Medical imaging covers a wide range of techniques that are used to look inside the body, including X-rays, MRI scans and ultrasound. A process called radiomics uses computer algorithms to process the data collected by these techniques to identify and precisely measure a large number of features that would not otherwise be quantifiable by human experts. By doing so, radiomics can automatically measure the radiographic characteristics of a tumor. For example, radiomics can establish the size, shape and texture of a tumor to help to diagnose cancer and guide its treatment.

Research has suggested that radiomics can predict certain clinical characteristics of cancer, such as how far through the body the cancer has spread, how likely it is to respond to treatment, and how likely a patient is to survive. However, these radiomic characteristics have not yet been precisely linked to the biological processes that drive how cancer develops and spreads.

Cancers develop as a result of genetic changes that activate “molecular pathways” in the cells and trigger processes such as cell division and inflammation. To work out exactly which changes are behind a particular tumor, a sample of the tumor from biopsy or surgery is analyzed using genomics techniques. Linking radiomics features to the molecular processes active in a tumor can generate further information that can complement the molecular data. Images are routinely collected on all cancer patients yet molecular data is not. Hence, in some cases, the images can be used to infer the molecular underpinnings of cancer in individual patients.

Grossmann et al. have now analyzed radiomic, genomic and clinical data collected from approximately 350 patients with lung cancer. The analysis revealed links between biological processes normally detected by genomics – in particular, inflammatory responses – and radiomics features. Furthermore, these features could also be associated with clinical characteristics, such as tumor type and patient survival rates. These results were further validated by using a technique called immunohistochemical staining on tumor tissue obtained by surgery.

Further investigation revealed that certain radiomics features can predict the state of molecular pathways that are key to cancer development (such as the inflammatory response). Furthermore, Grossmann et al. found that combining data from radiomics, genomics and clinical parameters predicts how the cancer will progress better than any of these parameters can predict on their own. These results demonstrate the complementary value of radiomic data to genomic and clinical data.

There are many different algorithms that can be used to process images for radiomics. Before radiomics can be used clinically to assess the biological processes underlying the tumors of patients, a specific algorithm needs to be decided upon and then tested in prospective clinical trials.

‘Precision medicine’ promotes the molecular characterization of a patient’s tumor with genomic approaches, which requires tissue extraction usually obtained via biopsy. A number of examples demonstrate successful translation of genomic information obtained from biopsies into clinical applications (Doroshow and Kummar, 2014), but these approaches also have inherent limitations, such as their invasive nature or sampling artifacts caused by intra-tumor heterogeneity (Sottoriva et al., 2013; Fisher et al., 2013; Gerlinger et al., 2012). These limitations can be addressed by medical imaging that has served as crucial diagnostic tool and treatment guidance in clinical oncology. In contrast to biopsies, medical imaging is usually noninvasive, can be applied longitudinally, and provides information about the entire visible tumor volume. In this way, medical imaging has the potential to characterize phenotypic information of tumors and thus complement molecular interrogation (Choi et al., 2016). As imaging is already used routinely throughout the course of treatment this facilitates ready access to this type of data. Therefore, imaging has the potential to serve as valuable diagnostic tool in clinical decision making by complementing biological interrogation or serving as a surrogate in settings where biospecimen-derived diagnostics is not feasible, such as in longitudinal monitoring.

Radiomics is an emerging field that translates these medical images into mineable data by extracting a large number of quantitative imaging features that objectively define tumor intensity, shape, size, and texture (Gillies et al., 2016; Aerts, 2016; Lambin et al., 2012; Kumar et al., 2012) in a robust and reproducible way (Zhao et al., 2016; Fried et al., 2014; Balagurunathan et al., 2014; Leijenaar et al., 2013). As this approach is applied to existing standard of care images, radiomics can be cost-effectively integrated with genomics or serve as surrogate in cases where biopsies are not feasible (O'Connor et al., 2015). Hence, such strategies can be of value for the development of clinical biomarkers for diagnosis, prognosis, and prediction of response to specific treatments (Choi et al., 2016; Huang et al., 2016a, 2016b; Aerts et al., 2016; Nicolasjilwan et al., 2015; Parmar et al., 2015a, Parmar et al., 2015b; Aerts et al., 2014; Chong et al., 2014; Coroller et al., 2015; Gevaert et al., 2012; Ganeshan et al., 2012; Win et al., 2013; Mattonen et al., 2016; Grossmann et al., 2017). Due to the enormous potential for precision medicine, an increasing number of studies have investigated associations between imaging and tumor biology in different cancer types (Aerts et al., 2014; Gevaert et al., 2012; Diehn et al., 2008; Grossmann et al., 2016; Gutman et al., 2015; Segal et al., 2007; Li et al., 2016; Yoon et al., 2015). However, these studies focused on specific genetic associations, or tended to be underpowered due to a limited number of available samples and lacked validation via independent datasets.

Here, we present a broad radiomic-genomic analysis in independent and large cohorts of patients with lung cancer. We rigorously investigated the mechanistic connections between imaging phenotypes and underlying molecular pathways. Furthermore, we validated key associations via immunohistochemical staining and related these associations to clinical factors. In addition, we developed and validated radiomic predictors of pathway activation status, and investigated the prognostic value of combining radiomic biomarkers with genetic and clinical data. In this study, we aimed at uncovering whether radiomic approaches have the potential to predict both molecular and clinical characteristics of tumors noninvasively and therefore have the potential to augment clinical decision-making using data extracted from standard of care medical images.

To uncover the mechanistic connections between radiomic phenotypes, molecular pathways, and clinical information, we performed an integrated radiomic-genomic analysis of a lung cancer discovery cohort (Dataset1, n = 262), and validated our results on an independent validation cohort (Dataset2, n = 89). We defined and extracted 636 radiomic features from CT scans (Figure 1A) quantifying tumor intensity, shape, and texture (Figure 1B), detailed in Supplementary file 1. Our approach to integrate radiomic, genomic, and clinical data is outlined in Figure 2 and clinical cohort characteristics are given in Table 1. Dataset1 can be downloaded as Figure 2—source data 1 and Dataset2 can be downloaded as Figure 2—source data 2. 

Figure 1

Download asset Open asset

Figure 2

Download asset Open asset

Figure 2—source data 1

Dataset1.

Spreadsheet containing radiomic data, normalized gene expression, and clinical data of the discovery cohort.

https://doi.org/10.7554/eLife.23421.005

Download elife-23421-fig2-data1-v3.xlsx

Figure 2—source data 2

Dataset2.

