| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Thoracic Imaging |
1 From the Department of Diagnostic Radiology, Osaka Medical Center for Cancer and Cardiovascular Diseases, Japan (M.T., K.K., J.A., C.K.); Department of Computer Science and Systems Engineering, Yamaguchi University, Japan (S.K.); Department of Radiology, Osaka Chuo Hospital, Japan (N.K.); and Department of Radiology, Osaka University Graduate School of Medicine, 2-2 Yamadaoka, Suita, Osaka 565-0871, Japan (M.T., T.J., N.T., O.H.). Received August 25, 2000; revision requested October 14; final revision received September 20, 2001; accepted December 11. Address correspondence to M.T. (e-mail: tubamoto@radiol.med.osaka-u.ac.jp).
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMATERIALS AND METHODS RESULTS DISCUSSIONREFERENCES |
|---|
MATERIALS AND METHODS: Chest radiographs of 75 patients with peripheral carcinomas 20 mm in diameter or smaller (26 localized bronchioloalveolar carcinomas [BACs], 49 other carcinomas) and 60 normal chest radiographs were retrospectively reviewed individually by 10 radiologists. The extent of GGO within the lesions at thin-section CT was reviewed retrospectively. The detection rates for localized BAC and other carcinomas on chest radiographs were calculated and were correlated with tumor size and extent of GGO.
RESULTS: The mean sensitivity for detection of small peripheral carcinomas was 58.5% ± 8.8 (standard error) for localized BAC and was 78.6% ± 5.1 for other carcinomas (P = .024). Lesions that were smaller than 15 mm in diameter and had an extent of GGO of 70% or greater at thin-section CT were more difficult to detect than tumors that had larger diameters or less extensive GGO (
2 = 8.13, df = 1, P = .004).
CONCLUSION: The detection of small peripheral carcinomas on chest radiographs is influenced by tumor size and extent of GGO as seen at thin-section CT.
© RSNA, 2002
Index terms: Lung, ground-glass opacification, 60.32 • Lung neoplasms, CT, 60.12118 • Lung neoplasms, diagnosis, 60.32
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODS RESULTS DISCUSSIONREFERENCES |
|---|
Several investigators have reported on thin-section computed tomographic (CT) imaging of small peripheral lung cancers, including localized bronchioloalveolar carcinoma (BAC) (13–18). According to these studies, localized BAC, which has a considerably good prognosis after treatment (19–21), demonstrates a greater extent of ground-glass opacity (GGO) on thin-section CT scans than do other adenocarcinomas (13).
We surmised that, due to the extent of GGO on thin-section CT scans of BAC tumors, these tumors would be more difficult to detect than other carcinomas on chest radiographs. We postulated that detection would be influenced not only by tumor size but also by the extent of GGO.
The purpose of our study was to evaluate the detection of small peripheral lung cancers on chest radiographs on the basis of tumor size and extent of GGO at thin-section CT.
|
MATERIALS AND METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODS RESULTS DISCUSSIONREFERENCES |
|---|
The chest radiographs of 60 subjects (36 men, 24 women; age range, 23–79 years; mean age, 55 years) with normal CT scans of the chest were selected so that the age and sex of the healthy subjects matched as closely as possible the age and sex of the lung cancer patients. Thus, the study included 135 chest radiographs: 60 normal radiographs and 75 radiographs of peripheral lung cancers 20 mm in diameter or smaller. Of the 75 lung cancers, 26 were localized BAC (compatible with type A, B of the Noguchi classification) (19), 32 were papillary adenocarcinoma (compatible with type C of the Noguchi classification), five were poorly differentiated adenocarcinoma (compatible with type D of the Noguchi classification), one was acinar adenocarcinoma (compatible with type E of the Noguchi classification), seven were papillary adenocarcinoma with a compressive and destructive growth pattern (compatible with type F of the Noguchi classification), two were squamous cell carcinoma, one was adenosquamous carcinoma, and one was large cell carcinoma. These 135 cases were selected for this study, and CT scans of the control group were reviewed and regarded as normal by one radiologist (M.T.) who was not included in either the interpretation of the chest radiographs or the estimation of the extent of GGO on the thin-section CT scans.
