Abstract
The objective of this feasibility study was to assess computed tomography (CT) texture analysis (CTTA) of pulmonary lesions as a predictor of overall survival in patients with suspected lung cancer on contrast-enhanced computed tomography (CECT). In a retrospective pilot study, 94 patients (52 men and 42 women; mean age, 67.2 ± 10.8 yrs) from 1 center with non-small cell lung cancer (NSCLC) underwent CTTA on the primary lesion by 2 individual readers. Both simple and multivariate Cox regression analyses correlating textural parameters with overall survival were performed. Statistically significant parameters were selected, and optimal cutoff values were determined. Kaplan–Meier plots based on these results were produced. Simple Cox regression analysis showed that normalized uniformity had a hazard ratio (HR) of 16.059 (3.861–66.788, P < .001), and skewness had an HR of 1.914 (1.330–2.754, P < .001). The optimal cutoff values for both parameters were 0.8602 and 0.1554, respectively. Normalized uniformity, clinical stage, and skewness were found to be prognostic factors for overall survival in multivariate analysis. Tumor heterogeneity, assessed by normalized uniformity and skewness on CECT may be a prognostic factor for overall survival.
Keywords: computed tomography, non-small cell lung cancer, texture analysis
1. Introduction
NSCLC is one of the most common types of cancer in the western part of the world and continues to be among the most lethal cancers. Over the last few years, the number of new cases and overall mortality have decreased in men, whereas both are increasing in women. In Denmark, 12% of all new cancer cases are diagnosed as lung cancer. The relative 5-year survival rate for the entire group is 11% in men and 14% in women; however, these are highly dependent on the time of diagnosis and stage of disease.[1]
Currently, the morphological features of conventional multi-detector computed tomography (MDCT) are the primary tools for staging lung cancer. 18-flourdeoxyglucose positron emission tomography/computed tomography (18-FDG-PET/CT) is often used in the diagnostic work-up of lung cancer. Evidence supports that the morphological features of multi-detector computed tomography (MDCT) and F18-FDG-PET/CT are equal in the assessment of tumor (T) and lymph node (N) stage.[2,3] However, F18-FDG-PET/CT is superior to CT in the assessment of metastatic spread (M). This is based on the 8th tumor-node-metastasis (TNM) classification, which currently does not include tumor heterogeneity, texture or biomarkers.[4]
CTTA, a noninvasive technique that quantifies the spatial pattern of pixel intensities on imaging, is an objective measure of tumor heterogeneity.[5] Tumor heterogeneity is a key feature of therapeutic resistance, which reflects intra-tumor variations in cell density, necrosis, and angiogenesis.[6] CTTA has been shown to provide prognostic information about survival and predictive information on recurrence and response to chemotherapy in patients with a variety of cancer forms.[7–11] If CT-derived prognostic biomarkers can be identified and incorporated into routine practice, they could potentially increase the utility of CT and provide useful and important information, which could help in making optimal treatment decisions and improve overall clinical outcomes.
For NSCLC specifically, texture analysis has been associated with metastatic spread, genetic mutation, differentiation between malignant and benign lymph nodes, and is a prognostic factor for overall survival.[7,12–16]
Although CTTA has been extensively used in research environments, it lacks adaptation to routine clinical use for diagnosis and prognostication.
The purpose of this feasibility study was to assess CTTA as a prognostic factor for overall survival, in patients with suspected lung cancer on CECT, based on the hypothesis that high heterogeneity in a tumor reflects poor overall survival.
2. Methods
2.1. Design and patients
This study was retrospective, following the STROBE guidelines, and included patients with NSCLC gathered between 2009 and 2011 at Aarhus University Hospital in a previous study. The Danish Board of Health granted permission to access the patient information retrospectively. All patients had previously provided written consent for the use of their images and data for the original study and subsequent analysis using new techniques.[3] The Danish Ethical Committee waived the need for new consent due to the retrospective nature of the study. Survival data were obtained for each patient using the Danish Lung Cancer Registry. Inclusion criteria: pathologically proven NSCLC, available survival data, and CECT suitable for texture analysis. The included patients were randomized for either a feasibility or validity dataset, with half of the cohort assigned to each.