Spreadsheet containing radiomic data, normalized gene expression, and clinical data of the validation cohort.

https://doi.org/10.7554/eLife.23421.006

Download elife-23421-fig2-data2-v3.xlsx

Association modules of radiomic features and molecular pathways

To investigate the main associations of radiomics and underlying molecular pathways, we developed association modules describing radiomic-pathway coherency. Bi-clustering allowed simultaneous grouping of coherently expressed features and pathways into a single module, thereby reducing dimensionality. Using this approach, we identified thirteen radiomic-pathway modules in Dataset1 that were independently validated in Dataset2 (FDR < 0.05). Figure 3A and Table 2 summarize these modules, while a detailed version of every module is given in Figure 3—source data 1.

Figure 3 with 2 supplements see all

Download asset Open asset

Figure 3—source data 1

Enlarged heatmaps of every module depicting normalized enrichment scores (NESs) of every pair of radiomic feature and molecular pathway clustered in a module.

https://doi.org/10.7554/eLife.23421.009

Download elife-23421-fig3-data1-v3.pdf

Table 2

Module	nr	np	Radiomic	Pathway
M1	6	7	Wavelet texture gray-level runs	Lipid and lipoprotein metabolism, Notch signaling, circadian clock
M2	58	5	Wavelet intensity entropy; Laplace of Gaussian intensity standard deviation	Immune system, p53
M3	4	17	Wavelet minimum intensity	Neural system, axon guidance
M4	25	14	Intensity variance and mean; wavelet minimum intensity min	Biological oxidations, signaling by insulin receptor, signaling by GPCR, neuronal system
M5	58	8	Wavelet texture gray-level runs; wavelet intensity range and median; (wavelet) texture information correlation and cluster tendency	Axon guidance and synaptic transmission, lipoprotein metabolism, cell type determination
M6	64	7	Laplace of Gaussian standard deviation; wavelet texture gray-level runs; wavelet texture cluster tendency	Circadian clock, signaling by Notch
M7	39	8	Laplace of Gaussian intensity entropy; wavelet intensity variance; Laplace of Gaussian texture information correlation	Mitochondria, Pol III transcription
M8	20	17	Laplace of Gaussian standard deviation	TCA cycle and electron transport, TGF-beta receptor signaling, response to stress, transcription regulation, protein synthesis,
M9	8	30	Intensity variance; wavelet intensity variance	Immune system, p53, cell cycle regulation checkpoints, cell-cell interaction, circadian clock
M10	5	83	Shape surface (SH); wavelet texture gray-level runs	Axon guidance, neuronal system, (innate) immune system, hemostasis, FGFR signaling, TGF-beta receptor signaling, Notch signaling, circadian clock
M11	17	66	Wavelet intensity range; wavelet texture information correlation	Hemostasis, neural system
M12	32	27	Wavelet texture entropy; intensity variance; wavelet texture cluster tendency	P53, immune system
M13	39	26	Intensity entropy	Gene expression regulation, Pol II/III transcription

In general, we found that distinct radiomic features were associated with distinct biological processes. For example, texture entropy and cluster features, as well as voxel intensity variance features were associated with the immune system, the p53 pathway, and other pathways involved in cell cycle regulation in modules M2, M9, and M12 (Table 2 and Figure 3A). In another module (M8), we found those features to also be associated with transforming growth factor beta (TGF-β) receptor signaling.

Further examples for radiomic-pathway links included two modules (M13 and M7) that were highly enriched for pathways involved in mitochondrial pathways, transcription, translation, and RNA regulatory mechanisms; with only one exception, all features in the larger module (M13) were voxel intensity entropy features. In addition to this feature type, the smaller module (M7) contained mainly textural variance and information correlation features.

Immunohistochemical investigation

To further investigate putative connections between radiomics, immune response pathways, and OS we performed immunohistochemical staining of 22 tumors for CD3, a T-cell co-receptor. These tumors were predicted to show relatively high or low immune response by a radiomics feature selected from the three modules (M2, M9, and M12) that were associated with OS. As represented in Figure 4, we found agreement between radiomics and pathology; cases that were pathologically scored to have high CD3 enrichment also expressed significantly higher radiomic values (one-sided Wilcoxon rank sum test, p=0.008). Furthermore, we tested the extent to which radiomic predictors of inflammation can be reproduced immunohistochemically. We built on our previous results suggesting that the radiomic shape feature sphericity predicts NFkB activation (module 10) and analyzed 24 stained tumors that were predicted to have relatively high or low NFkB activity for RelA, the p65 subunit of NFkB (Figure 4—figure supplement 1). Pathological assessment of enrichment for RelA revealed that those cases that indicated high RelA enrichment on average also had higher radiomic feature scores (one-sided Wilcoxon rank sum test, p=0.06).

Figure 4 with 1 supplement see all

Download asset Open asset

Prognostic value of radiomic signatures

To build on previously published results, we investigated prognostic value of an existing radiomic signature for survival of lung cancer. We fitted a Cox proportional-hazards model of this signature on Dataset1 and observed significant validation by the concordance-index (CI) on Dataset2 (CI = 0.60, Noether p=0.04). Furthermore, we tested combinations of clinical, genetic, and radiomic data and observed that the combinations of data types tended to result in higher performances than given by the individual data alone (Figure 5). In particular, the performance of a clinical model increased from CI = 0.65 (Noether p=0.001) to CI = 0.73 (p=2×10⁻⁹) when adding the radiomic and an existing gene signature (Hou et al., 2010); this increase was significant at p=0.001 by permutation test. This combined radiomic-genetic-clinical model also performed significantly better than the combined radiomic-clinical model (p=0.007) and the clinical-genetic model (p=0.01). Adding radiomics to clinical data alone did not result in a significant increase (p=0.3). We repeated this analysis with a novel radiomic survival signature and other published gene signatures (Yuan et al., 2004; Chen et al., 2007; Hsu et al., 2009), and found that the clinical-genetic-radiomic models consistently yielded the highest performances in nearly all cases (Figure 5—figure supplement 1 and Figure 5—figure supplement 2).

Figure 5 with 2 supplements see all

Download asset Open asset

Medical imaging plays a crucial role in cancer diagnosis, treatment, and response monitoring. Radiomics allows quantification of the radiographic phenotype of a tumor (Kuo and Jamshidi, 2014; Gillies et al., 2010; Rutman and Kuo, 2009), but the underlying connections of radiomics to tumor biology and clinical factors have not been elucidated yet. In this study, we identified novel and consistent associations between radiomic phenotype data, underlying molecular pathways, and clinical factors of patients with lung cancer in a North American cohort, and validated our findings in a European cohort and with immunohistochemical staining. In addition, we presented radiomic predictors for pathway activations, and demonstrated the complementary prognostic value of combining radiomic, genetic, and clinical information.