The imaging parameters and the screen-film systems used in this study for chest radiography were as follows: 120 kVp and MGC film (Konica, Tokyo, Japan) with an SRO-180 screen (Konica) before March 29, 1994 (used to obtain 32 chest radiographs—10 in healthy subjects and 22 in cancer patients); 130 kVp and UR-1 film (Fuji, Tokyo, Japan) with an HG-M screen (Fuji) between March 29, 1994, and February 22, 1997 (used to obtain 53 chest radiographs—20 in healthy subjects and 33 in cancer patients); and 140 kVp and UR-1 film (Fuji) with an HG-M2 screen (Fuji) after February 23, 1997 (used to obtain 50 chest radiographs—30 in healthy subjects and 20 in cancer patients). All images were obtained at a focus-to-film distance of 200 cm. The interval between chest radiography and the CT examination varied (range, 0–222 days; median, 30 days).
Lung Specimens
The lung containing the tumor was excised from each of the 75 patients, and inflated lung specimens were routinely prepared by injecting a 1% agar solution through the leading bronchi. Specimens were sliced at their maximum diameter. The tumor diameters in the pathologic specimens were measured by one of several surgeons. The lesions included in this study ranged in size from 5 to 20 mm (mean, 15.0 mm); specifically, the size of the localized BAC tumors ranged from 5 to 20 mm (mean, 14.9 mm), and the size of the non-BAC nodules ranged from 6 to 20 mm (mean, 15.1 mm) (P = .9).
Thin-Section CT
Film hard-copy thin-section CT images, with all references to patients’ names and hospital record numbers masked during interpretation, were available for all lesions included in this study. CT scanning was performed without the administration of intravenous contrast material. Scanning through the lesions was performed with a CT/T 9800 scanner (General Electric Yokogawa Medical Systems, Osaka, Japan) in 1.5-mm-thick sections, a TCT 900S scanner (Toshiba Medical Systems, Tokyo, Japan) in sequential 2-mm-thick sections at one breath hold, or a Somatom Plus scanner (Siemens Medical Systems, Erlangen, Germany) as helical scanning with 2-mm-thick sections at one breath hold. Five sections with 3-mm intersection spacing were obtained through the nodule with the GE 9800 scanner, five sections with 2-mm spacing were obtained with the Toshiba scanner, and 24–32 sections with 1-mm spacing were obtained with the Somatom Plus scanner. Images were reconstructed with an edge-enhancing algorithm. The images were photographed with a window level of -600 HU and a window width of 1,500 HU (ie, lung windows) and a window level of 0 HU and a window width of 300 HU (ie, mediastinal windows). Thin-section CT images in 70 of the 75 patients with cancer included in this study were also described in a previous study (13–15).
Interpretation of Chest Radiographs
All references to patients’ names and hospital record numbers were masked during interpretation. Ten radiologists—five senior board-certified chest radiologists (J.A., N.K., T.J., N.T., O.H.), and five residents (three 4th-year residents and two 2nd-year residents)—evaluated each of the 135 chest radiographs individually. Before the interpretation of the images, written instructional materials were distributed to the observers, and a personal teaching session was held with several cases that were not included in the actual study. The scoring method was also explained. Each observer recorded, on a prescribed worksheet, the presence of a focal abnormal area of opacity on a radiograph and rated the lesion, on the basis of their confidence in the presence or absence of a focal abnormal area of opacity, according to one of the following four levels: 1, definitely normal; 2, possibly normal but an abnormal finding is suspected; 3, probably abnormal; and 4, definitely abnormal. It was emphasized that when observers regarded a chest radiograph as normal, they should rate the lesion as 1 or 2, and that when they regarded a radiograph as abnormal, they should rate the lesion as 3 or 4. Observers were informed that each radiograph might contain no or only one area of abnormal opacity and that all lesions were proved to be small peripheral lung cancers. No mention was made of the frequency of nodules or normal studies.
