm, 3 to 3.9 cm and greater than 4 cm.
1,0
Sensitivity
0,8
0,6
0,4
Results Lesion detection
0,2
0,0 0,0
0,2
0,4
0,6
0,8
1,0
1 - Specificity Dual-energy
Digital radiography
Fig. 8 ROC curve reader 5 for lesion detection
Statistical analysis All analyses were performed using a software program (SPSS for Windows, version 12.0.0, Chicago, IL, and MedCalc for Windows, version 9.2.0.0 (MedCalc Software, Mariakerke, Belgium). Sensitivity, specificity, positive and negative predictive values as well as p-values were calculated for each individual reviewer. Locations of all false-positive findings were determined to serve as the negative control group. These 213 locations consisted of all positions that had been identified mistakenly as false positive by at least one of the observers during the review of chest radiographs. Although more false-positive findings were indicated in standard radiography combined with dual energy radiographs than in standard radiography alone, the difference was not significant (p=0.4). In addition, we performed a ROC analysis. For each
Table 1 shows the number of true-positive and falsepositive nodules as indicated by each of the reviewers by both techniques. The difference between chest radiography alone or in conjunction with dual energy images was not significant. In addition to the 149 true-positive findings on CT scans, 213 false-positive lung nodules were described. Locations of all false-positive findings were determined to serve as the negative control group. These 213 locations consisted of all positions that had been identified mistakenly as false positive by at least one of the observers during the review of chest radiographs. Although more false-positive findings were indicated in standard radiography combined with dual energy radiographs than in standard radiography alone, the difference was not significant (p=0.4). The cumulative area under the curve (Az) for all radiologists (all individual inputs counted as one) for the probability of nodule existence was 0.602 for digital radiography and 0.631 for digital radiography in conjunction with dual-energy images (Table 2; Figs. 4, 5, 6, 7 and 8). The cumulative sensitivity of digital radiography review alone was 35%, specificity 83%, positive predictive value 59% and negative predictive value 65%. These values for chest radiography review in conjunction with dual energy were 43%, 55%, 58% and 66%, respectively (Table 3). Results of a subgroup analysis for different lesion sizes revealed significance for nodules of 20 to 29 mm in size only (Table 5).
Table 3 Results for nodule detection for all readings (5 readers, 100 studies) True positive (mean)
False positive False negative True negative Sensitivity Specificity (mean) (mean) (mean) (%) (%)
PPV (%)
NPV (%)
320/500 (64)
229/500 (46)
421/500 (84)
Digital radiography 261/500 (48)
182/500 (36)
478/500 (96)
320/549 (58.1%) 261/443 (58.9%)
839/1,260 (66.5%) 882/1,360 (64.8%)
Dual energy
839/500 (168) 320/741 (43.1%) 882/500 (176) 261/739 (35.3%)
839/1,518 (55.2%) 882/1,064 (82.8%)
1837
Table 4 Results for nodule calcification for all readings (5 readers, 100 studies)
Dual energy
True positive (mean)
False positive (mean)
False negative (mean)
True negative (mean)
Sensitivity Specificity (%) (%)
PPV (%)
NPV (%)
34/500 (7)
54/500 (5)
106/500 (21)
547/500 (109)
67/500 (13)
112/500 (22)
532/500 (106)
34/140 (24.2%) 27/139 (19.4%)
34/98 (34.6%) 27/94 (28.7%)
547/653 (83.7%) 532/644 (82.6%)
Digital radiography 27/500 (5)
547/601 (91.0%) 532/599 (88.8%)
Determination of lesion calcification
Discussion
Of 149 pulmonary nodules determined in chest CT in 74 patients, 28 showed calcifications. The median size of these 28 calcified nodules was 9.5 mm (range 4–23 mm), median attenuation was 239 HU (range 177–456 HU). The area under the curve (Az) for all radiologists did not show significant differences between standard digital radiography alone or in conjunction with dual-energy radiography (Table 2). The cumulative sensitivity of lesion calcification for digital radiography review alone was 19%, specificity 89%, positive predictive value 29% and negative predictive value 83%. These values for chest radiography review in conjunction with dual energy were 24%, 91%, 34% and 84%, respectively (Table 4).
Although lung cancer is not the most common type of cancer, it is the leading cause of cancer death in the western hemisphere [8]. The incidence of a solitary pulmonary nodule on plain chest radiographs ranges from 0.09% to 0.2% in healthy subjects [9]. Conventional PA and lateral chest radiographs continue to be the most practical means for the initial detection and evaluation of cancer in the chest. However, one of the main shortcomings of chest radiography is poor sensitivity for the detection of solitary pulmonary nodules that are smaller than 2 cm in diameter [10]. Patient factors associated with an increased likelihood of malignancy include advanced age, a history of smoking and a history of prior malignancy [11]. Radiographic characteristics that have been proposed as useful to analyze pulmonary nodules include [1–4, 12]:
Determination of lesion malignancy
–
Out of all nodules (n=149) in all patients (n=73), 43 lesions were confirmed to be malignant (true positive) either by histological confirmation or by more than 25% increase by unidimensional measurement in follow-up CT. The remaining 106 pulmonary nodules could not be further differentiated; therefore, true-negative nodules could not be classified. Only those lesions that had been identified incorrectly were labeled false positive. ROC curves obtained from all radiologists did not show appreciable differences. Also the area under the curve (Az) for all radiologists did not show any statistically significant differences between standard digital radiography and dual energy radiography in any of the cases for determination of probability of malignancy. Table 2 shows the areas under the ROC curves for each reviewer for the determination of malignant lesions in both techniques.
–
Border characteristics: spiculated lesions are more like to be malignant, smooth border lesions are more likely to be benign Furuya et al. [12] analyzed the margin characteristics of 193 pulmonary nodules: 82% of the lobulated, 97% of the densely spiculated, 93% of the ragged and 100% of the halo nodules were malignant. Eighty percent of the tentacle or polygonal nodules were inflammatory, and 66% of the round ones were benign. The six types differed statistically as to the nature of the benignity/malignancy (p