Preliminary studies have previously investigated associations between imaging features, clinical factors, and molecular data for a number of cancer types as outlined in recent reviews (Gillies et al., 2016; Kuo and Jamshidi, 2014; Gillies et al., 2010; Rutman and Kuo, 2009; Cook et al., 2013). Our analysis builds on these studies in that we performed a rigorous classification of a comprehensive set of radiomic features in terms of underlying molecular pathways on a genome-wide scale and clinical factors in large and independent cohorts. Although the long-term vision is to augment clinical decision making, the current goal of our study is to satisfy the need of the radiomic and oncological community to better understand the underlying biological rationale of radiomic predictions. Furthermore, we are the first to publicly share all study data and analysis code with the growing radiomic and biomedical community to enable further translational research.

We identified and independently validated thirteen radiomic-pathway modules with coherent expression patterns, eleven of which were significantly associated with OS, stage, or histology. By basing these clinical associations exclusively on radiomic features, we could demonstrate that the associated molecular pathways robustly matched radiomic- based hypotheses. For example, based on radiomic features modules M2, M9, and M12 were prognostic and also associated with stage. These modules were highly enriched for immune system, p53, and cell-cycle regulation pathways, biological processes that are widely recognized to play key roles in lung cancer. For example, it has been established that cell cycle regulation is of utmost importance in lung cancer (Baldi et al., 2011). Furthermore, the status of p53 is reported to be a predictor of survival in lung cancer patients (Ahrendt et al., 2003) and a recent review has laid out how p53 can modulate innate immune system responses (Menendez et al., 2013). Radiomic features in these prognostic modules M2, M9, and M12 quantified textural entropy and dispersion image intensity values suggesting associations between textural heterogeneity, cell cycling, and prognosis. Therefore, these results suggest that noninvasive radiomic surrogates may benefit diagnostic methods in assessing cell cycling and immune system states of tumors.

We aimed at confirming our statistical results indicating connections between radiomics, immune response, and survival by immunohistochemical staining of lymphocytes in cases for which a relatively high or low immune response was predicted according to a radiomics score. We generally found high agreement between pathology and radiomics, especially in cases where immune response was predicted to be high. In cases of predicted low responders that showed high pathological immune response, the cause of disagreement may be a heavy distribution of CD3 clusters in the extreme periphery of the tumor with very little staining in the bulk of the tumor. In cases of predicted high responders that showed little to no immune response, this could be due to the lack of normal tissue margin around the edge of the tumor section or a sampling effect. Similarly, we stained tissue for RelA, the p65 subunit of NFkB, to validate radiomic predictions of inflammation. Overall, we found high agreement between pathology and radiomics, although at lower statistical significance. Future studies with whole mount sections stained with multi-plex phenotyping can help determine the relationship between a radiomic and a genetic immune or inflammation signature, and the gold standard.

A variety of textural features were also associated with stage and histology (module M5). Similar associations have been reported by Ganeshan et al. (2010), who suggested that 2D texture features of lung cancer CT scans could predict if tumor stage was II or above. Here, we found that texture features were enriched for axon guidance and lipoprotein metabolism. Furthermore, we observed strong associations between image intensity entropy features and pathways involved in gene expression, transcription regulation, and mitochondrial processes (M13 and M7). Previous research has suggested that imaging can detect consequences of an increase in the hypoxia-inducible factor as a result of absence of oxygen (Gillies et al., 2010). Hence, if extracting quantitative information about mitochondrial pathways from medical images leads to assessment of hypoxia status of a tumor, this may ultimately aid in clinical decision-making as alternative therapies for hypoxic tumor areas are being developed (Denny, 2010; Bryant et al., 2014). Indeed, previous work has indicated that CT pixel intensities correlate with hypoxia markers such as Glut-1 and pimonidazole (Ganeshan et al., 2013). Those two modules (M1 and M3) that were not associated with any of the tested clinical factors were relatively small modules; these modules suggested radiomic associations to circadian clock and neural system. The impact of these pathways to the clinical factors we tested is not apparent from current lung cancer literature, which could explain why these modules did not show clinical associations.

Our results further suggest that radiomic approaches could have the potential to predict molecular states of pathways. We found that the highest predictors of every module was also a suitable representative of the overall biological and radiomic themes of that module. Amongst these examples of pathways that showed high predictability in terms of radiomics, we found various pathways essential for tumorigenesis such as cell cycle pathways (e.g., G0 and early G1), signaling pathways (e.g., Notch and NfKB), and tumor suppressor pathways (e.g., COP1 autodegration and p53). Furthermore, we tested those radiomic pathway predictors for predictive value of driver mutations. Thereby, the highest performances were found for mutations in EGFR and KRAS, which is in line with current radiomic-genetic literature (Aerts et al., 2016; Gutman et al., 2015; Liu et al., 2016; Rizzo et al., 2016). Interestingly, however, the highest performance for the tumor suppressor and cell cycle regulator TP53 we found was given by a textural entropy feature that also predicted G0 and early G1 (module M12). In addition, features expressed selectivity for predicting mutations, which was suggested previously (Gutman et al., 2015). These results highlight the diagnostic potential, as ready information on pathway and mutation status may permit advanced patient stratification. Previous studies have indicated that gene expression can be predicted by imaging features (Gevaert et al., 2012; Segal et al., 2007; Gevaert et al., 2014). To our knowledge, however, no study has examined and independently validated radiomic models for specific pathways, including biological validation such as immunohistochemical staining.

Finally, we verified a previously described prognostic radiomic signature and observed that the best performance is achieved when combining radiomic, genetic, and clinical data. These results strongly suggest that radiomic data contain complementary prognostic information and are robust, as the published radiomic signature (Aerts et al., 2014) has not been tested on our data before. Notably, these prognostic improvements were relatively stable to substitution of radiomic or gene signatures. A related indication of improved survival predictions by combining imaging features and molecular data has been recently given for glioblastoma, however without validation (Nicolasjilwan et al., 2015). It is worth noting that for the first time we also demonstrate that radiomic prognostication generalizes across cohorts from different continents.

Three research tracks have recently been proposed for clinical translation of such imaging biomarkers (O'Connor et al., 2017), including biological validation, technical validation, and evaluation of cost-effectiveness. Our study conforms with several of these roadmap recommendations by advancing results on a previously proposed radiomic signature (Aerts et al., 2014) with additional biological validation and investigations on how genetic data and clinical factors impact this signature. Fixing a radiomic signature for technical validation and cost-effectiveness verification should be considered in subsequent studies to overcome additional translational gaps. Although the long-term vision would be to augment clinical decision making, the current goal of our study is to contribute in satisfying the need of the radiomic and oncological community to understand the underlying biology of radiomic predictions.

