Eur Radiol DOI 10.1007/s00330-015-3816-y

CHEST

Quantitative CT analysis of pulmonary ground-glass opacity nodules for distinguishing invasive adenocarcinoma from non-invasive or minimally invasive adenocarcinoma: the added value of using iodine mapping Ji Ye Son 1 & Ho Yun Lee 1 & Jae-Hun Kim 1 & Joungho Han 2 & Ji Yun Jeong 2,5 & Kyung Soo Lee 1 & O. Jung Kwon 3 & Young Mog Shim 4

Received: 8 January 2015 / Revised: 13 April 2015 / Accepted: 21 April 2015 # European Society of Radiology 2015

Abstract Objectives To determine whether quantitative analysis of iodine-enhanced images generated from dual-energy CT (DECT) have added value in distinguishing invasive adenocarcinoma from non-invasive or minimally invasive adenocarcinoma (MIA) showing ground-glass nodule (GGN). Methods Thirty-four patients with 39 GGNs were enrolled in this prospective study and underwent DECT followed by complete tumour resection. Various quantitative imaging parameters were assessed, including virtual non-contrast (VNC) imaging and iodine-enhanced imaging. Results Of all 39 GGNs, four were adenocarcinoma in situ (AIS) (10 %), nine were MIA (23 %), and 26 were invasive adenocarcinoma (67 %). When assessing only VNC imaging,

* Ho Yun Lee [email protected] * Young Mog Shim [email protected] 1

Department of Radiology and Center for Imaging Science, Samsung Medical Center, Sungkyunkwan University School of Medicine, 81 Irwon-Ro, Gangnam-gu Seoul 135-710, Korea

2

Department of Pathology, Samsung Medical Center, Sungkyunkwan University School of Medicine, Seoul 135-710, Korea

3

Division of Respiratory and Critical Medicine of the Department of Internal Medicine, Samsung Medical Center, Sungkyunkwan University School of Medicine, Seoul 135-710, Korea

4

Department of Thoracic and Cardiovascular Surgery, Samsung Medical Center, Sungkyunkwan University School of Medicine, 81 Irwon-Ro, Gangnam-gu Seoul 135-710, Korea

5

Department of Pathology, Kyungpook National University Medical Center, Kyungpook National University School of Medicine, Daegu 702-210, Korea

multivariate analysis revealed that mass, uniformity, and sizezone variability were independent predictors of invasive adenocarcinoma (odds ratio [OR]=19.92, P=0.02; OR=0.70, P= 0.01; OR =16.16, P = 0.04, respectively). After assessing iodine-enhanced imaging with VNC imaging, both mass on the VNC imaging and uniformity on the iodine-enhanced imaging were independent predictors of invasive adenocarcinoma (OR=5.51, P=0.04 and OR=0.67, P<0.01). The power of diagnosing invasive adenocarcinoma was improved after adding the iodine-enhanced imaging parameters versus VNC imaging alone, from 0.888 to 0.959, respectively (P=0.029). Conclusion Quantitative analysis using iodine-enhanced imaging metrics versus VNC imaging metrics alone generated from DECT have added value in distinguishing invasive adenocarcinoma from AIS or MIA. Key Points • Quantitative analysis using DECT was used to distinguish invasive adenocarcinoma. • Tumour mass and uniformity were independent predictors of invasive adenocarcinoma. • Diagnostic performance was improved after adding iodine parameters to VNC parameters. Keywords Pure ground glass opacity . Lung adenocarcinoma . Dual-energy CT . Iodine quantification . Histogram analysis

Abbreviations AIS Adenocarcinoma in situ ATS American Thoracic Society AUC Area under the receiver operating characteristic curve CT Computed tomography

Eur Radiol

DECT ERS GGN HU IASLC MIA OR ROC ROI VNC

Dual-energy CT European Respiratory Society Ground-glass opacity nodule Hounsfield unit International Association for the Study of Lung Cancer Minimally invasive adenocarcinoma Odds ratio Receiver operating characteristic Region of interest Virtual non-contrast

Introduction The persistent presence of ground-glass opacity nodules (GGN) on thin-section computed tomography (CT) usually suggests the presence of lung adenocarcinoma or its precursors, which include atypical adenomatous hyperplasia, adenocarcinoma in situ (AIS), and minimally invasive adenocarcinoma (MIA). According to one report, invasive adenocarcinomas may be seen on CT as large, pure GGNs greater than 16 mm in diameter [1]. Even with invasive characteristics (myofibroblastic changes, no vascular or lymphatic components), MIAs and invasive adenocarcinomas may still appear as pure GGNs on thin-section CT, making it difficult to distinguish them without further study or more invasive procedures. To identify pure GGN or part solid nodules with little solid component as invasive adenocarcinoma versus AIS or MIA, our previous study [2] revealed some significantly meaningful imaging variables using quantitative analysis of conventional non-contrast CT imaging metrics. We Fig. 1 Diagram of material decomposition (a), (c), and (d) are fixed points for constant density of air, fat, and soft tissue, respectively. Because (a) value could not be input in the Liver VNC application mode of Syngo Dual Energy, (b) was used a substituted value for (a) as material parameter. (x) is the degree of enhancement of ground glass opacity nodule. On the other hand, (y) is the degree of enhancement of solid nodule