2.2. CECT imaging protocol
CT was performed using a multiple-row detector CT scanner (Philips Brilliance CT 64-channel scanner; Philips Healthcare, Best, The Netherlands). The CT acquisition parameters were 64 × 0.625 mm collimation, kV 120 to 140, mAs/slice 150 to 250, rotation time 0.75, reconstruction thickness 2 mm, increment 1 mm, pitch 1.078, FOV 35 cm, and matrix 512 × 512. CT examinations included the chest and upper abdomen. Iodixanol 270 mg/mL (Visipaque® 270; GE Healthcare, Oslo, Norway), or iohexol 300 mg/mL (Omnipaque® 300; GE Healthcare) was injected intravenously at weight-adjusted doses of 2 mL/kg body weight to adjust for differences in distribution volume with an injection rate of 4 mL/s. A bolus-tracking technique with a region of interest (ROI) in the descending aorta at the level of the carina was used to adjust for differences in cardiac output. CT was performed after a delay of 15 seconds for the chest and upper abdomen (late arterial phase) and 65 seconds for the upper abdomen (portovenous phase) after a threshold of 200 HU was obtained.
2.3. CTTA of primary tumors
CTTA was performed using TexRAD, a research software (TexRAD Ltd, Cambridge, UK).[6] The technique comprised an image filtration-histogram approach; in which an initial filtration step using a Laplacian of Gaussian band-pass filter (non-orthogonal Wavelet) was employed to extract and enhance features of different size based on the spatial scale filter (SSF) values varying from 1.4 to 4.1 mm in diameter. Following image filtration, each filtered image texture map was quantified using histogram parameters such as entropy and uniformity (Fig. 1).
Figure 1.
Flowchart for included participants.
A specialist consultant in radiology with >5 years of experience in the use of the TexRAD software performed the analysis. The radiologists were blinded to all patient identifiers and clinical data, as well as to the CT and tissue sampling results. For each primary tumor, a ROI was drawn on all axial images, in 2D, and the final analysis was performed on the 3D volume of the drawn tumor. A semi-automated approach further refined the ROI enclosing the primary tumor to exclude air and fat using a thresholding procedure that removed pixels with values <–50 HU.
Thirty cases were randomly selected, and a second consultant in Radiology with >5 years of experience in the use of the TexRAD software reanalyzed these for interobserver agreement.
2.4. Statistical analysis
All analyses were conducted using the R software package 4.0.3 (R Core Team, 2022). In R the following packages were used: Dplyr (version 1.0.3, 2021), Survival (version 3.2-7. 2020), Survminer (version 0.4.9, 2021) and SurvMisc (version 0.5.5, 2018).[17–21] Texture parameters were normalized against coarse filters based on a previous publication suggesting that normalized features were more robust and had higher predictive value than standard texture features.[22] Simple Cox regression was used for the entire dataset to determine the parameters that were prognostic of survival. The feasibility dataset was used to determine the optimal thresholds for normalized and standard CTTA parameters, which were determined based on an algorithm in R using the Fisher exact method.[17] The differences between the survival curves for the groups over or under the threshold were evaluated using a nonparametric log-rank test (P < .05, considered significant). Interobserver agreement was assessed using Bland–Altman plots.
3. Results
3.1. Patients
In total, 94 patients were included in the study. Fifty-two males and 42 females with a mean age of 67.2 years (range, 56.4–78.0) were included. The cohort was randomized with 47 patients for the feasibility dataset and 47 patients for the validation dataset (Table 1, Fig. 2). Mean survival was 27.0 months and 27.6 months for the feasibility and validation datasets, respectively. Median survival was respectively 15.0 months and 21.8 months (Fig. 2).
Table 1.