Estimation of Extent of GGO at Thin-Section CT
The analysis of film hard-copy images was performed by three independent observers (including K.K.) who were board-certified chest radiologists and who were not included in the interpretation of the chest radiographs. Before the analysis, a teaching session was conducted with several cases not included in the analysis. The observers quantitated the percentage of the lesion occupied by areas of GGO as distinct from the percentage of soft-tissue attenuation of the lesion. This quantitation was subjective and was based on visual estimation; observers estimated the extent of GGO from 1% to 100%. GGO was defined as a phenomenon in which there was hazy increased attenuation of the lung tissue, but the visibility of the underlying vascular markings was preserved (22). To minimize the effect of volume averaging, the quantitation of the extent of GGO was made only at the level of the widest diameter of the lesion. The average of the values determined by the three observers was used as the value of the extent of GGO within a lesion.
Estimation of Calcification at Thin-Section CT
On the basis of film hard-copy images from thin-section CT photographed with mediastinal windows, the presence or absence of calcification in each tumor was also evaluated by a radiologist (M.T.) who was not included either in the interpretation of chest radiographs or in the estimation of the extent of GGO.
Analysis of Detection on the Basis of Tumor Size and Extent of GGO Estimated at Thin-Section CT
All tumors were classified into four groups on the basis of their size and extent of GGO. The study criteria—the tumor size cutoff point was 15 mm, and the cutoff point for the extent of GGO was 70%—were chosen because previous investigators have noted the difficulty in detecting tumors 10–15 mm in diameter or smaller on chest radiographs (7,9) and because the extent of GGO in most BAC tumors is 76% or greater (13). Lesions that five or more observers, out of five experienced observers and five residents, could detect and rated as 3 or 4 were regarded as conspicuous. Lesions that four or fewer observers could detect were regarded as subtle.
We also analyzed the distribution of the four levels of confidence across false-negative, true-negative, false-positive, and true-positive readings.
Statistical Analysis
Sensitivity and specificity values, standard errors, and intraclass correlation coefficients (ICCs) were calculated with the random-effects model (our study included two random effects: observers and chest radiographs). Comparisons of detection rates for the observers and the tumor types were performed with the Wilcoxon rank sum test.
The detection of small peripheral lung cancers on chest radiographs was analyzed on the basis of the tumor size and the extent of GGO at thin-section CT by means of logistic regression. Four groups were then defined by dichotomous strata of the two cutoff points—tumor size of 15 mm and extent of GGO of 70%—and the differences among the groups were compared with contrasts. Interobserver variability in grading the extent of GGO was estimated with an ICC.
|
RESULTS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODS RESULTS DISCUSSIONREFERENCES |
|---|
There was no statistically significant difference in detection rate between the experienced and the less-experienced observers. For five experienced observers, the mean sensitivity for detection of peripheral lung cancer was 62.3% ± 9.2 (ICC = 0.779) for localized BAC and 83.7% ± 4.4 (ICC = 0.617) for other cancer types; the mean specificity was 67.3% ± 5.5 (ICC = 0.357). For five less-experienced observers, the mean sensitivity was 54.6% ± 9.5 (ICC = 0.626) for localized BAC and 73.5% ± 6.2 (ICC = 0.579) for other cancer types; the mean specificity was 67.0% ± 10.6 (ICC = 0.096).
