Limited value of shape, margin and CT density in the discrimination between benign and malignant screen detected solid pulmonary nodules of the NELSON trial
Dong Ming Xu,Rob J. van Klaveren,Geertruida H. de Bock,Anne Leusveld,Yingru Zhao,Ying Wang,Rozemarijn Vliegenthart,Harry J. de Koning,Ernst T. Scholten,Johny Verschakelen,Mathias Prokop,Matthijs Oudkerk
DOI: https://doi.org/10.1016/j.ejrad.2007.08.027
IF: 4.531
2008-01-01
European Journal of Radiology
Abstract:Material and methods This study was institutional review board approved. For this study 405 participants of the NELSON lung cancer screening trial with 469 indeterminate or potentially malignant solid pulmonary nodules (>50 mm 3 ) were selected. The nodules were classified based on size, shape (round, polygonal, irregular) and margin (smooth, lobulated, spiculated). Mean nodule density and nodule volume were automatically generated by software. Analyses were performed by univariate and multivariate logistic regression. Results were presented as likelihood ratios (LR) with 95% confidence intervals (CI). Receiver operating characteristic analysis was performed for mean density as predictor for lung cancer. Results Of the 469 nodules, 387 (83%) were between 50 and 500 mm 3 , 82 (17%) >500 mm 3 , 59 (13%) malignant, 410 (87%) benign. The median size of the nodules was 103 mm 3 (range 50–5486 mm 3 ). In multivariate analysis lobulated nodules had LR of 11 compared to smooth; spiculated nodules a LR of 7 compared to smooth; irregular nodules a LR of 6 compared to round and polygonal; volume a LR of 3. The mean nodule CT density did not predict the presence of lung cancer (AUC 0.37, 95% CI 0.32–0.43). Conclusion In solid non-calcified nodules larger than 50 mm 3 , size and to a lesser extent a lobulated or spiculated margin and irregular shape increased the likelihood that a nodule was malignant. Nodule density had no discriminative power. Keywords Pulmonary nodules Margin Shape Density Multi-detector CT Lung cancer Screening 1 Introduction Lung cancer is today the most frequent cause of cancer deaths in the world [1] . In the United States, it was estimated that 171,600 people were diagnosed with this disease in 1999 (14.0% of the total number of cancer diagnosed), and about 158,900 people had died of this disease (28.2% of the total cancer death) [2] . More than two-thirds of these patients are diagnosed with locally advanced or metastatic disease, and their poor prognosis is due to late diagnosis and lack of effective treatment of metastatic disease. Less than 15% of the patients are surviving 5 years, and in several European countries the 5-year survival is far less [3] . Despite the optimal use of therapeutic resources, overall improvement in survival during the last decades has only been modest and major reductions in lung cancer mortality can only come from primary prevention, early detection and truly innovative treatments. Multi-detector low-dose spiral computer tomography (MDCT) is a relatively new technology that offers great potential as a screening tool because it is fast, potentially very accurate and relatively non-invasive. Therefore, the technology can be used for high throughput population screening purposes. Since the first publications of the use of the spiral CT in the early diagnosis of lung cancer, and especially the report from the Early Lung Cancer Action Program (ELCAP) in 1999 [4] , the interest in lung cancer screening by low-dose spiral CT has increased tremendously. In several screening studies it has been demonstrated that 55–80% of lung cancers detected by spiral CT screening are at a very early stage [5–8] . With the increasing use of the MDCT technology for lung cancer screening and the use of thinner slices, the number of screen detected nodules has increased substantially, which poses clinicians often for diagnostic dilemmas. The Fleischner Society [9] has recently issued recommendations for periodic follow-up scanning for this type of nodules, depending on their initial size. According to this recommendation, a nodule is benign if no growth has been observed after 24 months. Once growth has been demonstrated the likelihood of malignancy increases, but also benign lesions may grow [10] . If it were possible to identify subgroups of small pulmonary nodules less likely to be malignant, the follow-up rate for this type of lesions could be reduced. MDCT technology has not only led to an increased detection rate of pulmonary nodules, but at the same time improvements in spatial resolution and scanning time have given us the possibility to better analyse the internal and external characteristics of the nodules detected. Although many of these characteristics are already known for solitary pulmonary nodules (SPN) which are usually larger than 1 cm [11] , their applicability in the discrimination between benign and malignant in sub-centimeter solid screen detected pulmonary nodules is yet unknown. Purpose of this study is to evaluate prospectively the value of size, shape, margin and mean density in the discrimination between benign and malignant solid nodules detected at baseline screening of the Dutch-Belgian randomised lung cancer screening trial (NELSON). 