concluded that although the tumours appear to be pure GGNs on imaging, regional voxel heterogeneity within the tumours is increased as the malignancy progresses. Here, we hypothesized that the heterogeneity of the enhancement of the GGNs can also reflect the evolution of malignancy. Because the degree of enhancement is related to the vascularity of the solid lung nodule, increased enhancement suggests increased malignant tendency, according to many studies [3–5]. However, evaluating the contrast enhancement of pure GGNs is challenging because of their invisibility on the mediastinal window and low cellularity, making it difficult to place the region of interest (ROI) without misregistration on pre-contrast and post-contrast images. In this study, we quantified the enhancement of GGNs using dual-energy CT (DECT), which simultaneously offers a virtual non-contrast (VNC) and an iodine-enhanced image from a single examination after the administration of iodine contrast material [6, 7]. Therefore, we were able to obtain direct quantification of the iodine concentration (in mg/ml) in a lesion, even in a pure GGN. Thus, the aim of the present study was to determine whether quantitative analysis of iodine-enhanced images generated from DECT has added value in distinguishing invasive adenocarcinoma from AIS or MIA in pure GGNs or part solid nodules with little solid components.

Materials and methods Our institutional review board approved this single-centre prospective study, and informed consent was obtained from all patients.

Eur Radiol

Patients This study was performed as a part of an ongoing prospective clinical trial in which patients with early stage lung adenocarcinoma were enrolled and underwent pre-operative DECT scans beginning in 2011 (NCT01482585). Among the patients, we included only those with pure GGNs or part solid nodules with little solid components on pre-operative DECT imaging and those who were scheduled to undergo curative operation. Patients were excluded if they had GGNs showing≥5 mm in diameter of solid components in the mediastinal window virtual non-contrast (VNC) image generated from

Fig. 2 A 58-year-old male patient with adenocarcinoma in situ in the left upper lobe (a) VNC images show pure GGN (upper row), which disappears on the mediastinal setting image (not seen). The iodine map shows the degree of contrast enhancement of the pure GGN (lower row) (b) The two marginal histograms represent the VNC image-driven histogram graph (horizontal axis) and iodine-enhanced image-driven histogram graph (vertical axis) As compared to the VNC imagedriven histogram graphs, iodine-enhanced image-driven histogram

DECT because these nodules are usually considered as invasive adenocarcinomas [8, 9]. Overall, 34 patients with 39 nodules with little or no solid components were enrolled in our study. Imaging and analysis All patients underwent a CT examination using a dual-source CT system (Somatom Definition Flash; Siemens Healthcare, Forchheim, Germany) with the dual-energy technique. This DECT system was composed of two X-ray tubes and two corresponding 128-row detectors mounted in a perpendicular

graphs show a tendency to be less uniformed, as visualized in the scatter plot (c) Intensity size-zone matrix of AIS on VNC imaging (left) and iodine-enhanced imaging (right) The horizontal axis shows the size of the homogenous area. Divergent distribution on the horizontal axis indicates increased size-zone variability. The vertical axis shows intensity. Divergent distribution on the vertical axis indicates increased intensity variability (d) Pathological diagnosis was confirmed as adenocarcinoma in situ

Eur Radiol

arrangement. DECT scanning was obtained 90 seconds after the administration of contrast material (100 mL of iopamidol: Iomeron 300; Bracco, Milan, Italy) at a rate of 1.5 mL/sec using a power injector. This was followed by a 20 cc saline flush at a rate of 1.5 mL/sec. Imaging parameters were as follows: 105 mAs (effective) at 140 kV, 248 mAs (effective) at 80 kV, 32×0.6-mm collimation, a pitch of 0.7, a rotation time of 0.5 second, and a 512×512-pixel matrix. Imaging was performed from the thoracic inlet to the middle portion of the kidneys. Three different data sets were generated from the DECT imaging: the 80 kV, 140 kV, and enhanced weightedaverage images. The weighted-average images were generated by combining the 140-kV and 80-kV data sets with a weighting factor of 0.6 (60 % of the information derived from the 80 kV image and 40 % derived from the 140 kV image), and these were approximately 120 kV images. The VNC images and iodine-enhanced images (Fig. 1) were created from modifying the liver VNC application mode of the dedicated dual-energy post-processing

Fig. 2 continued.

software (Syngo Dual Energy; Siemens Medical Solutions, Forchheim, Germany). Since the GGNs are composed of a mixture of air and soft tissue, the Hounsfield Unit (HU) value of fat in the liver VNC application mode was replaced with air, which is a HU value at the interconnecting line between air and soft tissue [10, 11]. Thus, the material parameters were -110 HU for air at 80 kV, -115 HU for air at 140 kV, 60 HU for soft tissue at 80 kV, and 54 HU for soft tissue at 140 kV. Imaging data were reconstructed with a section thickness of 1 mm using a D30f (medium smooth) kernel for the iodineenhanced images and a D45f (medium sharp) kernel for the VNC images. CT scans were assessed independently by two chest radiologists (J.Y.S. and H.Y.L., with 3 and 12 respective years of experience in thoracic CT interpretation), who were unaware of the clinical and pathologic results. Tumour size and volume were measured on VNC imaging. Density, mass, CT attenuation values at the 75th and 97.5th percentiles on the histogram,