The observers used a four-level measure to classify the focal abnormal areas of opacity, and we evaluated the distribution of the four levels of confidence across true- or false-positive readings and true- or false-negative readings. There was a tendency that false-positive readings were more likely to have been given a level 3 classification (89.1%, 220 of 247 observations) than a level 4 classification (10.9%, 27 of 247 observations), and true-positive readings were more likely to have been given a level 4 classification (68.0%, 368 of 541 observations) than a level 3 classification (32.0%, 173 of 541 observations). There was also a tendency that true-negative readings were more likely to have been given a level 1 classification (61.4%, 245 of 399 observations) than a level 2 classification (38.6%, 154 of 399 observations); however, no similar tendency was observed with false-negative readings, 53.4% of which (87 of 163 observations) were given level 1 classifications, while 46.6% of which (76 of 163 observations) were given level 2 classifications.
Estimation of Extent of GGO at Thin-Section CT
The average extent of GGO within the BAC nodules, as estimated at thin-section CT, was 66.5% ± 28.1 (SD); the average extent within the non-BAC nodules was 29.2% ± 24.7. There was good agreement between the observers in estimating the percentage of the lesion which showed GGO; the maximal disagreement between the three observers was 25% (ICC, R = 0.868, P = .001).
Estimation of Calcification at Thin-Section CT
No calcification was observed within the tumors on thin-section CT images photographed with a mediastinal window.
Analysis of Detection on the Basis of Tumor Size and Extent of GGO Estimated at Thin-Section CT
According to the results of logistic regression analysis, there was no significant interaction between tumor size (P = .015) and extent of GGO (P = .048). On the basis of tumor size and the extent of GGO in the tumor, the tumors were classified into four groups. Group 1 included those tumors 15 mm in diameter or larger in which the extent of GGO was 70% or greater (n = 12) (Fig 1); group 2 included those tumors 15 mm in diameter or larger in which the extent of GGO was less than 70% (n = 31) (Fig 2); group 3 included those tumors smaller than 15 mm in diameter in which the extent of GGO was 70% or greater (n = 10) (Fig 3); and group 4 included those tumors smaller than 15 mm in diameter in which the extent of GGO was less than 70% (n = 22) (Fig 4). Three (25%) group 1 lesions, four (13%) group 2 lesions, seven (70%) group 3 lesions, and six (27%) group 4 lesions were regarded as subtle (ie, four or fewer observers, out of five experienced observers and five residents, could detect them and rated them as 3 or 4). On the other hand, 75% of group 1 lesions, 87% of group 2 lesions, 30% of group 3 lesions, and 72% of group 4 lesions were regarded as conspicuous (ie, five or more observers, out of five experienced observers and five residents, could detect them and rated them as 3 or 4). Lesions in group 3 (ie, those smaller than 15 mm in diameter in which the extent of GGO was 70% or greater) were more difficult to detect on chest radiographs than those in the other three groups (2 = 8.13, df = 1, P = .004); however, the differences among the other three groups were ambiguous.
|
|
|
|
|
|
|
|
|
DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODS RESULTS DISCUSSIONREFERENCES |
|---|
Our study also demonstrated that chest radiography is less sensitive for the detection of localized BAC than it is for the detection of other types of lung cancer, the majority of which were adenocarcinomas. This is because adenocarcinoma is the most common lung carcinoma in Japan, and almost all adenocarcinomas are peripherally located. Previous reports (13,15) have indicated that localized BAC has a greater extent of GGO than other adenocarcinomas (13). In our study, the average extent of GGO within BAC nodules, as estimated at thin-section CT, was 66.5% ± 28.1; the average extent of GGO within non-BAC nodules was 29.2% ± 24.7. Because localized BAC nodules have a greater extent of GGO, they are more difficult to detect on chest radiographs than are other carcinomas.
Other previous studies have shown that small peripheral lung cancers are often difficult to detect on chest radiographs for various reasons. Goldmeier (7) reported that malignant solitary nodules could not be detected on chest radiographs before they reached about 10 mm in diameter. Austin et al (9) reported that the mean diameter of tumors that were missed on chest radiographs was 16 mm ± 8. They also reported that the predominant characteristics of patients whose lung cancer was missed at radiography were female sex and location of the cancer in an upper lobe, especially on the right side (9). Kundel et al (23) defined three forms of false-negative radiologic perceptual error: search, recognition, and decision-making errors. Berbaum et al (24) have reported that satisfaction of search is the cause of lesions being missed at radiography.