Our study is limited by its retrospective nature. Imaging protocols are not standardized and hence variability in CT acquisition and reconstruction parameters is inherent in clinical practice. However, despite this, no corrections by cohort or scanner type were made in this study illustrating the translational aspect of our results that generalized across institutions. Hence, we expect that the performance of radiomics will further improve, as imaging data are becoming more standardized. In fact, multiple studies have already documented the robustness of radiomic feature extractions in terms of reproducibility and repeatability in test/re-test settings (Fried et al., 2014; Balagurunathan et al., 2014; Leijenaar et al., 2013; Parmar et al., 2015a; Aerts et al., 2014; Grove et al., 2015). Another limitation of this study is that the current cohorts mainly focused on early stage (I - III) tumors, hence generalization of radiomic-genomic associations to late stage tumors should be drawn with precaution only. However, most radiomic applications do focus on early stage tumors as the current radiomic approach requires segmentation of tumors which for late stage tumors remains to be of particular complexity. Furthermore, although our study provides multiple facets of validation, immunohistochemical validation was restricted to considerably smaller sample sizes as compared to our statistical validations due to limited availability of frozen tissue. Prospective protocols can ensure availability of sufficient tissue for additional validation.

Biological material investigated in this study has been acquired by single-needle biopsies, thus the interpretation of our genomic data is limited due to heterogeneity of lung cancer tumors. However, as our results validated in independent data and because known drivers of tumorigenesis were among the main pathways found to be associated with radiomic features, this suggests that these associations have been established in an early evolutionary step in tumorigenesis and are therefore reasonable representatives of the overall tumor. Prospective studies with defined spatial matchings of biopsies and/or single cell analyses could provide deeper insight into whether the strengths of these associations can be further increased. Prospective studies will also be required to assess clinical utility of combining radiomic, genomic, and clinical data into prognostic models.

In conclusion, this study presented novel and consistent associations between radiomics, molecular pathways, and clinical factors. We applied an independent discovery and validation design on large patient cohorts from different continents with enough variability that allowed confidence in the generalization of our results. Furthermore, we performed biological validation and demonstrated that radiomics predicts molecular pathway status and thus improves the prognostic performances of clinical and gene signatures. The clinical impact of our results is illustrated by the fact that it advances the molecular knowledge of automated radiomic characterization of tumors, information currently not used clinically. This may provide opportunities to improve decision-support at low additional cost as imaging is routinely used in clinical practice as standard of care.

1. 1. Aerts HJ
   2. Grossmann P

   (2014) 89 NSCLC patients with gene expression profiles and matching CT imaging data available at TCIA

   Publicly available at the NCBI Gene Expression Omnibus (accession no: GSE58661).

   https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE58661

1. 1. Aerts HJ
   2. Velazquez ER
   3. Leijenaar RT
   4. Parmar C
   5. Grossmann P
   6. Carvalho S
   7. Cavalho S
   8. Bussink J
   9. Monshouwer R
   10. Haibe-Kains B
   11. Rietveld D
   12. Hoebers F
   13. Rietbergen MM
   14. Leemans CR
   15. Dekker A
   16. Quackenbush J
   17. Gillies RJ
   18. Lambin P

   (2014) Decoding tumour phenotype by noninvasive imaging using a quantitative radiomics approach

   Nature Communications 5:4006.

   https://doi.org/10.1038/ncomms5006

   • PubMed
   • Google Scholar
2. 1. Aerts HJ

   (2016) The potential of radiomic-based phenotyping in precision medicine: a review

   JAMA Oncology 2:1636.

   https://doi.org/10.1001/jamaoncol.2016.2631

   • PubMed
   • Google Scholar
3. 1. Aerts HJ
   2. Grossmann P
   3. Tan Y
   4. Oxnard GG
   5. Rizvi N
   6. Schwartz LH
   7. Zhao B

   (2016) Defining a radiomic response phenotype: a pilot study using targeted therapy in NSCLC

   Scientific Reports 6:33860.

   https://doi.org/10.1038/srep33860

   • PubMed
   • Google Scholar
4. 1. Ahrendt SA
   2. Hu Y
   3. Buta M
   4. McDermott MP
   5. Benoit N
   6. Yang SC
   7. Wu L
   8. Sidransky D

   (2003) p53 mutations and survival in stage I non-small-cell lung Cancer: results of a prospective study

   JNCI Journal of the National Cancer Institute 95:961–970.

   https://doi.org/10.1093/jnci/95.13.961

   • PubMed
   • Google Scholar
5. 1. Balagurunathan Y
   2. Gu Y
   3. Wang H
   4. Kumar V
   5. Grove O
   6. Hawkins S
   7. Kim J
   8. Goldgof DB
   9. Hall LO
   10. Gatenby RA
   11. Gillies RJ

   (2014) Reproducibility and prognosis of quantitative features extracted from CT Images

   Translational Oncology 7:72–87.

   https://doi.org/10.1593/tlo.13844

   • PubMed
   • Google Scholar
6. 1. Baldi A
   2. De Luca A
   3. Esposito V
   4. Campioni M
   5. Spugnini EP
   6. Citro G

   (2011) Tumor suppressors and cell-cycle proteins in lung Cancer

   Pathology Research International 2011:1–12.

   https://doi.org/10.4061/2011/605042

   • PubMed
   • Google Scholar
7. 1. Benjamini Y
   2. Hochberg Y

   (1995) Controlling the False Discovery Rate: a Practical and Powerful Approach to multiple testing

   Journal of the Royal Statistical Society. Series B, Statistical Methodology 57:289–300.

   https://doi.org/10.2307/2346101

   • Google Scholar
8. 1. Bergmann S
   2. Ihmels J
   3. Barkai N

   (2003) Iterative signature algorithm for the analysis of large-scale gene expression data

   Physical Review E 67:031902.

   https://doi.org/10.1103/PhysRevE.67.031902

   • PubMed
   • Google Scholar
9. 1. Bradley AP

   (1997) The use of the area under the ROC curve in the evaluation of machine learning algorithms

   Pattern Recognition 30:1145–1159.

   https://doi.org/10.1016/S0031-3203(96)00142-2

   • Google Scholar
10. 1. Bryant JL
    2. Meredith SL
    3. Williams KJ
    4. White A

    (2014) Targeting hypoxia in the treatment of small cell lung Cancer

    Lung Cancer 86:126–132.