Patient demographics including age, sex, tumor pathology, clinical stage, smoking history and treatment.
| Demographics | ||||
|---|---|---|---|---|
| Feasibility dataset | Validity dataset | Full dataset | ||
| Age | Yrs of age | 67.8 (CI 95% 64.3–71.2) | 66.5 (CI 95% 63.6–69.4) | 67.2 (CI 95% 64.9–69.4) |
| Sex | Male | 27 | 25 | 52 |
| Female | 20 | 22 | 42 | |
| Pathology | Adenocarcinoma | 29 | 22 | 51 |
| Planocellular carcinoma | 15 | 15 | 30 | |
| Large cell carcinoma | 0 | 6 | 6 | |
| Blended | 3 | 4 | 7 | |
| Stage | Ia | 8 | 9 | 17 |
| Ib | 10 | 9 | 19 | |
| IIa | 6 | 6 | 12 | |
| IIb | 1 | 1 | 2 | |
| IIIa | 6 | 6 | 12 | |
| IIIb | 4 | 2 | 6 | |
| IV | 11 | 10 | 21 | |
| Unknown | 1 | 4 | 5 | |
| Smoking history | Pack yrs | 29.9 (CI 95% 23.7–36.1) | 37.2 (CI 95% 31.5–42.9) | 33.8 (CI 95% 29.6–38.1) |
| Treatment | Surgery | 26 | 26 | 52 |
| Chemotherapy | 18 | 21 | 39 | |
| Radiation/Chemotherapy | 1 | 0 | 1 | |
| Missing data | 2 | 0 | 2 | |
Figure 2.
TexRAD research software showing the drawn ROI (Top left) and the representation of fine (Bottom left), medium (Top right) and coarse textures (Bottom right). TexRAD indicates grading of textures in color: red = fine, green = medium and blue = coarse. ROI = region of interest.
3.2. Texture analysis
The selection of relevant texture parameters was based on a simple Cox regression. Only 2 parameters were prognostic for overall survival with normalized uniformity (HR 16.06, confidence interval [CI] 95: 3.86–66.79, P-value < .001) and standard skewness (HR 1.91, CI 95: 1.33–2.75, P-value < .001). The remaining texture parameters and clinical parameters, including the total number of voxels, mean intensity, skewness, kurtosis, smoking history, and uniformity, were not associated with overall patient survival (Table 2).
Table 2.
Results for simple Cox regressions.
| Simple Cox regressions | ||||
|---|---|---|---|---|
| Parameters | HR | 95% CI | P-value | Test of proportional hazards P > chi2 |
| Normalized uniformity (medium/coarse filter) | 16.06 | 3.86–66.79 | <.001 | 0.6802 |
| Treatment | 3.10 | 1.54–6.25 | .002 | 0.1516 |
| Skewness medium filter | 1.91 | 1.33–2.75 | <.001 | 0.1992 |
| Clinical stage | 1.33 | 1.19–1.50 | <.001 | 0.2015 |
| Entropi coarse filter | 1.43 | 1.06–1.95 | .02 | 0.6401 |
3.3. Optimal cutoffs
Using the Fisher exact method, optimal cutoff points for normalized uniformity were determined to be 0.8602 for the feasibility dataset and 0.1554 for skewness medium filters.
3.4. Kaplan–Meier analysis for both groups
Based on the optimal threshold for both normalized uniformity for medium texture and standard skewness for medium texture, we found similar Kaplan–Meier curves for both the feasibility and validation datasets, thereby validating the model (Figs. 3 and 4).
Figure 3.
Kaplan–Meier analysis for skewness medium filter, with the determined cutoff value 0.1554, showing similar curves for the feasibility and validity data set.
Figure 4.
Kaplan–Meier analysis for normalized uniformity, with the determined cutoff value 0.8602, showing similar curves for the feasibility and validity data set.
Accompanying risk tables can be found in Table 3.
Table 3.