Eur Radiol

texture parameters (uniformity, entropy, intensity variability and size-zone variability), and skewness/kurtosis were calculated on both VNC images and iodine-enhanced images. For nodule segmentation, a ROI covering the largest possible area of each whole tumour was drawn on the VNC images. The ROI was drawn freehand around the tumour using an electronic cursor and mouse. Large vessels and pulmonary arteries were excluded from the ROIs. This process was repeated for each contiguous transverse level until the entire tumour had been

covered. Next, voxel-based CT numbers were collected from the lesion segmentation. Segmentation masks obtained on VNC images were overlapped on iodineenhanced images of each nodule, and voxel-based CT numbers were collected within this ROI on the iodine map. To measure tumour density and volume, the computer automatically calculated the density from the mean attenuation of the total voxels and volume by multiplying the number of voxels by the unit volume of a voxel [12]. Tumour mass (in

Fig. 3 A 56-year-old female patient with invasive adenocarcinoma in the left upper lobe (a) VNC images show pure GGN (upper row), which disappears on the mediastinal setting image (not seen). The iodine map shows the degree of contrast enhancement of the pure GGN (lower row) (b) Two marginal histograms represent the VNC image-driven histogram graph (horizontal axis) and iodine-enhanced image-driven histogram graph (vertical axis) As compared to the VNC image-driven histograms, the iodine-enhanced image-driven histograms show homogeneous tendency, as visualized on scatter plot. Compared to a scatter plot of AIS (Fig. 2b),

the scatter plot of invasive adenocarcinoma shows a broader distribution of attenuation values (c) Intensity size-zone matrix of the invasive adenocarcinoma on VNC imaging (left) and iodine-enhanced imaging (right). Compared to the distributions of matrix points for AIS (Fig. 2), the distribution of matrix points for invasive adenocarcinoma shows more divergence both in the horizontal axis and the vertical axis (d) Pathological diagnosis was confirmed to be acinar predominant invasive adenocarcinoma with a 10 mm invasive component

Eur Radiol

grams) was calculated by multiplying tumour volume (in cubic centimetres) by mean tumour density [13]. Texture analysis was performed by a radiology physicist (J.H.K.) with four years of experience. Within this ROI, entropy (irregularity), uniformity (distribution of grey level), intensity variability, and size-zone variability were calculated to evaluate tumour heterogeneity. Entropy, uniformity, intensity variability, and size-zone variability are all potential indicators of heterogeneity of the tumour. Entropy and uniformity reflect local tumoral heterogeneity whereas intensity variability and size-zone variability reflect regional tumoral heterogeneity. Entropy is a measure of texture irregularity, while uniformity reflects how close the image is to being a uniform distribution of the grey levels, in which higher entropy and lower uniformity represent increased heterogeneity [14]. These parameters are defined below where l is the number of grey levels (e.g., l=1 to k indicates grey level from 1 to k) in

Fig. 3 continued.

and p(l) is the probability of the occurrence of the grey level l based on the image histogram technique: Entropy ðEÞ ¼ −

k X

½pðlÞlog2 ½pðlÞUniformity ðUÞ

l¼1

¼

k X

½pðlÞ2

l¼1

Intensity variability is a measure of differences in density between an individual voxel and its neighbours, describing local tumour heterogeneity proportional to variations of density between individual voxels. Size-zone variability corresponds to regional heterogeneity, such as the variation in density between regions and the variation in the size and alignment of homogeneous areas. For intensity variability and size-zone variability, voxel values within the segmented tumours were

Eur Radiol

resampled to yield 16 discrete values in order to reduce the image noise and normalize the intensity across subjects by clustering voxels with similar intensities [15]. From the discrete tumour images (16 grey levels), the grey level size zone matrix was computed. The value of the matrix’s (m, n) is defined by the number of homogenous regions given the homogeneous tumour size (n) related to their intensity (m). For example, BMatrix’s (3, 5) is 6.^ means that there are six homogenous regions, in which the grey level of each cluster is 3, and the size of each cluster is five voxels. This grey level size zone matrix was used to compute the variability in the size and the intensity of homogeneous tumour regions [15, 16] (Figs. 2 and 3). Next, a histogram of the voxel-based CT numbers was displayed, and CT attenuation values at the 75th and 97.5th percentiles of pixel attenuation value (HU) on the histogram (e.g., a 75th percentile value means a CT attenuation value of the 75 % of pixels calculated from the pixel with the minimum CT attenuation value) of the sorted values was computed [2, 17]. Additionally, a spreadsheet of all values was created, which was used to compute histogram distribution parameters, specifically, kurtosis and skewness. Skewness describes the degree of asymmetry of a histogram; a histogram with a long tail to the right has a positive skewness value, and a perfectly symmetric distribution has a skewness value of zero. Kurtosis describes how sharply peaked a histogram is; a histogram that is more peaked than a normal distribution has a positive kurtosis value, and a normal distribution has a kurtosis of zero [2, 18]. To measure tumour density and volume, and to analyze the texture and the histograms from the CT data in the tumour regions, we developed software that was implemented using MATLAB R2009a (The Mathworks, Natick, MA). Pathologic evaluation Approximately 10 mm of tumour tissue was placed on a slide. All slides were examined to produce high-quality resolution digital images (0.25 lm/pixel at 40·) using the Aperio Slide Scanning System (ScanScope T3; Aperio Technologies Inc., Vista, CA, USA) [19]. Two experienced lung pathologists (J.Y.J. and J.H. with 7 and 19 respective years of experience in lung pathology) jointly interpreted all tissue sections via virtual slides using ImageScope viewing software (Aperio Technologies, Inc.) and a high-resolution monitor [19]. For each case, the specimens were reviewed according to the International Association for the Study of Lung Cancer (IASLC), the American Thoracic Society (ATS), and the European Respiratory Society (ERS) International Multidisciplinary Lung Adenocarcinoma Classification Criteria [20], and comprehensive histological subtyping was performed for the primary tumour in a semi-quantitative manner to the nearest 5 % level, adding up to a total of 100 % subtype components per tumour. The extent of the invasive components was measured, and the predominant subtype

