Radiomic Analysis of Pulmonary Nodules for Distinguishing Malignancy From Benignancy: The Value of Using Iodine Maps From Dual-Energy Computed Tomography : Journal of Computer Assisted Tomography

Secondary Logo

Journal Logo

Chest Imaging

Radiomic Analysis of Pulmonary Nodules for Distinguishing Malignancy From Benignancy: The Value of Using Iodine Maps From Dual-Energy Computed Tomography

Zhong, Yan MM; Xu, Hai MD; Zhang, Wei MD; Li, Hai MD; Yu, Tong-Fu MD; Yuan, Mei MD

Author Information
Journal of Computer Assisted Tomography 46(6):p 878-883, 11/12 2022. | DOI: 10.1097/RCT.0000000000001360
  • Free


Lung cancer remains the leading cause of cancer-related death around the word.1 Early and accurate diagnosis of lung cancer can improve the survival benefit.2,3 Therefore, the accuracy and reproducibility of diagnosis for discriminating between benign and malignant nodules are crucial. However, there is a lack of a noninvasive method for the diagnosis of lung cancer with relatively high accuracy.

The availability of conventional computed tomography (CT) has an important impact on capturing morphological and enhancement features of pulmonary nodules in a noninvasive way. To some degree, the nonspecific and similar CT image appearance makes it difficult to differentiate between benign and malignant nodules.4–6

A recently introduced dual-energy CT (DECT) has provided an effective method for the quantification of iodine enhancement and distribution of iodine. A few studies have reported that iodine-related attenuation on DECT is a reliable quantitative parameter for the assessment of underlying tumor perfusion and angiogenesis, which is associated with the increased microvessel density in malignancy.7–9 Therefore, DECT iodine intake measurement had been used in multiple systems, including the lung, gastrointestinal tract, liver, kidney, and so on.10–12 Lin et al13 evaluated the parameters from DECT gemstone spectral imaging and demonstrated that the parameters (mean slope rate, iodine concentration, and normalized iodine concentration) of malignant solid pulmonary nodules were significantly lower than those of inflammatory solid pulmonary nodules, on the contrary, significantly higher than those of tuberculoma.

However, until now, few studies have evaluated the uptake parameters from DECT iodine overlap map for differentiating benign nodules independently. Moreover, the potential high-dimensional quantitative information of iodine distribution from iodine overlap map has not been fully investigated. Radiomics is an emerging and promising method for extraction high-throughput quantitative imaging features of lesions.14 Wang et al15 showed that radiomics had potential for predicting benign or malignant status of pulmonary nodules by quantification of tumor heterogeneity. However, tumor angiogenesis cannot be assessed properly if radiomics were extracted from noncontrast-enhanced CT images. According to the study of Bae et al,16 quantitative radiomics value derived from contrast-enhanced DECT showed potential ability for prediction of pathological invasiveness and prognosis in lung adenocarcinoma. Iodine map derived from mixed DECT images provides distribution information of iodinated contrast media, and radiomics enables the quantitative evaluation of internal heterogeneity on iodine map. As a result, it may be able to detect the heterogeneous iodinated distribution and more accurate to detect potential malignant features of pulmonary nodules.

Therefore, we hypothesized that radiomics based on iodine map from DECT may provide heterogeneous iodinated information for discriminating malignant pulmonary nodules. The purpose of our study was to evaluate the predictive role of iodine map–based radiomics in discriminating malignant pulmonary nodules from benign ones.



Our institutional review board approved this study, and informed consent was waived.

During October 2017 and May 2018, 432 patients underwent thoracic DECT scans in our institution. Patients met the following criteria were retrospectively included in this study: (a) patients with solitary noncalcified nodules (a pulmonary nodule is defined as any pulmonary lesion characterized by a well-defined, discrete, approximately circular opacity 3 cm or less in diameter); (b) patients received anti-inflammatory therapies for at least 2 weeks before DECT scan, and lesions showed no or slight changes (less than 20% changes in sum according to revised RECIST guideline version 1.1) for at least 3 months of follow-up, which were difficult for clinical diagnosis; and (c) pathologically confirmed benign or malignant diagnosis by CT-guided transcutaneous needle biopsy or surgery or transbronchial biopsy within 6 weeks after DECT. Patients were excluded if they met the following criteria: (a) no simultaneous nonenhanced CT scan (n = 153); (b) distant metastasis or lymph node metastasis (n = 64); (c) radiation and chemotherapy history before scanning (n = 59); (d) unsatisfactory imaging quality due to respiratory artifact during examination (n = 33); and (e) nodules with extensive cavities (n = 15). Finally, a total of 117 nodules in 109 patients were enrolled in our institution as the model-development cohort.