Accompanying risk tables for Kaplan–Meyer analysis for normalized uniformity and Skewness medium texture.
| Accompanying risk tables for Kaplan–Meyer analysis | ||||
|---|---|---|---|---|
| Patients at risk, n | ||||
| Normalized uniformity, feasibility group | ||||
| 19 | 4 | 2 | 1 | >0.8602 |
| 28 | 19 | 16 | 14 | <0.8602 |
| 0 | 500 | 1000 | 1500 | Survival (d) |
| Normalized uniformity, validity group | ||||
| 21 | 8 | 4 | 2 | >0.8602 |
| 26 | 20 | 13 | 9 | <0.8602 |
| 0 | 500 | 1000 | 1500 | Survival (d) |
| Skewness medium texture, feasibility group | ||||
| 17 | 7 | 3 | 3 | >–0.1554 |
| 30 | 16 | 15 | 12 | <–0.1554 |
| 0 | 500 | 1000 | 1500 | Survival (d) |
| Skewness medium texture, validity group | ||||
| 30 | 15 | 5 | 2 | >–0.1554 |
| 17 | 13 | 12 | 9 | <–0.1554 |
| 0 | 500 | 1000 | 1500 | Survival (d) |
3.5. Multiple Cox regression
Multiple Cox regression analysis suggested that standardized uniformity for medium filters (hazard ratio 14.42, CI 95: 2.78–74.92, P-value = .001), clinical stage (hazard ratio 1.25, CI 95: 1.03–1.51, P-value = .022), and skewness on medium filters (hazard ratio 1.51, CI 95: 1.01–2.27, P-value = .044) were prognostic factors for overall survival (Table 4). The proportional hazards assumption was tested in the analysis and showed no violation.
Table 4.
Results for multivariate Cox regression.
| Multivariate Cox regression | |||
|---|---|---|---|
| Parameters | HR | 95% CI | P-value |
| Variables included in the Cox analysis | |||
| Normalized uniformity (medium/coarse filter) | 14.42 | 2.78–74.92 | .001 |
| Clinical stage | 1.25 | 1.03–1.51 | .022 |
| Skewness medium filter | 1.51 | 1.01–2.27 | .044 |
| Parameters | HR | P-value | |
| Variables excluded in the Cox analysis | |||
| Age | 1.02 | .232 | |
| Metastatic stage | 1.28 | .594 | |
| Uniformity medium filter | 862.87 | .503 | |
3.6. Interobserver variability
We found good interobserver agreement with narrow limits of agreement, and an even spread around 0. The mean difference between readers for normalized uniformity was –0.023, with limits of ±0.334. For the skewness medium filter, the mean difference between readers was –0.38 with limits of ±0.2.66 (Fig. 5).
Figure 5.
Bland–Altman plots for normalized uniformity and skewness medium filter. (A) Normalized uniformity shows a few outliers, but also very narrow limits of agreement and an even spread around 0. (B) Skewness medium filter showing similar findings with narrow limits of agreement and an even spread around 0.
4. Discussion
In this retrospective study, we found that normalized uniformity texture parameters, skewness for medium filters, and entropy for coarse filters were significant prognostic factors for patient survival using simple Cox regression and Kaplan–Meier analysis. For normalized uniformity and skewness, the association with overall survival was maintained in the multiple Cox regression.
The publication of the new 8th edition of TNM staging shows that with each new iteration, we gain a new understanding of the factors that influence survival in lung cancer patients; however, TNM staging does not consider imaging biomarkers.[4] Some studies, including the one presented here, have shown a possible impact of texture parameters in the prognostication and staging of lung cancers.[7,14,24,25] An earlier study that focused on unenhanced CT derived from PET/CT examinations showed coarse uniformity to be the best prognostic factor for survival.[7] Most likely, the discrepancy between the findings presented in the current study, which found normalized uniformity to be the best predictor, and previous studies was caused by the addition of contrast to the examinations. This indicates that contrast administration can severely affect the texture parameters. Furthermore, differences in acquisition parameters could be another explanation. This suggests that to implement texture analysis in clinical practice, it is imperative to reach a consensus on the international harmonization of scanning protocols and contrast policies.