was recorded. When evaluating the predominant pattern, the area of central fibrosis was disregarded. Statistical analysis Patient demographics and CT parameters were compared among the three different pathologic subtypes (AIS, MIA, and invasive adenocarcinoma) by using one-way ANOVAs with Bonferroni post hoc testing. Bonferroni corrections were also used to account for multiple comparisons. When multiple GGNs were present in one patient, we did not account for within-patient correlations because each GGN was considered as an independent synchronous lesion [21]. Multivariate logistic regression analyses were used to identify the independent factors for the stratification of invasive and minimally or non-invasive adenocarcinoma. Characteristics with P values of less than 0.05 on one-way ANOVA were used as the input variables for the multiple logistic regression analysis. First, a Table 1 Clinicopathologic characteristics of lung adenocarcinoma with little solid components on CT (39 tumours of 34 patients) Characteristics Gender (%) Male Female Median age (y) Smoking habits (%) Nonsmoker Current/former smoker p-T status† T1a T1b T2a p-N status N0 N1 Histopathology† AIS MIA Invasive adenocarcinoma Lepidic predominant Acinar predominant Papillary predominant Micropapillary predominant Type of operation† Segmentectomy Lobectomy

Number of patients/nodules

18 (53) 16 (47) 57 (36-71)* 19 (56) 15 (44) 26 (66.5) 10 (25.5) 3 (8) 34 (100) 0 (0) 4 (10) 9 (23) 3 (8) 21 (54) 1 (2.5) 1 (2.5) 15 (38) 24 (62)

Note: Unless otherwise indicated, data in parentheses are percentages. AIS: adenocarcinoma in situ, MIA: minimally invasive adenocarcinoma * Data in parentheses indicate the range † Tumour number (n=39)

12.1±5.4 1.24±1.55 0.33±0.12 1.17±0.03 0.35±0.36 1.49±1.91 0.45±0.71 1.53±1.72 3.45±2.09 10.29±11.21 -481±119 65.8±27.2 -419±126 131.3±21.5 0.0057±0.0042 0.0120±0.0030 8.00±0.97 6.44±0.54 5760±4904 4843±3911 0.26±0.17 0.34±0.42

9.3±6.4 0.43±0.64 0.37±0.14 1.17±0.01 0.19±0.30 0.39±0.53

-0.04±0.19 0.37±0.84 2.08±0.15 3.96±2.54 -529±150 71.9±13.4 -434±160 123.8±10.1

0.0068±0.0035 0.0141±0.0039 7.53±0.97 6.44±0.41 3071±2908 2334±1987 0.09±0.08 0.11±0.13

0.0023±0.0009 0.0085±0.0024 9.07±0.45 7.29±0.37 18068±16316 11890±7486 0.59±0.41 0.57±0.42

0.16±0.53 0.95±0.74 2.73±0.86 5.78±3.34 -351±113 76.1±21.1 -177±137 161.8±35.4

22.2±9.1 4.16±3.60 0.52±0.14 1.17±0.02 1.70±1.27 3.88±3.29

56.7±7.4 13 : 13

Invasive Adenocarcinoma (n=26)

<0.01* <0.01* <0.01* <0.01* 0.03* <0.01* 0.02* 0.07

0.27 0.16 0.13 0.11 <0.01* 0.49 <0.01* 0.01*

<0.01* 0.02* <0.01* 0.95 0.01* 0.03*

0.86 0.15

P

<0.01* <0.01* <0.01* <0.01* 0.08 0.02* 0.04*

<0.01* 0.04*

1.00 1.00 1.00 0.64 0.87 1.00 1.00 1.00 1.00

0.02*

<0.01* 0.12

1.00 1.00

1.00

<0.01* 0.06 <0.01*

1.00 1.00

P2

1.00 1.00 1.00

1.00 0.78

P1

<0.01* <0.01* <0.01* <0.01* 0.15 0.03* 0.02*

<0.01* 0.09*

0.02*

0.04* 0.09

0.02* 0.09 0.15

1.00 0.19

P3

Bi^ indicates parameters obtained based on iodine-enhanced imaging

P3 indicates P values for post hoc analyses of AIS versus invasive adenocarcinoma