For external validation, patients who undertaken thoracic DECT scans in another institution were enrolled with the same inclusion and exclusion criteria as the testing data set.

Dual-Energy Computed Tomography

All patients underwent CT examination using a dual-source CT scanner (Somaton Force; Siemens Healthcare, Forchheim, Germany). Nonenhanced CT scans were performed by using singe-energy CT (collimation, 192 × 0.6 mm; tube voltage 110 kV; gantry rotation time, 0.5 seconds; reconstruction slice thickness, 1.5 mm). Dual-energy CT scanning was obtained 40 and 100 seconds after the administration of nonionic iodinated contrast material (Ultravist 370 mg/mL; Bayer, Germany) at a rate of 4.0 mL/s using a power injector (Medrad; Stellant, Bayer, Germany), followed by a 30 mL of saline chaser at the same injection rate. The scan parameters were as followed: tube voltage of 90 kVp and 150 Sn kVp, tube current adjusted to voltage, 192 × 0.6 mm collimation and gantry rotation time of 0.5 seconds. Three types of data sets were generated from DECT: 90 kVp, 150 Sn kVp, and weighted-average image (the ratio from tube A and tube B was 0.6:0.4). Virtual nonenhanced images and iodine distribution images were created from modified virtual noncontrast mode of dedicated dual-energy postprocessing software (Syngo.via, version VA30A; Siemens Healthcare). Image data were reconstructed using a medium-soft convolution kernel (Q40) with a 1.0-mm slice thickness. We regarded iodine distribution images as iodine maps.

Image Postprocessing

Two thoracic radiologists with 8 and 6 years of experience, for each patient, separately segmented the entire-tumor volume of interest. Using the lungCAD software (Syngo.via, version VA30A), each radiologist separately drew a line along the longest diameter of the tumor on enhanced weighted-average images (arterial phase, 40 seconds after the administration of nonionic iodinated contrast material). The lungCAD software, using this drawn line, segmented the entire volume of the tumor automatically. The radiologists then separately manually adjusted the nodule segmentation in each image. The segmented volume of interest (VOI) was propagated to nonenhanced CT images and iodine maps to cover the same VOI. Eight iodine uptake parameters by DECT were measured automatically by an in-house software (Syngo.via, version VA30A), including mean value of virtual noncontrast (NCM), enhancement median on contrast-enhanced image (CM), total iodine concentration, total iodine uptake, vital iodine concentration, vital iodine uptake, mean value of vital part on contrast enhanced image (VM), and mean value of iodine image of the whole targeted nodules. Here, “vital” means the remaining part of the tumor after removing necrotic area of the tumor.

Then, the iodine maps of VOI were loaded by our in-house software (Analysis Kit, version 3.0.0; GE Healthcare). A total of 87 radiomics features were extracted automatically with the following 3 textual categories: (a) histogram; (b) gray level co-occurrence matrix and gray level run length matrix; and (c) wavelet features. Radiomics feature selections on the basis of reproducibility and redundancy were performed to prioritize these high-dimensional features. First, concordance correlation coefficient was used to test the reproducibility and stability of each feature. Features with concordance correlation coefficient value greater than 0.90 were initially selected.

As sample size was relatively small, least absolute shrinkage and selection operator (LASSO) logistic regression classifiers were trained (cohort 1) and validated (cohort 2) using leaving-one-out 10-fold (10 repeat iterations) cross-validation approach, in which, except one fold for validation (cohort 2), the other 6-folds were applied for training (cohort 1). This procedure was repeated until each case in the database was used once in the validating set.

Statistical Analysis

The reproducibility of parameters derived from the volume measurements independently created by each of the 2 radiologists was assessed by the intraclass correlation coefficient. Intraclass correlation coefficient values of 0.40 to 0.60, 0.61 to 0.80, and 0.81 to 1.00 indicate, good, moderate, and excellent reproducibility, respectively.

Continuous variables were expressed as mean ± standard deviation. The ability to discriminate between benign and malignant nodules was analyzed by Mann-Whitney U test for nonnormally distributed variables, and independent samples t test for normally distributed variables. P value less than 0.05 was considered statistically significant.