The reason why tumor heterogeneity is a prognostic factor for survival is currently not understood; however, an explanation might be that aggressive tumors have increased hypoxia and hence increased micro necrosis, giving rise to increased heterogeneity on CECT. It has been previously shown that some texture parameters correlates with tumor hypoxia assessed histopathologically.[5]
Several studies investigating various techniques for assessing tumor heterogeneity have been published.[13,25,26] The most recent technique implemented is radiomics, which uses both histogram analysis, such as the texture analysis presented in this study, and more advanced analysis, including gray-level co-occurrence matrix, run-length matrix, and morphological features of the lesion. In the present study, only histogram analysis was included, as all the parameters were well described and comprehensible. Radiomics features have previously been correlated with survival, the presence of distant metastases, and as a predictor of treatment response.[13,25–27]
As lung cancers are heterogeneous in nature, with multiple cell clones within the same tumor, we chose to assess the full volume of the tumor, even though prior studies and studies of tumors originating in other organ systems have suggested that a single-slice ROI sufficiently accounts for the inherent heterogeneity of lung cancers.
The cohort included patients with a variety of NSCLC pathologies, including adenocarcinomas (ADCs), squamous cell carcinomas (SCCs), large cell carcinomas, and blended tumors. Some studies have suggested that texture analysis/radiomics can differentiate between adenocarcinomas (ADCs) and squamous cell carcinomas (SCCs).[28] Most exhibit almost perfect differentiation on training data, but there is a significant drop in performance on clinical validation data.[29] The differences in texture parameters between various NSCLC pathologies could account for some of the variance in survival found in the current study.
The results presented in this feasibility study, in combination with previous results, show the possible use of imaging biomarkers as part of the prognostication for survival. In particular the current 8th TNM classification only considers morphological features, and the results presented here suggest that imaging biomarkers can further refine the classification of individual groups.
The present study has some limitations. First, the sample size was relatively small for survival analysis. Second, there was selection bias. The reason behind this was to ensure differences in acquisition parameters and contrast media policies, did not affect the measured heterogeneity within the lesions. Finally, a significant limitation is that all CECT come from a single center and from the same CT system.
In conclusion, we showed that for CECT, lung tumor heterogeneity assessed by normalized uniformity can be used as a prognostic factor for survival. It has the potential to be used as a biomarker in future revisions of the TNM classification. The limited sample size warrants further studies investigating various T stages within the TNM classification and in larger groups of patients.
Author contributions
Conceptualization: Michael Brun Andersen, Hans Henrik Torp Madsen, Finn Rasmussen.
Data curation: Michael Brun Andersen, Stefan Walbom Harders, Jesper Thygesen, Balaji Ganeshan, Finn Rasmussen.
Formal analysis: Michael Brun Andersen, Stefan Walbom Harders.
Project administration: Michael Brun Andersen.
Software: Balaji Ganeshan.
Supervision: Hans Henrik Torp Madsen, Finn Rasmussen.
Writing – original draft: Michael Brun Andersen.
Writing – review & editing: Michael Brun Andersen, Stefan Walbom Harders, Jesper Thygesen, Balaji Ganeshan, Hans Henrik Torp Madsen, Finn Rasmussen.
Abbreviations:
- 18-FDG-PET/CT =
- 18-flourdeoxyglucose positron emission tomography/computed tomography
- CECT =
- contrast enhanced computed tomography
- CI =
- confidence interval
- CTTA =
- CT texture analysis
- NSCLC =
- non-small cell lung cancer
- ROI =
- region of interest
- TNM =
- tumor-node-metastasis
The authors received salaries from their individual institutions. No additional funding was received from the work presented in this manuscript.