P2 indicates P values for post hoc analyses of MIA versus invasive adenocarcinoma

P1 indicates P values for post hoc analyses of AIS versus MIA

In terms of size variables and histogram analysis variables, P values are Bonferroni-corrected P values (Bonferroni-correction, P<.05 ÷ 2 for size variables and P<.05 ÷ 7 for histogram variables)

* P values were calculated with one-way ANOVAs

*P<0.05

**Data indicate the number of individuals

Unless otherwise indicated, data are means±standard deviation

Note: Classified According to the International Multidisciplinary Lung Adenocarcinoma Classification system

57.9±7.5 6:3

Minimally Invasive Adenocarcinoma(MIA) (n=9)

56.8±3.3 4:0

Adenocarcinoma In Situ (AIS) (n=4)

Characteristics of lung adenocarcinoma with little solid component on CT in reference to invasion status (n=39)

Age (y) Male-to-female ratio** CT parameters Size (mm) Volume (cm3) Density i-Density Mass (g) i-Mass (g) Histogram analysis Skewness i-Skewness Kurtosis i-Kurtosis 75th percentile (HU) i-75th percentile (HU) 97.5th percentile (HU) i-97.5th percentile (HU) Texture analysis Uniformity i-Uniformity Entropy i-Entropy Intensity variability i-Intensity variability Size-zone variability i-Size-zone variability

Variable

Table 2

Eur Radiol

Eur Radiol

multiple logistic regression was performed using significant parameters derived from only VNC images. Then, we performed a second multiple logistic regression with significant parameters from the first multiple logistic regression model and significant iodine-enhanced imaging parameters. Also, for multivariate analyses, logistic regression analysis was used with multi-colinearity examinations by using the variance inflation factor. Spearman correlation analyses were performed to evaluate correlations between the extent of the invasive components and all imaging variables. Finally, ROC analysis was also performed to evaluate the differentiating performance of multiple logistic regression models in discriminating invasive adenocarcinoma from AIS or MIA. Statistical significance was evaluated with software (SPSS, Version 19.0, 2010; SPSS, Chicago, Ill). A P value less than 0.05 indicated a statistically significant difference.

Results Patient population and surgical outcomes Details on clinicopathologic characteristics are shown in Table 1. Of all 39 tumours, four were AIS (10 %), nine were MIA (23 %), and 26 were invasive adenocarcinoma (67 %). Subtype classification of the 26 invasive adenocarcinomas showed three lepidic-predominant (11 %), 21 acinar-predominant (81 %), one papillary-predominant (4 %), and one micropapillary predominant (4 %) adenocarcinoma. No tumours showed lymphatic, vascular, perineural, or pleural invasion. The median extent of invasion in the 26 invasive adenocarcinomas was 15 mm (range: 6–28 mm). Of all 39 tumours, 36 tumours (92 %) showed pure GGNs without solid components on CT images. The remaining three tumours, which showed less than 5 mm of solid components, were all invasive adenocarcinomas. The absolute amount of enhancement of normal lung tissue was 12.18 (range: 6.3–22.2), which is significantly lower than that of GGN, 65.24 (range: 46.7–111.1) suggesting normally perfused surrounding lung tissue is less likely to have a significant effect on the result. Fifteen tumours were removed via sublobar resection (wide wedge resection or segmentectomy). The remaining 24 tumours were removed via lobectomy. Of all 26 invasive adenocarcinomas, 20 were removed via lobectomy. The remaining six tumours were less than 2 cm in size or showed less than 5 mm of a solid component on preoperative CT imaging. These remaining six tumours were removed via sublobar resection. Of all AIS, three were removed via sublobar resection, and the remaining tumour, which was associated with another GGN in the same lobe, was removed via lobectomy.

Comparison between three different lung adenocarcinomas based on the amount of invasive component Table 2 presents comparisons of all CT parameters according to the three pathologic subtypes (Figs. 2 and 3). Compared to AIS or MIA, invasive adenocarcinoma showed significant differences in nodule size, mass, CT numbers at the 75th and 97.5th percentiles, uniformity, and entropy on VNC images. On iodine-enhanced imaging, compared to AIS or MIA, invasive adenocarcinoma also showed significant differences in uniformity, entropy, CT numbers at the 97.5th percentiles, intensity, and variability (all Ps<0.05). According to multivariate logistic regression analysis, when applying VNC image parameters alone, mass, uniformity, and size-zone variability were independent predictors of invasive adenocarcinoma (odds ratio [OR]=19.92, P=0.02; OR=0.70, P=0.01; OR=16.16, P=0.04, respectively). After adding the iodine-enhanced imaging parameters to the independent predictors identified via VNC imaging, mass on the VNC images and uniformity on iodine-enhanced images were independent predictors of invasive adenocarcinoma (OR= 5.51, P=0.04 and OR=0.67, P<0.01) (Table 3).