The significant features of iodine uptake parameters and radiomics in the training cohort were selected by the LASSO logistic regression to constitute iodine uptake model and radiomics model. The LASSO logistic regression model was used with penalty parameter tuning that was conducted by leaving-one-out 10-fold cross-validation based on minimum criteria, which is suitable for regression analysis of a high-dimensional data set.17 The performance of each model was evaluated using the validation cohort. The predictive performances of iodine uptake parameters and radiomics features were evaluated by receiver operating characteristic (ROC) curves and the area under the curve (AUC) in both the training and external validation sets. Diagnostic accuracy, sensitivity, and specificity of significant features were calculated and compared at a cutoff point that maximized the value of the Youden index.

Statistical analyses were performed with software (SPSS, Version 19.0, 2010; SPSS, Chicago, III).


The interobserver ICC values for the volume measurements independently created by each of the 2 radiologists was excellent with values greater 0.80.

Patient Characteristics

A total of 109 patients (52 men; mean age, 61.9 ± 9.2; 57 women, mean age, 56.1 ± 12.8) with 117 nodules were included in our institution as the model-development cohort. There were 62 benign and 55 malignant nodules according to pathologic diagnoses. The benign nodules included granulomatous inflammation (n = 28), organizing pneumonia (n = 28), hamartoma (n = 3), sclerosing hemangioma (n = 2), and lymph node (n = 1). The malignant diagnoses included adenocarcinoma (n = 46), squamous carcinoma (n = 6), and small cell carcinoma (n = 3). Table 1 summarizes the comparisons of age, sex, smoking history, maximum lesion diameter, and lobar location between groups A and B. No significant difference was found on any characteristic between group A (benign) and B (malignant, all P > 0.05).

TABLE 1 - Patient and Lesion Characteristics
Characteristics Group A (Benign) Group B (Malignant) P
No. patients 60 49
No. nodules 62 55
Age, y 59.02 ± 12.54 61.98 ± 9.86 0.268
Sex, men/women 29/31 23/26 0.897
Smoking history, yes/no 26/34 20/29 0.715
Maximum lesion diameter, cm 1.81 ± 0.47 1.86 ± 0.45 0.507
Lobar location, upper/middle/lower 32/9/21 30/10/15 0.748

In testing data set, 47 patients with surgically resected solitary nodules (25 benign and 22 malignant) from another hospital were included for external validation.

Enhancement Pattern

Malignant nodules showed significantly higher whole-tumor enhancement and mean iodine value than inflammatory nodules (P < 0.001). The accuracy of enhancement value based on whole-tumor iodine uptake map (mean iodine) was 72.35% with sensitivity and specificity of 69.4% and 83.1%, respectively.

Iodine Uptake Model

All 8 parameters from DECT were included into modeling with ICC values greater than 0.80. As showing in Table 2, all 8 features were significant higher in malignant nodules compared with benign nodules (P < 0.01). There are 2 best performing features (NCM and VM) selected by LASSO logistic regression, and both of them showed high accuracy of 73.9% and 74.2%, respectively. The mean value of vital part showed higher AUC and sensitivity than NCM (AUC, 0.728 vs 0.708; sensitivity, 66.9% vs 61.2%; Table 3). Iodine uptake model composed of these 2 features showed great potential to discriminate malignant nodules from benign nodules with an AUC of 0.804 in the training data set. In the validation data set, the accuracy for the prediction model was 80.6% with AUC value of 0.756 (Fig. 1).

TABLE 2 - Comparison of Quantitative Iodine Uptake Parameters
Variable Group A (Benign) Group B (Malignant) P
NCM, HU 12.97 ± 19.96 21.17 ± 19.98 <0.001
CM, HU 16.75 ± 15.60 31.12 ± 15.72 <0.001
TIC, mg/mL 0.70 ± 0.67 1.30 ± 0.66 <0.001
TIU, mg 11.05 ± 41.42 29.98 ± 46.15 <0.001
VIC, mg/mL 0.71 ± 0.67 1.36 ± 0.66 <0.001
VIU, mg 11.18 ± 42.05 30.58 ± 46.66 <0.001
VM, HU 17.02 ± 15.77 32.57 ± 15.65 <0.001
Mean iodine, HU 17.56 ± 10.85 39.99 ± 19.14 <0.001
Vital indicates the remaining part of the tumor after removing necrotic area of the tumor.
CM indicates mean value of contrast image; TIC, total iodine concentration; TIU, total iodine uptake; VIC, vital iodine concentration; VIU, vital iodine uptake.