2 Materials and methods 2.1 Participants The participants of the NELSON study underwent baseline screening for lung cancer by low-dose multi-detector CT. They were between 50 and 75 years of age and were recruited via population registries through mail. They had to be current or former smokers with a smoking history of >15 cigarettes/day for >25 years or >10 cigarettes/day for >30 years. Ineligible were those applicants who already underwent a pneumonectomy and those with a history of breast cancer, melanoma or hypernephroma. People with a history of other cancers were only eligible if curatively treated at least 5 years ago without signs of recurrence at the time of inclusion. The NELSON study was approved by the Medical Ethical Committees of all institutions and all subjects provided written informed consent. From April 2004 until July 2006, 7310 participants underwent baseline screening. For this particular study a subgroup of 405 participants with 387 indeterminate solid pulmonary nodules between 50 and 500 mm 3 (≤0 mm in diameter) and 82 screen-positive solid pulmonary nodules with a volume > 500 mm 3 (>10 mm in diameter) have been selected. The nodules had to be located purely intra-parenchymal without pleural-, fissure-attachment or juxtavascular location. For all these nodules either a 1-year follow-up or histological proof of benignancy or malignancy was available. 2.2 Data acquisition At all four screening sites 16-detector MDCT scanners were used (Mx8000 IDT or Brilliance 16P, Philips Medical Systems, Cleveland, OH, USA, or Sensation-16, Siemens Medical Solutions, Forchheim, Germany). Scanning of the entire chest was performed in caudo-cranial direction. Scan data were obtained in spiral mode, with 16 mm× 0.75 mm collimation and pitch 1.5. No contrast was used. Low-dose settings were applied. Depending on the body weight (<50, 50–80 and >80 kg), the kVp settings were 80–90, 120 and 140 kVp, respectively, to achieve a Computed Tomography Dose Index Volume (CTDIvol) of approximately 0.8, 1.6 and 3.2 mGy, respectively. The mAs settings were adjusted accordingly dependent on the machine used. To minimize breathing artefacts, scans were performed at suspended maximal inspiration after appropriate instruction of the subjects. Data were reconstructed at 1.0 mm slice thickness, with 0.7 mm reconstruction increment. 2.3 Image analysis All CT images were read twice independently. First readings were done by a radiologist with an experience in reading thoracic CT scans varying from 1 year to more than 20 years. Second readings were done by radiologists with 6 years of experience. In case of a discrepancy between the first and second reader, a third radiologist with more than 15 years of experience with lung CT made the final decision. The Syngo Lungcare © (Leonardo © workstation, Somaris/5 VB 10A Siemens Medical Solutions, Erlangen, Germany) software package designed to aid the radiologist in the diagnosis of pulmonary nodules was used in addition to visual readings. Baseline and follow-up images were reviewed and displayed simultaneously on one workstation. All images were interpreted both at lung window and mediastinal settings. 2.4 Nodule features The non-calcified solid nodules detected at baseline screening were classified into four different categories based on size and benign characteristics as described earlier [12] . Indeterminate nodules (Category III) were defined as nodules with volumes between 50 and 500 mm 3 , and screen-positive nodules (Category IV) as nodules larger than 500 mm 3 . In this study, all solid nodules were further classified based on shape (round/oval, polygonal or irregular) [13] , margin (smooth, lobulated or spiculated/irregular) [14] , and mean CT density in Hounsfield units (HU) (≤0, 0–100 or >100 HU). A nodule was polygonal when the entire lesion surface was surrounded by concave margins only [15] . Spiculation was defined as the presence of thicker strands extending from the nodule margin into the lung parenchyma without reaching the pleural surface [16] . Lobulation was defined as an abrupt bulging of the lesion contour [17] . Mean CT density of the nodule was calculated automatically by the Lungcare © software instead of specifying a region of interest. Volumes were calculated by three-dimensional (3D) volumetric computer software (Lungcare © software). In case of inappropriate segmentation, the radiologist was able to enter manual measurements as well, which then overruled the automatically generated volume calculations as described earlier [12] . Because cancer risk was absent at 1 year of follow-up in indeterminate solid nodules attached to either a vessel, the pleura or a fissure [18] , only purely intra-parenchymal indeterminate solid nodules and screen positive nodules have been included in this study. Nodules were classified as benign