Table 3 Multivariate analysis for differentiating invasive adenocarcinoma from AIS or MIA Invasive adenocarcinoma vs. AIS or MIA Variable VNC image alone+ Size (mm)

Odds Ratio

0.90

Mass (g) 19.92 75th percentile (HU) 1.96 97.5th percentile (HU) 1.01 Uniformity 0.70 Entropy 1.11 Size-zone variability 16.16 Adding iodine-enhanced image** Mass (g) 5.51 Uniformity 0.61 Size-zone variability 1.07 i-97.5th percentile (HU) 1.05 i-Uniformity 0.67 i-Entropy 1.18 i-Intensity variability 1.00

95 % CI

P

0.63 – 1.30

0.59

1.79 – 222.57 0.87 – 3.07 0.98 – 1.05 0.56 – 0.87 0.00 – 2.52 1.01 – 257.63

0.02* 0.45 0.44 0.01* 0.79 0.04*

1.49 – 24.87 0.38 – 1.57 0.94 – 1.23 0.96 – 1.15 0.52 – 0.85 0.00 – 2.92 0.99 – 1.00

0.04* 0.36 0.58 0.26 < 0.01* 0.91 0.59

Note: CI confidence interval. * P<0.05 ** Multivariate analysis after adding iodine-enhanced imaging parameters to the significant VNC imaging parameters +

Multivariate analysis with VNC imaging parameters

Eur Radiol

Correlation between imaging parameters and pathology The relationships between all parameters and the extent of invasion on pathology are shown in Table 4. Of all VNC image parameters, tumour size, volume, density, mass, CT attenuation values at the 75th and 97.5th percentiles on the histogram, entropy, intensity variability, and size-zone variability correlated positively with the extent of invasion. On iodine-enhanced images, mass, entropy, intensity variability, and size-zone variability also correlated positively with the extent of invasion. However, on VNC and iodine-enhanced images, uniformity was negatively correlated with the extent of invasion. Predictive probability of quantitative CT parameters for classifying pathology Based on multivariate analysis, we investigated whether we could accurately identify invasive adenocarcinoma from AIS or MIA by combining significant predictive factors (Table 3). ROC analysis showed that the area under the curve (AUC) was 0.888 when VNC imaging parameters Table 4 Correlation of imaging biomarker features to the extent of invasion on pathology Imaging variable

Extent of invasion on pathology

Size (mm) Volume (cm3) Density i-Density Mass (g) i-Mass (g) Skewness i-Skewness Kurtosis i-Kurtosis 75th percentile (HU) i-75th percentile (HU) 97.5th percentile (HU) i-97.5th percentile (HU) Uniformity i-Uniformity Entropy

0.71** 0.53** 0.43** -0.07 0.51** 0.54** -0.07 0.01 -0.05 -0.10 0.45** 0.08 0.61** 0.45** -0.60** -0.65** 0.69**

i-Entropy Intensity variability i-Intensity variability Size-zone variability i-Size-zone variability

0.71** 0.51** 0.61** 0.49** 0.45**

Note: Data are R values (correlation coefficients) * P<.05. ** P<.01.

were applied alone (combination of mass, uniformity, and size-zone variability). For the combination of independent predictors from both VNC images and iodine-enhanced images (mass on VNC images and uniformity on iodineenhanced images), the AUC value of the ROC was 0.959 (Fig. 4). The power of diagnosing invasive adenocarcinoma was improved after adding the iodine-enhanced imaging parameters compared to VNC imaging alone, from 0.888 to 0.959 (P=0.029).

Discussion Analysis using conventional non-contrast CT imaging metrics can help quantify tumour heterogeneity by assessing the greyscale distribution and by predicting tumour cellularity [17, 22]. In our previous study using conventional noncontrast CT imaging metrics, the 75th percentile CT attenuation value and entropy were identified as predictors for invasive adenocarcinoma from AIS or MIA lesions showing pure GGN or part solid nodules with little solid components [2]. However, no previous studies assessing GGN have ever evaluated the amount of the nodule enhancement. In this study, we found that quantitative analysis using iodine-enhanced imaging metrics generated from DECT have added value in distinguishing invasive adenocarcinoma from AIS or MIA showing pure GGN or part solid nodules with

Fig. 4 Receiver operating characteristic (ROC) curve for predicting invasive adenocarcinoma from non-invasive or minimally invasive adenocarcinoma. The AUC, based on the logistic model of VNC imaging parameters (mass, uniformity, and size-zone variability) alone, was 0.888. After adding the iodine imaging parameters (mass in the VNC image and uniformity in the iodine-enhanced image), the AUC was 0.959, showing the most significant diagnostic accuracy