TABLE 3 - Effectiveness of Iodine Signature and Radiomics Signature in Prediction of Malignant Nodules
P Cutoff Value AUC SEN SPE ACC
Mean iodine, HU <0.001 >27.32 0.685 0.694 0.831 0.724
Iodine signature <0.001 >0.506 0.756 0.750 0.850 0.806
 NCM <0.001 >18.11 0.708 0.612 0.775 0.739
 VM <0.001 >27.16 0.728 0.669 0.744 0.742
Radiomic signature <0.001 >0.478 0.800 0.846 0.938 0.897
 shape_Sphericity <0.001 <0.558 0.819 0.810 0.750 0.766
 glszm_SmallAreaHighGrayLevelEmphasis <0.001 >2.131 0.783 0.690 0.808 0.755
Vital indicates the remaining part of the tumor after removing necrotic area of the tumor.
ACC indicates accuracy; SEN, sensitivity; SPE, specificity.

A, The ROC curve effectiveness of the iodine uptake model in the training cohort to detect the malignant nodules. B, The ROC curves effectiveness of the radiomics signature in the training cohort. C, The ROC curve effectiveness of the iodine uptake model in the testing cohort. D, The ROC curves effectiveness of the radiomics signature in the testing cohort.


Among the 87 radiomics features, 62 features were initially included with both interobserver and intraobserver ICC values greater than 0.90. Two most valuable features with nonzero coefficients in the LASSO logistic regression model were subsequently selected (Fig. 2) and used to build the radiomics model.

Radiomics feature Selection. A, Selection of tuning parameter (λ) in the LASSO model used 10-fold cross-validation via minimum criteria. Likelihood deviance was plotted versus log(λ). The dotted vertical lines were drawn at the optimal values by using the minimum criteria and the 1 standard error of the minimum criteria. Two features (original shape sphericity and original glszm small area high gray level emphasis) were selected with the smallest binomial deviance. C, Least absolute shrinkage and selection operator coefficient profiles of radiomics feature. Vertical line was potted at the value selected using 10-fold cross-validation and 5 features with nonzero coefficients were shown. Figure 2 can be viewed online in color at

The model ultimately included 2 features: original shape sphericity and original glszm small area high gray level emphasis. The radiomics model indicated favorable predictive efficacy with an AUC of 0.957 in the training cohort and 0.800 in the validation cohort (Fig. 3). In the validation data set, the accuracy for the prediction model was 89.7% with sensitivity and specificity of 84.6% and 93.8%, respectively.


The 4-best performing features (NCM, VM, original shape sphericity, and original glszm small area high-gray level emphasis) selected from iodine uptake and radiomics features showed potential association with discriminating of malignancy (P < 0.05; Table 3). Regression models of iodine uptake and radiomics features were generated by multivariate logistic regression analysis, and the Akaike information criterion was used as a measure of goodness of fit. The accuracy of the iodine uptake model and radiomics model were 80.6% and 89.7%, respectively. In consideration of accuracy, radiomics model was superior to the iodine uptake features in discriminating malignant nodules from benign nodules (Table 3).


This is the first study using radiomics features extracted from DECT iodine overlay map to differentiate malignant and benign nodules. Our study showed that radiomics model from DECT iodine overlay map was an independent predictor of malignance and showed better discriminative ability than the iodine uptake model to identify high-risk patients with malignant nodules.

Previous studies8,18 have suggested that an enhancement value of 20 HU was a critical value for differential diagnosis of benign and malignant nodules. However, the measurement of enhancement value can be easily affected by measurement region of interest of whole volume. Meanwhile, matrix and contrast agent also have impacts on CT measurements. Thus, in previous studies, the specificities of contrast-enhanced CT in discriminating benign and malignant nodules varied from 52% to 93%. The present study assessed a series of enhancement value and iodine uptake parameters based on whole tumor and demonstrated that iodine concentration, uptake, and enhancement were significantly higher in malignant nodules than that in benign ones. The accuracy of enhancement value based on whole-tumor iodine uptake map (mean iodine) was 72.35% with sensitivity and specificity of 69.4% and 83.1%, respectively. Lung cancers typically presented as hypervascularity accompanied with increased permeability and leakiness.19 Thus, iodine concentration and iodine uptake, which reflect to blood supply and vascular permeability, were significantly higher in malignant nodules.