    https://doi.org/10.1016/j.lungcan.2014.08.003

    • PubMed
    • Google Scholar
11. 1. Chen HY
    2. Yu SL
    3. Chen CH
    4. Chang GC
    5. Chen CY
    6. Yuan A
    7. Cheng CL
    8. Wang CH
    9. Terng HJ
    10. Kao SF
    11. Chan WK
    12. Li HN
    13. Liu CC
    14. Singh S
    15. Chen WJ
    16. Chen JJ
    17. Yang PC

    (2007) A five-gene signature and clinical outcome in non-small-cell lung Cancer

    New England Journal of Medicine 356:11–20.

    https://doi.org/10.1056/NEJMoa060096

    • PubMed
    • Google Scholar
12. 1. Choi ER
    2. Lee HY
    3. Jeong JY
    4. Choi YL
    5. Kim J
    6. Bae J
    7. Lee KS
    8. Shim YM

    (2016) Quantitative image variables reflect the intratumoral pathologic heterogeneity of lung adenocarcinoma

    Oncotarget 7:67302–67313.

    https://doi.org/10.18632/oncotarget.11693

    • PubMed
    • Google Scholar
13. 1. Chong Y
    2. Kim JH
    3. Lee HY
    4. Ahn YC
    5. Lee KS
    6. Ahn MJ
    7. Kim J
    8. Shim YM
    9. Han J
    10. Choi YL

    (2014) Quantitative CT variables enabling response prediction in neoadjuvant therapy with EGFR-TKIs: are they different from those in neoadjuvant concurrent chemoradiotherapy?

    PLoS One 9:e88598.

    https://doi.org/10.1371/journal.pone.0088598

    • PubMed
    • Google Scholar
14. 1. Cook GJ
    2. Yip C
    3. Siddique M
    4. Goh V
    5. Chicklore S
    6. Roy A
    7. Marsden P
    8. Ahmad S
    9. Landau D

    (2013) Are pretreatment 18F-FDG PET tumor textural features in non-small cell lung Cancer associated with response and survival after chemoradiotherapy?

    Journal of Nuclear Medicine 54:19–26.

    https://doi.org/10.2967/jnumed.112.107375

    • PubMed
    • Google Scholar
15. 1. Coroller TP
    2. Grossmann P
    3. Hou Y
    4. Rios Velazquez E
    5. Leijenaar RT
    6. Hermann G
    7. Lambin P
    8. Haibe-Kains B
    9. Mak RH
    10. Aerts HJ

    (2015) CT-based radiomic signature predicts distant metastasis in lung adenocarcinoma

    Radiotherapy and Oncology 114:345–350.

    https://doi.org/10.1016/j.radonc.2015.02.015

    • PubMed
    • Google Scholar
16. 1. Csárdi G
    2. Kutalik Z
    3. Bergmann S

    (2010) Modular analysis of gene expression data with R

    Bioinformatics 26:1376–1377.

    https://doi.org/10.1093/bioinformatics/btq130

    • PubMed
    • Google Scholar
17. 1. De Jay N
    2. Papillon-Cavanagh S
    3. Olsen C
    4. El-Hachem N
    5. Bontempi G
    6. Haibe-Kains B
    7. Jay ND

    (2013) mRMRe: an R package for parallelized mRMR ensemble feature selection

    Bioinformatics 29:2365–2368.

    https://doi.org/10.1093/bioinformatics/btt383

    • PubMed
    • Google Scholar
18. 1. Denny WA

    (2010) Hypoxia-activated prodrugs in Cancer therapy: progress to the clinic

    Future Oncology 6:419–428.

    https://doi.org/10.2217/fon.10.1

    • PubMed
    • Google Scholar
19. 1. Diehn M
    2. Nardini C
    3. Wang DS
    4. McGovern S
    5. Jayaraman M
    6. Liang Y
    7. Aldape K
    8. Cha S
    9. Kuo MD

    (2008) Identification of noninvasive imaging surrogates for brain tumor gene-expression modules

    PNAS 105:5213–5218.

    https://doi.org/10.1073/pnas.0801279105

    • PubMed
    • Google Scholar
20. 1. Doroshow JH
    2. Kummar S

    (2014) Translational research in oncology--10 years of progress and future prospects

    Nature Reviews Clinical Oncology 11:649–662.

    https://doi.org/10.1038/nrclinonc.2014.158

    • PubMed
    • Google Scholar
21. 1. Fawcett T

    (2006) An introduction to ROC analysis

    Pattern Recognition Letters 27:861–874.

    https://doi.org/10.1016/j.patrec.2005.10.010

    • Google Scholar
22. 1. Fisher R
    2. Pusztai L
    3. Swanton C

    (2013) Cancer heterogeneity: implications for targeted therapeutics

    British Journal of Cancer 108:479–485.

    https://doi.org/10.1038/bjc.2012.581

    • PubMed
    • Google Scholar
23. 1. Fried DV
    2. Tucker SL
    3. Zhou S
    4. Liao Z
    5. Mawlawi O
    6. Ibbott G
    7. Court LE

    (2014) Prognostic value and reproducibility of pretreatment CT texture features in stage III non-small cell lung Cancer

    International Journal of Radiation Oncology*Biology*Physics 90:834–842.

    https://doi.org/10.1016/j.ijrobp.2014.07.020

    • PubMed
    • Google Scholar
24. 1. Ganeshan B
    2. Abaleke S
    3. Young RC
    4. Chatwin CR
    5. 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.

    https://doi.org/10.1102/1470-7330.2010.0021

    • PubMed
    • Google Scholar
25. 1. Ganeshan B
    2. Panayiotou E
    3. Burnand K
    4. Dizdarevic S
    5. Miles K

    (2012) Tumour heterogeneity in non-small cell lung carcinoma assessed by CT texture analysis: a potential marker of survival

    European Radiology 22:796–802.

    https://doi.org/10.1007/s00330-011-2319-8

    • PubMed
    • Google Scholar
26. 1. Ganeshan B
    2. Goh V
    3. Mandeville HC
    4. Ng QS
    5. Hoskin PJ
    6. Miles KA

    (2013) Non-small cell lung Cancer: histopathologic correlates for texture parameters at CT

    Radiology 266:326–336.

    https://doi.org/10.1148/radiol.12112428

    • PubMed
    • Google Scholar
27. 1. Gautier L
    2. Cope L
    3. Bolstad BM
    4. Irizarry RA

    (2004) Affy--analysis of Affymetrix GeneChip data at the probe level

    Bioinformatics 20:307–315.