One of the authors (BG) is the co-founder/co-inventor of TexRAD texture analysis software used in this study and a shareholder (not an employee) of Feedback Plc., a UK based company that owns, develops, and markets TexRAD texture analysis software. The remaining authors have no conflicts of interest to disclose.
The datasets generated during and/or analyzed during the current study are available from the corresponding author on reasonable request.
How to cite this article: Andersen MB, Harders SW, Thygesen J, Ganeshan B, Torp Madsen HH, Rasmussen F. Potential impact of texture analysis in contrast enhanced CT in non-small cell lung cancer as a marker of survival: A retrospective feasibility study. Medicine 2022;101:48(e31855).
Contributor Information
Stefan Walbom Harders, Email: swharders@gmail.com.
Jesper Thygesen, Email: jesthy@rm.dk.
Balaji Ganeshan, Email: B.ganeshan@ucl.ac.uk.
Hans Henrik Torp Madsen, Email: Hansmad@rm.dk.
Finn Rasmussen, Email: finnrasm@rm.dk.
References
- [1].Danckert B, Ferlay J, Engholm G, et al. NORDCAN: Cancer incidence, mortality, prevalence and survival in the Nordic countries, version 8.2.. 2019. Available at: http://www-dep.iarc.fr/NORDCAN/DK/frame.asp. [access date May 28, 2019]. [Internet].
- [2].Chao F, Zhang H. PET/CT in the staging of the non-small-cell lung cancer. J Biomed Biotechnol. 2012. Available at: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3346692/. [access date May 28, 2019]. [Internet]. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [3].Harders SW, Madsen HH, Hjorthaug K, et al. Mediastinal staging in non-small-cell lung carcinoma: computed tomography versus F-18-fluorodeoxyglucose positron-emission tomography and computed tomography. Cancer Imaging. 2014;14:23. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [4].Goldstraw P, Chansky K, Crowley J, et al. The IASLC lung cancer staging project: proposals for revision of the TNM stage groupings in the forthcoming (Eighth) Edition of the TNM classification for lung cancer. J Thorac Oncol. 2016;11:39–51. [DOI] [PubMed] [Google Scholar]
- [5].Ganeshan B, Goh V, Mandeville HC, et al. Non-small cell lung cancer: histopathologic correlates for texture parameters at CT. Radiology. 2013;266:326–36. [DOI] [PubMed] [Google Scholar]
- [6].Nelson DA, Tan TT, Rabson AB, et al. Hypoxia and defective apoptosis drive genomic instability and tumorigenesis. Genes Dev. 2004;18:2095–107. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [7].Ganeshan B, Panayiotou E, Burnand K, et al. Tumour heterogeneity in non-small cell lung carcinoma assessed by CT texture analysis: a potential marker of survival. Eur Radiol. 2012;22:796–802. [DOI] [PubMed] [Google Scholar]
- [8].Mulé S, Thiefin G, Costentin C, et al. Advanced hepatocellular carcinoma: pretreatment contrast-enhanced CT texture parameters as predictive biomarkers of survival in patients treated with Sorafenib. Radiology. 2018;288:445–55. [DOI] [PubMed] [Google Scholar]
- [9].Zhang H, Graham CM, Elci O, et al. Locally advanced squamous cell carcinoma of the head and neck: CT texture and histogram analysis allow independent prediction of overall survival in patients treated with induction chemotherapy. Radiology. 2013;269:801–9. [DOI] [PubMed] [Google Scholar]
- [10].Ng F, Ganeshan B, Kozarski R, et al. Assessment of primary colorectal cancer heterogeneity by using whole-tumor texture analysis: contrast-enhanced CT texture as a biomarker of 5-year survival. Radiology. 2013;266:177–84. [DOI] [PubMed] [Google Scholar]