Among the series of enhancement and iodine uptake parameters, the results in our study indicated that mean value of virtual noncontrast image (NCM) and mean vital (VM) of nodules showed to be best performing features with accuracy of 73.9% and 74.2%, respectively in differentiating malignant nodules from benign ones after feature selection. The NCM and VM of malignant lesions were significantly higher than that of benign lesions. For mean value of NCM, malignant nodules were usually with higher density probably because of invasive growth and fibrosis formation. However, Xu et al20 have shown that the malignant solid nodules tending to have a lower mean density than the benign solid nodules. Our study was inconsistent with this result. This might be explained by that we have excluded nodules with extensive cavities, which were more often seen in malignant nodules. As for VM value, according to previous studies, heterogeneity, especially necrosis of tumor, was significantly associated with aggressive tumor behavior.21–23 As far as we know, angiogenesis established in malignant lesions usually could not support the rapid growth of neoplasm and necrosis within malignant lesions would occur resulted from subsequent reduction of oxygen delivery. Thus, the remaining area of lesions after eliminating necrosis represented the real part of angiogenesis relatively.

The iodine uptake model consisted of 2 parameters achieved good diagnostic performance with an accuracy of 0.806 and an AUC of 0.756. Although there are several studies4–6 focusing on morphology of nodules in common CT images to discriminate malignant nodules, the accuracy of these study was not satisfied, indicating that iodine quantification showed superior performance than qualitative appearance and may potentially be used in clinical practice regarding the differential diagnosis of pulmonary nodules.

The radiomics model in our study composed of selected 2 features (original shape sphericity and original glszm small area high gray level emphasis) yield good performance in differentiation. The greater than 90% specificity demonstrated that benign lesions would not be easily classified into false malignant group. If a benign lesion was diagnosed as a malignant one, the patients will be performed unnecessary treatment, which may cause excessive medical care. The sensitivity greater than 90% indicated that the radiomics model is quite sensitive for discriminating malignant lesions, which is valuable for clinical lung cancer screening. In all, it provided reasonable discrimination in the training cohort with an AUC of 0.957. The radiomics model was tested using the validation cohort, which showed favorable discrimination based on an AUC of 0.800. It indicated that radiomics model could predict malignant nodules preoperatively in a noninvasive and quantitative way. In addition, the prediction power of radiomic model is larger than iodine uptake features (ACC, 0.897 vs 0.806). Thus, the radiomic model derived from iodine map showed superior diagnostic value than dual-energy features.

Compared with the previous study Chia et al,24 in which the radiomic features were extracted from noncontrast enhancement CT images, the accuracy (84%) of the selected 4-feature radiomics model for predicting malignant pulmonary nodules was lower than that in our study (84% vs 90%). In addition, a previous study25 of constructing a radiomics signature derived from iodine overlay map in predicting survival outcomes in patients with lung cancer also showed that their prediction model achieved better performance than studies using radiomics extracted from common CT images. Thus, we presumed that radiomic features derived from iodine map capturing the status of blood supply within a tumor may provide more information than that from noncontrast enhancement images.

Our study had several limitations. First, this is a retrospective study performed in a single center with a relatively small sample. Second, our analysis was only available for arterial-phase images (40 seconds after the administration of nonionic iodinated contrast material), which impeded a quantitative assessment and comparison of venous-phase images (100 seconds). Further study is needed to analyze arterial-phase and venous-phase images comprehensively. Third, the performance of radiomics features extracted from iodine maps was not compared with that of routine CT images. A multiparametric radiomics signature using several CT images (noncontrast, contrast-enhanced, and iodine map images) is needed to be evaluated for predicting malignant nodules in the future. Lastly, radiologists' performance was not calculated in our study. The performance of the radiologists and other features is needed to be compared in further study.


Radiomics model derived from iodine map had added value to reflect heterogeneity of tumor perfusion and showed superior performance in discriminating between malignancy and benignancy than iodine uptake features. This technique may have potential and favorable application values to discriminate malignant and benign nodules and optimize clinical decision-making in the future.