In our study, true-positive and true-negative readings were more likely to have been made with a high level of confidence (ie, with a level 4 or level 1 rating). In contrast, false-positive readings were likely to have been made with a low level of confidence (eg, level 3). However, false-negative readings did not show such a tendency. It is possible that because this study included faint opacities, observers misinterpreted some of them with high level of confidence.
No statistically significant difference was seen in detection rate with regard to the observers’ level of expertise in chest radiograph interpretation (P = .063). The agreement among all observers was high (ICC = 0.626). From this point of view, the detection of small peripheral lung cancers on chest radiographs was estimated correctly. It should be noted, however, that the observers were aware that they were looking for small lung tumors. Therefore it is likely that the rate of detection of tumors was greater in the current study than would be seen in clinical practice. In fact, the specificity was relatively low, and the agreement of specificities among the observers was low (ICC = 0.172).
Our study had several limitations. It was retrospective, and, because the study interval was long, it included several screen-film systems. It is possible that advances in screen-film systems for digital radiography may result in increased sensitivity for the detection of small lung nodules. We did not analyze the causes of the misinterpretation of images, because it was difficult to estimate the causes objectively; however, we had an impression that the most of the cases of misinterpretation were due to incorrect summation of the original structure.
In conclusion, chest radiography is less sensitive for the detection of localized BAC compared with other types of lung cancer, most of which, in our study, were adenocarcinomas. Lesions that are smaller than 15 mm in diameter and in which the extent of GGO is 70% or greater are difficult to detect on chest radiographs.
|
ACKNOWLEDGMENTS |
|---|
|
FOOTNOTES |
|---|
Author contributions: Guarantor of integrity of entire study, M.T.; study concepts and design, M.T., K.K., S.K.; literature research, M.T.; clinical studies, J.A., N.K., T.J., N.T., O.H.; data acquisition, M.T.; data analysis/interpretation, M.T., S.K.; statistical analysis, M.T., S.K.; manuscript preparation, M.T., K.K.; manuscript definition of intellectual content and editing, M.T.; manuscript revision/review, M.T., K.K., J.A., C.K.; manuscript final version approval, M.T.
|
REFERENCES |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODS RESULTS DISCUSSIONREFERENCES |
|---|
This article has been cited by other articles:
|
|
 |
|
  K Ono, T Yoshitake, K Akahane, Y Yamada, T Maeda, M Kai, and T Kusama Comparison of a digital flat-panel versus screen-film, photofluorography and storage-phosphor systems by detection of simulated lung adenocarcinoma lesions using hard copy images Br. J. Radiol., October 1, 2005; 78(934): 922 - 927. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 J. K. Leader, T. E. Warfel, C. R. Fuhrman, S. K. Golla, J. L. Weissfeld, R. S. Avila, W. D. Turner, and B. Zheng Pulmonary Nodule Detection with Low-Dose CT of the Lung: Agreement Among Radiologists Am. J. Roentgenol., October 1, 2005; 185(4): 973 - 978. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 Y.-H. Yu, W.-H. Hsu, N.-Y. Hsu, T.-Y. Lin, T.-C. Hsia, S.-S. Sun, and C.-H. Kao The Use of Dual Phase 201Tl SPECT for Differentiating Pulmonary Malignancies from Benign Lesions Jpn. J. Clin. Oncol., August 1, 2004; 34(8): 445 - 451. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
L. Berlin Potential Legal Ramifications of Whole-Body CT Screening: Taking a Peek into Pandora's Box Am. J. Roentgenol., February 1, 2003; 180(2): 317 - 322. [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
| RADIOLOGY | RADIOGRAPHICS | RSNA JOURNALS ONLINE |