- [11].Chee CG, Kim YH, Lee KH, et al. CT texture analysis in patients with locally advanced rectal cancer treated with neoadjuvant chemoradiotherapy: a potential imaging biomarker for treatment response and prognosis. PLoS One. 2017;12:e0182883. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [12].Yang X, Pan X, Liu H, et al. A new approach to predict lymph node metastasis in solid lung adenocarcinoma: a radiomics nomogram. J Thorac Dis. 2018;10(Suppl 7):S807–19. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [13].Zhou H, Dong D, Chen B, et al. Diagnosis of distant metastasis of lung cancer: based on clinical and radiomic features. Transl Oncol. 2017;11:31–6. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [14].Miles KA. How to use CT texture analysis for prognostication of non-small cell lung cancer. Cancer Imaging. 2016;16:10. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [15].Weiss GJ, Ganeshan B, Miles KA, et al. Noninvasive image texture analysis differentiates K-ras mutation from pan-wildtype NSCLC and is prognostic. PLoS One. 2014;9:e100244. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [16].Andersen MB, Harders SW, Ganeshan B, et al. CT texture analysis can help differentiate between malignant and benign lymph nodes in the mediastinum in patients suspected for lung cancer. Acta Radiol Stockh Swed 1987. 2016;57:669–76. [DOI] [PubMed] [Google Scholar]
- [17].R: A language and environment for statistical computing. Vienna, Austria: R Foundation for Statistical Computing. 2020. Available at: https://www.R-project.org/ [accessed December 2020]. [Google Scholar]
- [18].dplyr: a grammar of data manipulation. [accessed December 2020].
- [19].Therneau TM, Grambsch PM. A package for survival analysis in R. 2000. Available at: https://CRAN.R-project.org/package=survival [accessed December 2020].
- [20].Kassambara A, Kosinski M, Biecek P. survminer: drawing survival curves using “ggplot2.” 2021. Available at: https://CRAN.R-project.org/package=survminer [accessed December 2020].
- [21].Dardis C. survMisc: miscellaneous functions for survival data. 2018. Available at: https://CRAN.R-project.org/package=survMisc [accessed December 2020].
- [22].Miles KA, Ganeshan B, Griffiths MR, et al. Colorectal cancer: texture analysis of portal phase hepatic CT images as a potential marker of survival. Radiology. 2009;250:444–52. [DOI] [PubMed] [Google Scholar]
- [23].Budczies J, Klauschen F, Sinn BV, et al. Cutoff finder: a comprehensive and straightforward web application enabling rapid biomarker cutoff optimization. PLoS One. 2012;7:e51862. Available at: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3522617/. [access date May 29, 2019]. [Internet]. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [24].Ravanelli M, Agazzi GM, Ganeshan B, et al. CT texture analysis as predictive factor in metastatic lung adenocarcinoma treated with tyrosine kinase inhibitors (TKIs). Eur J Radiol. 2018;109:130–5. [DOI] [PubMed] [Google Scholar]
- [25].Virginia BM, Laura F, Silvia R, et al. Prognostic value of histogram analysis in advanced non-small cell lung cancer: a radiomic study. Oncotarget. 2017;9:1906–14. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [26].Shi L, He Y, Yuan Z, et al. Radiomics for response and outcome assessment for non-small cell lung cancer. Technol Cancer Res Treat. 2018;17:1533033818782788. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [27].van Timmeren JE, Leijenaar RTH, van Elmpt W, et al. Survival prediction of non-small cell lung cancer patients using radiomics analyses of cone-beam CT images. Radiother Oncol. 2017;123:363–9. [DOI] [PubMed] [Google Scholar]
- [28].Liu H, Jing B, Han W, et al. A comparative texture analysis based on NECT and CECT images to differentiate lung adenocarcinoma from squamous cell carcinoma. J Med Syst. 2019;43:59. [DOI] [PubMed] [Google Scholar]
- [29].Bashir U, Kawa B, Siddique M, et al. Non-invasive classification of non-small cell lung cancer: a comparison between random forest models utilising radiomic and semantic features. Br J Radiol. 2019;92:2019015920190159. [DOI] [PMC free article] [PubMed] [Google Scholar]