1. Siegel RL, Miller KD, Jemal A. Cancer statistics. CA Cancer J Clin. 2017;67:7–30.
2. Parmar C, Leijenaar RTH, Grossmann P, et al. Radiomic feature clusters and prognostic signatures specific for lung and head & neck cancer. Sci Rep. 2015;5:11044.
3. MacMahon H, Austin JH, Gamsu G, et al. Guidelines for management of small pulmonary nodules detected on CT scans: a statement from the Fleischner society. Radiology. 2005;237:395–400.
4. Swensen SJ, Viggiano RW, Midthun DE, et al. Lung nodule enhancement at CT: multicenter study. Radiology. 2000;214:73–80.
5. Erasmus J. Solitary pulmonary nodules: part I. Morphologic evaluation for differentiation of benign and malignant lesions. Radiographics. 2000;20.
6. Ohno Y, Nishio M, Koyama H, et al. Dynamic contrast-enhanced CT and MRI for pulmonary nodule assessment. Am J Roentgenol. 2014;202:515–529.
7. Wang M, Li BG, Sun H, et al. Correlation study between dual source CT perfusion imaging and the microvascular composition of solitary pulmonary nodules. Lung Cancer. 2019;130:115–120.
8. Yi CA, Lee KS, Kim EA. Solitary pulmonary nodules: dynamic enhanced multi-detector row CT study and comparison with vascular endothelial growth factor and microvessel density. Radiology. 2004;233:191–199.
9. Viallard C, Larrivee B. Tumor angiogenesis and vascular normalization: alternative therapeutic targets. Angiogenesis. 2017;20:409–426.
10. Yan WQ, Xin YK, Jing Y, et al. Iodine quantification using dual-energy computed tomography for differentiating thymic tumors. J Comput Assist Tomogr. 2018;42.
11. Anno G, Becker CR, Michael S, et al. Single-phase dual-energy CT allows for characterization of renal masses as benign or malignant. Invest Radiol. 2010;1.
12. Benjamin K, Wichmann JL, Sophia P, et al. Iodine quantification to distinguish hepatic neuroendocrine tumor metastasis from hepatocellular carcinoma at dual-source dual-energy liver CT. Eur J Radiol. 2018;105:20–24.
13. Lin JZ, Zhang L, Zhang CY, et al. Application of gemstone spectral computed tomography imaging in the characterization of solitary pulmonary nodules: preliminary result. J Comput Assist Tomogr. 2016;40:907–911.
14. Zhong Y, Yuan M, Zhang T, et al. Radiomics approach to prediction of occult mediastinal lymph node metastasis of lung adenocarcinoma. Am J Roentgenol. 2018;1–5.
15. Wang J, Liu X, Dong D, et al. Prediction of malignant and benign of lung tumor using a quantitative radiomic method. Conf Proc IEEE Eng Med Biol Soc. 2016;1272–1275.
16. Bae JM, Jeong JY, Lee HY, et al. Pathologic stratification of operable lung adenocarcinoma using radiomics features extracted from dual energy CT images. Oncotarget. 2017;8:523–535.
17. Tibshirani RJ. Regression shrinkage and selection via the LASSO. J R Stat Soc. 1996;73:273–282.
18. Swensen SJ, Viggiano RW, Midthun DE, et al. Lung nodule enhancement at CT: multicenter Study1. Radiology. 2000;214:73–80.
19. Li GJ, Gao J, Wang GL, et al. Correlation between vascular endothelial growth factor and quantitative dual-energy spectral CT in non-small-cell lung cancer. Clin Radiol. 2016;71:363–368.
20. Xu DM, Klaveren RJV, Bock GHD, et al. Role of baseline nodule density and changes in density and nodule features in the discrimination between benign and malignant solid indeterminate pulmonary nodules. Eur J Radiol. 2009;70:492–498.
21. Swinson DE, Jones JL, Richardson D, et al. Tumour necrosis is an independent prognostic marker in non-small cell lung cancer: correlation with biological variables. Lung Cancer. 2002;37:235–240.
22. Park SY, Lee HS, Jang HJ, et al. Tumor necrosis as a prognostic factor for stage IA non-small cell lung cancer. Ann Thorac Surg. 2011;91:1668–1673.
23. Kilicgun A, Turna A, Sayar A, et al. Very important histopathological factors in patients with resected non-small cell lung cancer: necrosis and perineural invasion. Thorac Cardiovasc Surg. 2010;58:93–97.
24. Chia HC, Chih KC, Chih YT, et al. Radiomic features analysis in computed tomography images of lung nodule classification. PLoS One. 2018;13:e0192002.
25. Jooae C, Min LS, Kyung HD, et al. Prognostic value of radiomic analysis of iodine overlay maps from dual-energy computed tomography in patients with resectable lung cancer. Eur Radiol. 2018.

pulmonary nodule; radiomics; dual-energy computed tomography; iodine quantification

Copyright © 2022 Wolters Kluwer Health, Inc. All rights reserved.