Eur Radiol

little solid components compared to non-enhanced imaging metrics. Often on CT, strong enhancement of a malignant nodule is associated with increased perfusion and permeability of the capillaries, which results from increased microvessel density [23]. The process of tumour neo-angiogenesis plays a vital role in tumour progression and metastasis [24]. Many previous studies have revealed dynamic enhancement of pulmonary nodules on CT to help differentiate malignant nodules from benign nodules [23, 25, 26]. For example, Yi et al. [23] reported that malignant nodules show significantly higher peak and net enhancements, and the extent of the enhancement reflects the underlying microvessel density and tumour angiogenesis. Still, evaluation of enhancement of GGNs with too much air components and too little solid components is not possible with conventional CT. However, using DECT, we can measure the net iodine value of the pure GGNs, which reflects the level of underlying tumour angiogenesis. To the best of our knowledge, this is the first paper using DECT to examine whether quantitative analysis of iodine maps of GGNs can distinguish invasive adenocarcinoma from AIS or MIA. In our study, we could quantify the enhancement extent of the GGNs with the help of the DECT technique, which allowed the pure iodine component of GGN to be differentiated from an air component of the lesion. Such results may imply that the DECT technique allows for quantification of the extent of the nodule enhancement even in pure GGNs and this facilitates stratifying the GGN internal vascular changes over time and a more accurate evaluation of the aggressiveness of the tumour. As seen in Figs. 2 and 3, the shapes of the histograms are different and show no linear correlation between VNC images and iodine-enhanced images, even in the same tumour. This suggests that vascularity of the tumour may or may not conform to the tumour density in the same area of the tumour. The extent of neovascularization can be quantified very sensitively by iodine-enhanced images generated from DECT. Furthermore, we evaluated both the mean value of enhancement and the heterogeneity of vascularity via texture and histogram analysis. Finally, we observed significantly reduced uniformity of the enhancement in a step-by-step fashion from AIS to MIA to invasive adenocarcinoma, and the uniformity of enhancement was found to be an independent predictor for invasive adenocarcinoma after multivariate analysis. When assessing tumour vasculature, an important component of heterogeneity within tumours is irregularity of the distribution of tumour blood vessels because a heterogeneous vascular supply will result in areas of hypoxia [27]. On CT images, the difference in density is more prominent between tumour blood vessels and necrotic tissue after contrast administration, which results in increased intratumoral heterogeneity. Although adenocarcinoma is heterogeneous on microscopy, adenocarcinoma showing GGN seems to be homogeneous on CT. Thus, we hypothesized that microscopic heterogeneity

can be represented as heterogeneity of density or enhancement on CT. So, we used these parameters to measure tumoral heterogeneity. A limitation of this study is the relatively small patient population. Future studies with a substantially larger number of patients are necessary to further validate the findings of our present study. Another potential limitation is that the pathologic invasive component was evaluated in a subjective manner. Thunnissen et al. [28] assessed the reproducibility of lung adenocarcinoma invasion using an international group of pulmonary pathologists, and concluded that there is fair reproducibility when distinguishing invasive from in situ (wholly lepidic) patterns. Nevertheless, we attempted to reduce interobserver and intra-observer variability by using virtual microscopy [19]. Recent studies showed that virtual microscopy is a reliable and more reproducible technology as compared with conventional microscopy [19, 29–31]. Also in our study, digital pathology offered a rigorous and reproducible method for quantifying invasive and noninvasive components of histopathology. The manual outlining of ROIs may have subjective differences, depending on who draws them. However, automatic volumetry has not been accurate enough to be used for GGNs. Thus, we manually excluded all blood vessels, bronchial walls, and air-bronchograms when drawing the boundaries of the lesions. In conclusion, the addition of iodine-enhanced imaging metrics to conventional non-contrast imaging improves the power of diagnosing invasive adenocarcinoma appearing as pure GGN or part solid nodules with little solid component when compared to conventional non-contrast imaging alone. Acknowledgments The scientific guarantor of this publication is Ho Yun Lee. The authors of this manuscript declare no relationships with any companies, whose products or services may be related to the subject matter of the article. This study has received funding through grants from the Korean Foundation for Cancer Research (KFCR-CB-2011-02-02). Dr. Seonwoo Kim at the Biostatistics Unit of Samsung Biomedical Research Institute kindly provided statistical advice for this manuscript. Institutional Review Board approval was obtained (IRB 2011 09-083). Written informed consent was obtained from all patients in this study. Methodology: prospective, diagnostic study, performed at one institution.

References 1.

2.

3.

Lim HJ, Ahn S, Lee KS et al (2013) Persistent pure ground-glass opacity lung nodules >/= 10 mm in diameter at CT scan: histopathologic comparisons and prognostic implications. Chest 144:1291– 1299 Son JY, Lee HY, Lee KS et al (2014) Quantitative CT analysis of pulmonary ground-glass opacity nodules for the distinction of invasive adenocarcinoma from pre-invasive or minimally invasive adenocarcinoma. PLoS One 9, e104066 Swensen SJ, Brown LR, Colby TV, Weaver AL (1995) Pulmonary nodules: CT evaluation of enhancement with iodinated contrast material. Radiology 194:393–398

Eur Radiol 4.

5. 6.

7.

8.

9.

10.

11.

12.

13. 14.

15.

16.

17.