    https://doi.org/10.1093/bioinformatics/btg405

    • PubMed
    • Google Scholar
28. 1. Gerlinger M
    2. Rowan AJ
    3. Horswell S
    4. Math M
    5. Larkin J
    6. Endesfelder D
    7. Gronroos E
    8. Martinez P
    9. Matthews N
    10. Stewart A
    11. Tarpey P
    12. Varela I
    13. Phillimore B
    14. Begum S
    15. McDonald NQ
    16. Butler A
    17. Jones D
    18. Raine K
    19. Latimer C
    20. Santos CR
    21. Nohadani M
    22. Eklund AC
    23. Spencer-Dene B
    24. Clark G
    25. Pickering L
    26. Stamp G
    27. Gore M
    28. Szallasi Z
    29. Downward J
    30. Futreal PA
    31. Swanton C

    (2012) Intratumor heterogeneity and branched evolution revealed by multiregion sequencing

    New England Journal of Medicine 366:883–892.

    https://doi.org/10.1056/NEJMoa1113205

    • PubMed
    • Google Scholar
29. 1. Gevaert O
    2. Xu J
    3. Hoang CD
    4. Leung AN
    5. Xu Y
    6. Quon A
    7. Rubin DL
    8. Napel S
    9. Plevritis SK

    (2012) Non-small cell lung Cancer: identifying prognostic imaging biomarkers by leveraging public gene expression microarray data--methods and preliminary results

    Radiology 264:387–396.

    https://doi.org/10.1148/radiol.12111607

    • PubMed
    • Google Scholar
30. 1. Gevaert O
    2. Mitchell LA
    3. Achrol AS
    4. Xu J
    5. Echegaray S
    6. Steinberg GK
    7. Cheshier SH
    8. Napel S
    9. Zaharchuk G
    10. Plevritis SK

    (2014) Glioblastoma multiforme: exploratory radiogenomic analysis by using quantitative image features

    Radiology 273:168–174.

    https://doi.org/10.1148/radiol.14131731

    • PubMed
    • Google Scholar
31. 1. Gillies RJ
    2. Anderson AR
    3. Gatenby RA
    4. Morse DL

    (2010) The biology underlying molecular imaging in oncology: from genome to anatome and back again

    Clinical Radiology 65:517–521.

    https://doi.org/10.1016/j.crad.2010.04.005

    • PubMed
    • Google Scholar
32. 1. Gillies RJ
    2. Kinahan PE
    3. Hricak H

    (2016) Radiomics: images are more than Pictures, they are data

    Radiology 278:563–577.

    https://doi.org/10.1148/radiol.2015151169

    • PubMed
    • Google Scholar
33. 1. Grossmann P
    2. Gutman DA
    3. Dunn WD
    4. Holder CA
    5. Aerts HJ

    (2016) Imaging-genomics reveals driving pathways of MRI derived volumetric tumor phenotype features in glioblastoma

    BMC Cancer 16:611.

    https://doi.org/10.1186/s12885-016-2659-5

    • PubMed
    • Google Scholar
34. 1. Grossmann P
    2. Narayan V
    3. Chang K
    4. Rahman R
    5. Abrey L
    6. Reardon DA
    7. Schwartz LH
    8. Wen PY
    9. Alexander BM
    10. Huang R
    11. Aerts H

    (2017) Quantitative Imaging biomarkers for risk stratification of patients with recurrent glioblastoma treated with Bevacizumab

    Neuro-Oncology.

    https://doi.org/10.1093/neuonc/nox092

    • PubMed
    • Google Scholar
35. 1. Grove O
    2. Berglund AE
    3. Schabath MB
    4. Aerts HJ
    5. Dekker A
    6. Wang H
    7. Velazquez ER
    8. Lambin P
    9. Gu Y
    10. Balagurunathan Y
    11. Eikman E
    12. Gatenby RA
    13. Eschrich S
    14. Gillies RJ

    (2015) Quantitative computed tomographic descriptors associate tumor shape complexity and intratumor heterogeneity with prognosis in lung adenocarcinoma

    PLoS One 10:e0118261.

    https://doi.org/10.1371/journal.pone.0118261

    • PubMed
    • Google Scholar
36. 1. Gutman DA
    2. Dunn WD
    3. Grossmann P
    4. Cooper LA
    5. Holder CA
    6. Ligon KL
    7. Alexander BM
    8. Aerts HJ

    (2015) Somatic mutations associated with MRI-derived volumetric features in glioblastoma

    Neuroradiology 57:1227–1237.

    https://doi.org/10.1007/s00234-015-1576-7

    • PubMed
    • Google Scholar
37. 1. Harrell FE
    2. Califf RM
    3. Pryor DB
    4. Lee KL
    5. Rosati RA

    (1982) Evaluating the yield of medical tests

    JAMA: The Journal of the American Medical Association 247:2543–2546.

    https://doi.org/10.1001/jama.1982.03320430047030

    • PubMed
    • Google Scholar
38. 1. Hou J
    2. Aerts J
    3. den Hamer B
    4. van Ijcken W
    5. den Bakker M
    6. Riegman P
    7. van der Leest C
    8. van der Spek P
    9. Foekens JA
    10. Hoogsteden HC
    11. Grosveld F
    12. Philipsen S

    (2010) Gene expression-based classification of non-small cell lung carcinomas and survival prediction

    PLoS One 5:e10312.

    https://doi.org/10.1371/journal.pone.0010312

    • PubMed
    • Google Scholar
39. 1. Hsu YC
    2. Yuan S
    3. Chen HY
    4. Yu SL
    5. Liu CH
    6. Hsu PY
    7. Wu G
    8. Lin CH
    9. Chang GC
    10. Li KC
    11. Yang PC

    (2009) A four-gene signature from NCI-60 cell line for survival prediction in non-small cell lung Cancer

    Clinical Cancer Research 15:7309–7315.

    https://doi.org/10.1158/1078-0432.CCR-09-1572

    • PubMed
    • Google Scholar
40. 1. Huang YQ
    2. Liang CH
    3. He L
    4. Tian J
    5. Liang CS
    6. Chen X
    7. Ma ZL
    8. Liu ZY

    (2016a) Development and Validation of a Radiomics Nomogram for Preoperative prediction of Lymph Node Metastasis in Colorectal Cancer

    Journal of Clinical Oncology 34:2157–2164.

    https://doi.org/10.1200/JCO.2015.65.9128

    • PubMed
    • Google Scholar
41. 1. Huang Y
    2. Liu Z
    3. He L
    4. Chen X
    5. Pan D
    6. Ma Z
    7. Liang C
    8. Tian J
    9. Liang C

    (2016b) Radiomics signature: a potential biomarker for the prediction of Disease-Free Survival in Early-Stage (I or II) Non-Small Cell lung Cancer

    Radiology 281:947–957.