or malignant based on histological examinations of trans-thoracic needle biopsies or examination of surgical specimens. Nodules were also classified as benign if the volume doubling time (VDT) was >600 days or if the volume decreased at 1 year of follow-up. 2.5 Statistical analysis The Chi-square test for independence was used for comparisons of the presence of lung cancer among nodules within different categories of size, shape, margin or density. Univariate and multivariate logistic regression analyses were used to investigate the effect of the different nodule characteristics on the presence of lung cancer. Results are presented as likelihood ratios (LR) with 95% confidence intervals. Because mean nodule density and size were not normally distributed, a logarithmic transformation was performed before statistical analysis. Correlation analysis between density and nodule volume for lung cancer positive and negative cases was performed by Pearson's test. Receiver operating characteristic (ROC) analysis was performed for density as a predictor for the presence of lung cancer. p values < 0.05 were considered statistically significant. All values are expressed as means ± standard deviation (S.D.) and all analyses have been performed in SPSS version 14.0. 3 Results Between April 2004 and July 2006, in 405 participants 469 solid purely intra-parenchymal nodules of Category III and IV were detected which met the inclusion criteria. The mean age of the participants was 62 years (±5 years), 93% were males, 7% females. Of these 469 nodules, 387 (82%) were Category III and 82 (18%) Category IV, 59 (13%) malignant and 410 (87%) benign. The median size of the nodules was 103 mm 3 (range 50–5486 mm 3 ; median diameter 4.7 mm). The median CT density for benign nodules was 39 HU (range, −195 to 192 HU), and 11 HU (range, −128 to 7 HU) for malignant nodules ( p = 0.002). In Table 1 the distribution of the nodule characteristics is presented. For all variables except for nodule density, there was correlation with lung cancer ( p = 0.000), which was confirmed in univariate analysis ( Table 2 ). Lobulated, spiculated, irregular and Category IV nodules had an increased LR for lung cancer compared to smooth, round or polygonal and Category III nodules, respectively. The 95% CI intervals were very large, however, except for tumor volume. In multivariate analysis only tumor volume remained highly significant with a small 95% CI, while margin and shape just reached statistical significance, again with very wide 95% CI's ( Table 3 ). The ROC curve ( Fig. 1 ) illustrates that mean nodule CT density could not be used as a predictor for lung cancer (AUC 0.37, 95% CI 0.32–0.43). For lung cancer positive cases, the correlation coefficient between tumor volume and mean density was −0.05 ( p = ns), and for lung cancer negative cases 0.06 ( p = ns) ( Fig. 2 A and B ). Of the 469 solid nodules, 72 had a round contour, smooth margin and negative CT density (<0 HU). None of them became malignant at 1 year of follow-up. 4 Discussion The differentiation of benign form malignant pulmonary nodules detected within the context of a lung cancer screening programme is of great importance in order to reduce costs, anxiety and unnecessary additional investigations. It is therefore a great challenge to the specialist because the nodule remains frequently indeterminate at the end of the diagnostic process and surgical resection is often needed before a definitive diagnosis can be made. In a previous study, Xu et al. have demonstrated that in solid indeterminate nodules between 50 and 500 mm 3 lung cancer risk is absent at 1 year of follow-up in nodules attached to either a vessel, the pleura or a fissure while all lung cancer cases originated from purely intra-parenchymal nodules without attachment to these structures [18] . In the current study we investigated prospectively whether size, margin, shape and CT density could help in the discrimination between benign and malignant nodules. Although univariate analysis showed that lobulated, spiculated, irregular and Category IV nodules had an increased likelihood for lung cancer compared to smooth, round or polygonal and Category III nodules, respectively, the 95% CI intervals were very large, except for tumor volume. In multivariate analysis only tumor volume (size) remained highly significantly associated with the presence of lung cancer with small 95% CI's, while margin and shape just reached statistical significance, again with wide 95% CI's. Furthermore, we demonstrated that nodule density could not be used as a predictor, neither for lung cancer negative nor for lung cancer positive cases, although the density of malignant nodules appeared to be significant lower than that of benign nodules. The correlation between size and cancer risk found in our study has also been reported by other investigators [6,19–21] . The chance of being malignant is 0.9% for nodules between 4 and 7 mm, 18% for nodules 8–20 mm and 50% for those larger than 20 mm, while cancer