Swensen SJ, Brown LR, Colby TV, Weaver AL, Midthun DE (1996) Lung nodule enhancement at CT: prospective findings. Radiology 201:447–455 Zhang M, Kono M (1997) Solitary pulmonary nodules: evaluation of blood flow patterns with dynamic CT. Radiology 205:471–478 Chae EJ, Song JW, Krauss B et al (2010) Dual-energy computed tomography characterization of solitary pulmonary nodules. J Thorac Imaging 25:301–310 Brooks RA (1977) A quantitative theory of the Hounsfield unit and its application to dual energy scanning. J Comput Assist Tomogr 1: 487–493 Travis WD, Brambilla E, Noguchi M et al (2011) International association for the study of lung cancer/american thoracic society/ european respiratory society: international multidisciplinary classification of lung adenocarcinoma: executive summary. Proc Am Thorac Soc 8:381–385 Naidich DP, Bankier AA, MacMahon H et al (2013) Recommendations for the management of subsolid pulmonary nodules detected at CT: a statement from the Fleischner society. Radiology 266:304–317 Chae EJ, Song JW, Seo JB, Krauss B, Jang YM, Song KS (2008) Clinical utility of dual-energy CT in the evaluation of solitary pulmonary nodules: initial experience. Radiology 249:671–681 Gupta R, Phan CM, Leidecker C et al (2010) Evaluation of dualenergy CT for differentiating intracerebral hemorrhage from iodinated contrast material staining. Radiology 257:205–211 de Hoop B, Gietema H, van de Vorst S, Murphy K, van Klaveren RJ, Prokop M (2010) Pulmonary ground-glass nodules: increase in mass as an early indicator of growth. Radiology 255:199–206 Mull RT (1984) Mass estimates by computed tomography: physical density from CT numbers. AJR Am J Roentgenol 143:1101–1104 Goh V, Ganeshan B, Nathan P, Juttla JK, Vinayan A, Miles KA (2011) Assessment of response to tyrosine kinase inhibitors in metastatic renal cell cancer: CT texture as a predictive biomarker. Radiology 261:165–171 Tixier F, Le Rest CC, Hatt M et al (2011) Intratumor heterogeneity characterized by textural features on baseline 18F-FDG PET images predicts response to concomitant radiochemotherapy in esophageal cancer. J Nucl Med 52:369–378 Chong Y, Kim JH, Lee HY et al (2014) Quantitative CT variables enabling response prediction in neoadjuvant therapy with EGFRTKIs: are they different from those in neoadjuvant concurrent chemoradiotherapy? PLoS One 9, e88598 Ikeda K, Awai K, Mori T, Kawanaka K, Yamashita Y, Nomori H (2007) Differential diagnosis of ground-glass opacity nodules: CT number analysis by three-dimensional computerized quantification. Chest 132:984–990

18.

19.

20.

21.

22.

23.

24.

25.

26.

27.

28.

29.

30.

31.

Chandarana H, Rosenkrantz AB, Mussi TC et al (2012) Histogram analysis of whole-lesion enhancement in differentiating clear cell from papillary subtype of renal cell cancer. Radiology 265:790–798 Ozluk Y, Blanco PL, Mengel M, Solez K, Halloran PF, Sis B (2012) Superiority of virtual microscopy versus light microscopy in transplantation pathology. Clin Transpl 26:336–344 Travis WD, Brambilla E, Noguchi M et al (2011) International association for the study of lung cancer/american thoracic society/ european respiratory society international multidisciplinary classification of lung adenocarcinoma. J Thorac Oncol 6:244–285 Kim HK, Choi YS, Kim J, Shim YM, Lee KS, Kim K (2010) Management of multiple pure ground-glass opacity lesions in patients with bronchioloalveolar carcinoma. J Thorac Oncol 5:206– 210 Ganeshan B, Abaleke S, Young RC, Chatwin CR, Miles KA (2010) Texture analysis of non-small cell lung cancer on unenhanced computed tomography: initial evidence for a relationship with tumour glucose metabolism and stage. Cancer Imaging 10:137–143 Yi CA, Lee KS, Kim EA et al (2004) Solitary pulmonary nodules: dynamic enhanced multi-detector row CT study and comparison with vascular endothelial growth factor and microvessel density. Radiology 233:191–199 Weidner N, Semple JP, Welch WR, Folkman J (1991) Tumor angiogenesis and metastasis–correlation in invasive breast carcinoma. N Engl J Med 324:1–8 Swensen SJ, Morin RL, Schueler BA et al (1992) Solitary pulmonary nodule: CT evaluation of enhancement with iodinated contrast material–a preliminary report. Radiology 182:343–347 Yamashita K, Matsunobe S, Takahashi R et al (1995) Small peripheral lung carcinoma evaluated with incremental dynamic CT: radiologic-pathologic correlation. Radiology 196:401–408 Nelson DA, Tan TT, Rabson AB, Anderson D, Degenhardt K, White E (2004) Hypoxia and defective apoptosis drive genomic instability and tumorigenesis. Genes Dev 18:2095–2107 Thunnissen E, Beasley MB, Borczuk AC et al (2012) Reproducibility of histopathological subtypes and invasion in pulmonary adenocarcinoma. An international interobserver study. Mod Pathol 25:1574–1583 Kayser G, Kayser K (2013) Quantitative pathology in virtual microscopy: history, applications, perspectives. Acta Histochem 115: 527–532 Neltner JH, Abner EL, Schmitt FA et al (2012) Digital pathology and image analysis for robust high-throughput quantitative assessment of Alzheimer disease neuropathologic changes. J Neuropathol Exp Neurol 71:1075–1085 Onozato ML, Klepeis VE, Yagi Y, Mino-Kenudson M (2012) A role of three-dimensional (3D) reconstruction in the classification of lung adenocarcinoma. Stud Health Technol Inform 179:250–256