    https://doi.org/10.1148/radiol.2016152234

    • PubMed
    • Google Scholar
42. 1. Hänzelmann S
    2. Castelo R
    3. Guinney J

    (2013) GSVA: gene set variation analysis for microarray and RNA-seq data

    BMC Bioinformatics 14:7.

    https://doi.org/10.1186/1471-2105-14-7

    • PubMed
    • Google Scholar
43. 1. Irizarry RA
    2. Hobbs B
    3. Collin F
    4. Beazer-Barclay YD
    5. Antonellis KJ
    6. Scherf U
    7. Speed TP

    (2003) Exploration, normalization, and summaries of high density oligonucleotide array probe level data

    Biostatistics 4:249–264.

    https://doi.org/10.1093/biostatistics/4.2.249

    • PubMed
    • Google Scholar
44. 1. Joshi-Tope G
    2. Gillespie M
    3. Vastrik I
    4. D'Eustachio P
    5. Schmidt E
    6. de Bono B
    7. Jassal B
    8. Gopinath GR
    9. Wu GR
    10. Matthews L
    11. Lewis S
    12. Birney E
    13. Stein L

    (2005) Reactome: a knowledgebase of biological pathways

    Nucleic Acids Research 33:D428–D432.

    https://doi.org/10.1093/nar/gki072

    • PubMed
    • Google Scholar
45. 1. Kobayashi S
    2. Yoneda-Kato N
    3. Itahara N
    4. Yoshida A
    5. Kato JY

    (2013) The COP1 E3-ligase interacts with FIP200, a key regulator of mammalian autophagy

    BMC Biochemistry 14:1.

    https://doi.org/10.1186/1471-2091-14-1

    • PubMed
    • Google Scholar
46. 1. Kumar V
    2. Gu Y
    3. Basu S
    4. Berglund A
    5. Eschrich SA
    6. Schabath MB
    7. Forster K
    8. Aerts HJ
    9. Dekker A
    10. Fenstermacher D
    11. Goldgof DB
    12. Hall LO
    13. Lambin P
    14. Balagurunathan Y
    15. Gatenby RA
    16. Gillies RJ

    (2012) Radiomics: the process and the challenges

    Magnetic Resonance Imaging 30:1234–1248.

    https://doi.org/10.1016/j.mri.2012.06.010

    • PubMed
    • Google Scholar
47. 1. Kuo MD
    2. Jamshidi N

    (2014) Behind the numbers: decoding molecular phenotypes with radiogenomics--guiding principles and technical considerations

    Radiology 270:320–325.

    https://doi.org/10.1148/radiol.13132195

    • PubMed
    • Google Scholar
48. 1. Lambin P
    2. Rios-Velazquez E
    3. Leijenaar R
    4. Carvalho S
    5. van Stiphout RG
    6. Granton P
    7. Zegers CM
    8. Gillies R
    9. Boellard R
    10. Dekker A
    11. Aerts HJ

    (2012) Radiomics: extracting more information from medical images using advanced feature analysis

    European Journal of Cancer 48:441–446.

    https://doi.org/10.1016/j.ejca.2011.11.036

    • PubMed
    • Google Scholar
49. 1. Leijenaar RT
    2. Carvalho S
    3. Velazquez ER
    4. van Elmpt WJ
    5. Parmar C
    6. Hoekstra OS
    7. Hoekstra CJ
    8. Boellaard R
    9. Dekker AL
    10. Gillies RJ
    11. Aerts HJ
    12. Lambin P

    (2013) Stability of FDG-PET radiomics features: an integrated analysis of test-retest and inter-observer variability

    Acta Oncologica 52:1391–1397.

    https://doi.org/10.3109/0284186X.2013.812798

    • PubMed
    • Google Scholar
50. 1. Li H
    2. Zhu Y
    3. Burnside ES
    4. Drukker K
    5. Hoadley KA
    6. Fan C
    7. Conzen SD
    8. Whitman GJ
    9. Sutton EJ
    10. Net JM
    11. Ganott M
    12. Huang E
    13. Morris EA
    14. Perou CM
    15. Ji Y
    16. Giger ML

    (2016) MR Imaging Radiomics signatures for predicting the risk of breast Cancer recurrence as given by Research Versions of MammaPrint, Oncotype DX, and PAM50 gene assays

    Radiology 281:382–391.

    https://doi.org/10.1148/radiol.2016152110

    • PubMed
    • Google Scholar
51. 1. Liberzon A
    2. Subramanian A
    3. Pinchback R
    4. Thorvaldsdóttir H
    5. Tamayo P
    6. Mesirov JP

    (2011) Molecular signatures database (MSigDB) 3.0

    Bioinformatics 27:1739–1740.

    https://doi.org/10.1093/bioinformatics/btr260

    • PubMed
    • Google Scholar
52. 1. Liu Y
    2. Kim J
    3. Balagurunathan Y
    4. Li Q
    5. Garcia AL
    6. Stringfield O
    7. Ye Z
    8. Gillies RJ

    (2016) Radiomic features are associated with EGFR mutation status in lung adenocarcinomas

    Clinical Lung Cancer 17:441–448.

    https://doi.org/10.1016/j.cllc.2016.02.001

    • PubMed
    • Google Scholar
53. 1. Mattonen SA
    2. Palma DA
    3. Johnson C
    4. Louie AV
    5. Landis M
    6. Rodrigues G
    7. Chan I
    8. Etemad-Rezai R
    9. Yeung TP
    10. Senan S
    11. Ward AD

    (2016) Detection of local Cancer recurrence after stereotactic ablative Radiation therapy for lung Cancer: physician Performance Versus Radiomic Assessment

    International Journal of Radiation Oncology*Biology*Physics 94:1121–1128.

    https://doi.org/10.1016/j.ijrobp.2015.12.369

    • PubMed
    • Google Scholar
54. 1. Menendez D
    2. Shatz M
    3. Resnick MA

    (2013) Interactions between the tumor suppressor p53 and immune responses

    Current Opinion in Oncology 25:85–92.

    https://doi.org/10.1097/CCO.0b013e32835b6386

    • PubMed
    • Google Scholar
55. 1. Mirsadraee S
    2. Oswal D
    3. Alizadeh Y
    4. Caulo A
    5. van Beek E

    (2012) The 7th lung Cancer TNM classification and staging system: review of the changes and implications

    World Journal of Radiology 4:.