risk is <0.2% for nodules less than 3 mm in diameter [9] . However, other features in addition to size can be helpful in the discrimination between benign and malignant pulmonary nodules [11,22] . The higher LR of cancer in nodules with a spiculated margin compared to round or polygonal nodules in our study is consistent with the results of previous studies in SPN's [22–24] , in which a spiculated contour occurs significantly more often in malignant SPNs. However, 95% CI intervals were wide, also limiting its use for SPN's in clinical practice. In pathological studies a spiculated contour was found to be due to thickened interlobular septa, fibrosis caused by obstruction of pulmonary vessels or lymphatic channels filled with tumor cells or extension to the pleura with pleural infiltration [23,24] . Nevertheless, also in benign SPN such as in inflammatory pseudotumors or tuberculomas spiculated edges may be found due to a desmoplastic response. With regard to the shape of benign and malignant SPN there is no agreement in the literature. Seemann reported that both benign and malignant SPN's could exhibit smooth edges [11,23] , while others found round or polygonal margins more often in benign lesions [16,22,25] . The pathologic substrate for lobulation in malignant lesions is the presence of nodular excrescences of the tumor at its advancing edge or tumor growth in islands at the periphery of the nodule, indicating that the nodule has uneven growth rates [10,24] . In our study of solid non-calcified pulmonary nodules, we found that the cancer cases had a significant lower mean density compared to benign nodules ( p < 0.05) most likely due to a lepidic growth pattern, the inclusion of air or an air bronchogram or regressive changes such as necrosis or hemorrhage within the tumor. This was consistent with the results of studies performed in usually larger SPN's [11,23,26,27] . Despite these findings, we do not recommend the use of nodule density in the differentiation between malignant and benign nodules because of the large overlap and the wide range of densities observed among the malignant nodules. In multivariate analysis nodule density was also not significant. Other investigators [11,23] also concluded that benign and malignant SNP's showed large variability in nodule density, particularly in the malignant group, with large overlap, and that size, shape and margin had only limited value in the identification of malignant from benign pulmonary nodules. Siegelman et al. suggested that a representative density value above 164 HU could be used to separate benign from malignant lesions, primarily because of diffuse calcifications in benign nodules [28] . However, this conclusion was not supported by the study by Seemann et al. [11] , who demonstrated that both benign and malignant solitary pulmonary nodules showed large range in lesion densities, particularly in the malignant group, with large overlap. Our results in small screen detected nodules are more or less consistent with the conclusions for larger SPN's, showing that size, shape and margin had only limited value in the identification of malignant from benign pulmonary nodules [23–25] . In the literature, there are several nodule classifications in use based on the external and internal features of the nodules, such as ragged, densely spiculated, somewhat irregular with slight spiculation or irregular with spiculation, but so far no standard classification is being used. According to our experience, some of these classifications are too complex and subjective for research purposes. For example, it is difficult to differentiate ragged nodule from lobulated and the same holds true for slightly and densely spiculated or coarse spiculation. This was also concluded by Li et al. who found large variability in the classification of the CT features by three different radiologists, especially for the margins of the nodule [25] . In order to minimize the inter-observer variability, we used a more simplified nodule classification as described in the method section. Future studies should focus on the further development of computer aided diagnosis (CAD) algorithms for small screen detected pulmonary nodules [29] . In our study, none of the round spherical nodules with a smooth margin and negative CT density became malignant at 1 year of follow-up, but the numbers are yet too small to draw final conclusions. Also changes in nodule shape and density over time might be associated with malignancy, and need further investigation. In conclusion, in solid non-calcified nodules larger than 50 mm 3 , size and to a lesser extent lobulated or spiculated margin and irregular shape increased the likelihood that the nodule was malignant. Nodule density had no discriminative power. Conflict of interest Not reported. Acknowledgements We would like to thank Peter Paul Taverne, Professor. Dr. Kris Nackaerts, René Vernhout, Estelle J.K. Noach, Wim Tukker, Peter van Ooijen and Fenneke van de Wiel for their efforts concerning data collection and management, providing us the population data from the central NELSON databases, English editing of the manuscript, data archiving and transferring. References [1] A. Pasic P.E. Postmus T.G. Sutedja What is early lung cancer? A review of the literature Lung Cancer 45 3 2004 267 277 [2] S.H. Landis T. Murray S. Bolden P.A. Wingo Cancer statistics, 1999 CA Cancer J Clin 49 1 1999 8 31 [3] R.J. van Klaveren J.D.F. Habbema J.H. Pedersen H.J. de Koning M. Oudkerk H.C. Hoogsteden Lung cancer screening by low-dose spiral computed tomography Eur Respir J 18 5 2001 857 866 [4] C.I. Henschke D.I. McCauley D.F. Yankelevitz Early Lung Cancer Action Project: overall design and findings from baseline screening Lancet 354 9173 1999 99 105 [5] S. Diederich D. Wormanns M. Semik Screening for early lung cancer with low-dose spiral CT: prevalence in 817 asymptomatic smokers Radiology 222 3 2002 773 781 [6] J.K. Gohagan P.M. Marcus R.M. Fagerstrom Final results of the Lung Screening Study, a randomized feasibility study of spiral CT versus chest X-ray screening for lung cancer Lung Cancer 47 1 2005 9 15 [7] J.L. Mulshine D.C. Sullivan Clinical practice. Lung cancer screening N Engl J Med 352 26 2005 2714 2720 [8] S.J. Swensen J.R. Jett T.E. Hartman Lung cancer screening with CT: Mayo Clinic experience Radiology 226 3 2003 756 761 [9] H. MacMahon J.H. Austin G. Gamsu Guidelines for management of small pulmonary nodules detected on CT scans: a statement from the Fleischner Society Radiology 237 2 2005 395 400 [10] H.T. Winer-Muram The solitary pulmonary nodule Radiology 239 1 2006 34 49 [11] M.D. Seemann A. Staebler T. Beinert Usefulness of morphological characteristics for the differentiation of benign from malignant solitary pulmonary lesions using HRCT Eur Radiol 9 3 1999 409 417 [12] D.M. Xu H. Gietema H.J. de Koning Nodule management protocol of the NELSON randomised lung cancer screening trial Lung Cancer 54 2 2006 177 184 [13] S. Matsuoka Y. Kurihara K. Yagihashi H. Niimi Y. Nakajima Peripheral solitary pulmonary nodule: CT findings in patients with pulmonary emphysema Radiology 235 1 2005 266 273 [14] J.W. Gurney Determining the likelihood of malignancy in solitary pulmonary nodules with Bayesian analysis. Part I. Theory Radiology 186 2 1993 405 413 [15] S. Takashima S. Sone F. Li Y. Maruyama M. Hasegawa M. Kadoya Indeterminate solitary pulmonary nodules revealed at population-based CT screening of the lung: using first follow-up diagnostic CT to differentiate benign and malignant lesions Am J Roentgenol 180 5 2003 1255 1263 [16] S. Takashima S. Sone F. Li Small solitary pulmonary nodules (<or=1 cm) detected at population-based CT screening for lung cancer: reliable high-resolution CT features of benign lesions Am J Roentgenol 180 4 2003 955 964 [17] S. Takashima Y. Maruyama M. Hasegawa CT findings and progression of small peripheral lung neoplasms having a replacement growth pattern Am J Roentgenol 180 3 2003 817 826 [18] Xu DM, Zaag-Loonen vd HJ, Oudkerk M, et al. Cancer risk during one year of follow-up in smooth or attached solid indeterminate nodules detected at baseline CT screening in the NELSON study. Radiology, in press. [19] C.I. Henschke D.F. Yankelevitz D.P. Naidich CT screening for lung cancer: suspiciousness of nodules according to size on baseline scans Radiology 231 1 2004 164 168 [20] S.J. Swensen J.R. Jett J.A. Sloan Screening for lung cancer with low-dose spiral computed tomography Am J Respir Crit Care Med 165 4 2002 508 513 [21] S.J. Swensen J.R. Jett T.E. Hartman CT screening for lung cancer: five-year prospective experience Radiology 235 1 2005 259 265 [22] K. Furuya S. Murayama H. Soeda New classification of small pulmonary nodules by margin characteristics on high-resolution CT Acta Radiol 40 5 1999 496 504 [23] M.D. Seemann O. Seemann W. Luboldt Differentiation of malignant from benign solitary pulmonary lesions using chest radiography, spiral CT and HRCT Lung Cancer 29 2 2000 105 124 [24] C.V. Zwirewich S. Vedal R.R. Miller N.L. Muller Solitary pulmonary nodule: high-resolution CT and radiologic–pathologic correlation Radiology 179 2 1991 469 476 [25] F. Li S. Sone H. Abe H. MacMahon K. Doi Malignant versus benign nodules at CT screening for lung cancer: comparison of thin-section CT findings Radiology 233 3 2004 793 798 [26] J.C. Wang S. Sone L. Feng Rapidly growing small peripheral lung cancers detected by screening CT: correlation between radiological appearance and pathological features Br J Radiol 73 873 2000 930 937 [27] Z. Yang S. Sone S. Takashima F. Li T. Honda T. Yamanda Small peripheral carcinomas of the lung: thin-section CT and pathologic correlation Eur Radiol 9 9 1999 1819 1825 [28] S.S. Siegelman E.A. Zerhouni F.P. Leo N.F. Khouri F.P. Stitik CT of the solitary pulmonary nodule Am J Roentgenol 135 1 1980 1 13 [29] F. Li M. Aoyama J. Shiraishi Radiologists’ performance for differentiating benign from malignant lung nodules on high-resolution CT using computer-estimated likelihood of malignancy Am J Roentgenol 183 5 2004 1209 1215