    https://doi.org/10.4329/wjr.v4.i4.128

    • PubMed
    • Google Scholar
56. 1. Nicolasjilwan M
    2. Hu Y
    3. Yan C
    4. Meerzaman D
    5. Holder CA
    6. Gutman D
    7. Jain R
    8. Colen R
    9. Rubin DL
    10. Zinn PO
    11. Hwang SN
    12. Raghavan P
    13. Hammoud DA
    14. Scarpace LM
    15. Mikkelsen T
    16. Chen J
    17. Gevaert O
    18. Buetow K
    19. Freymann J
    20. Kirby J
    21. Flanders AE
    22. Wintermark M
    23. TCGA Glioma Phenotype Research Group

    (2015) Addition of MR imaging features and genetic biomarkers strengthens glioblastoma survival prediction in TCGA patients

    Journal of Neuroradiology 42:212–221.

    https://doi.org/10.1016/j.neurad.2014.02.006

    • PubMed
    • Google Scholar
57. 1. O'Connor JP
    2. Rose CJ
    3. Waterton JC
    4. Carano RA
    5. Parker GJ
    6. Jackson A
    7. O’Connor JPB

    (2015) Imaging intratumor heterogeneity: role in therapy response, resistance, and clinical outcome

    Clinical Cancer Research 21:249–257.

    https://doi.org/10.1158/1078-0432.CCR-14-0990

    • PubMed
    • Google Scholar
58. 1. O'Connor JP
    2. Aboagye EO
    3. Adams JE
    4. Aerts HJ
    5. Barrington SF
    6. Beer AJ
    7. Boellaard R
    8. Bohndiek SE
    9. Brady M
    10. Brown G
    11. Buckley DL
    12. Chenevert TL
    13. Clarke LP
    14. Collette S
    15. Cook GJ
    16. deSouza NM
    17. Dickson JC
    18. Dive C
    19. Evelhoch JL
    20. Faivre-Finn C
    21. Gallagher FA
    22. Gilbert FJ
    23. Gillies RJ
    24. Goh V
    25. Griffiths JR
    26. Groves AM
    27. Halligan S
    28. Harris AL
    29. Hawkes DJ
    30. Hoekstra OS
    31. Huang EP
    32. Hutton BF
    33. Jackson EF
    34. Jayson GC
    35. Jones A
    36. Koh DM
    37. Lacombe D
    38. Lambin P
    39. Lassau N
    40. Leach MO
    41. Lee TY
    42. Leen EL
    43. Lewis JS
    44. Liu Y
    45. Lythgoe MF
    46. Manoharan P
    47. Maxwell RJ
    48. Miles KA
    49. Morgan B
    50. Morris S
    51. Ng T
    52. Padhani AR
    53. Parker GJ
    54. Partridge M
    55. Pathak AP
    56. Peet AC
    57. Punwani S
    58. Reynolds AR
    59. Robinson SP
    60. Shankar LK
    61. Sharma RA
    62. Soloviev D
    63. Stroobants S
    64. Sullivan DC
    65. Taylor SA
    66. Tofts PS
    67. Tozer GM
    68. van Herk M
    69. Walker-Samuel S
    70. Wason J
    71. Williams KJ
    72. Workman P
    73. Yankeelov TE
    74. Brindle KM
    75. McShane LM
    76. Jackson A
    77. Waterton JC
    78. O’Connor JPB

    (2017) Imaging biomarker roadmap for Cancer studies

    Nature Reviews Clinical Oncology 14:169–186.

    https://doi.org/10.1038/nrclinonc.2016.162

    • PubMed
    • Google Scholar
59. 1. Pardoll DM

    (2012) The blockade of immune checkpoints in Cancer immunotherapy

    Nature Reviews Cancer 12:252–264.

    https://doi.org/10.1038/nrc3239

    • PubMed
    • Google Scholar
60. 1. Parmar C
    2. Leijenaar RT
    3. Grossmann P
    4. Rios Velazquez E
    5. Bussink J
    6. Rietveld D
    7. Rietbergen MM
    8. Haibe-Kains B
    9. Lambin P
    10. Aerts HJ

    (2015a) Radiomic feature clusters and prognostic signatures specific for lung and Head & Neck Cancer

    Scientific Reports 5:11044.

    https://doi.org/10.1038/srep11044

    • PubMed
    • Google Scholar
61. 1. Parmar C
    2. Grossmann P
    3. Rietveld D
    4. Rietbergen MM
    5. Lambin P
    6. Aerts HJ

    (2015b) Radiomic machine-learning Classifiers for prognostic biomarkers of head and neck cancer

    Frontiers in Oncology 5:272.

    https://doi.org/10.3389/fonc.2015.00272

    • PubMed
    • Google Scholar
62. 1. Pencina MJ
    2. D'Agostino RB

    (2004) Overall C as a measure of discrimination in survival analysis: model specific population value and confidence interval estimation

    Statistics in Medicine 23:2109–2123.

    https://doi.org/10.1002/sim.1802

    • PubMed
    • Google Scholar
63. 1. Postow MA
    2. Callahan MK
    3. Wolchok JD

    (2015) Immune checkpoint blockade in Cancer therapy

    Journal of Clinical Oncology 33:1974–1982.

    https://doi.org/10.1200/JCO.2014.59.4358

    • PubMed
    • Google Scholar
64. Software

    1. R Development Core Team

    (2013) R: A Language and Environment for Statistical Computing

    R Foundation for Statistical Computing, Austria.

    http://www.R-project.org
65. 1. Rabbani MG
    2. Hossain SA
    3. Islam KK
    4. Uddin SN

    (2014) Constitutive photomorphogensis Protein1 (COP1) mediated p53 pathway and its oncogenic role

    Biomedical Research and Therapy 1:142–151.

    https://doi.org/10.7603/s40730-014-0022-5

    • Google Scholar
66. 1. Rizzo S
    2. Petrella F
    3. Buscarino V
    4. De Maria F
    5. Raimondi S
    6. Barberis M
    7. Fumagalli C
    8. Spitaleri G
    9. Rampinelli C
    10. De Marinis F
    11. Spaggiari L
    12. Bellomi M

    (2016) CT Radiogenomic characterization of EGFR, K-RAS, and ALK mutations in non-small cell Lung cancer

    European Radiology 26:32–42.

    https://doi.org/10.1007/s00330-015-3814-0

    • PubMed
    • Google Scholar
67. 1. Rutman AM
    2. Kuo MD

    (2009) Radiogenomics: creating a link between molecular diagnostics and diagnostic imaging

    European Journal of Radiology 70:232–241.

    https://doi.org/10.1016/j.ejrad.2009.01.050

    • PubMed
    • Google Scholar
68. 1. Schröder MS
    2. Culhane AC
    3. Quackenbush J
    4. Haibe-Kains B

    (2011) Survcomp: an R/Bioconductor package for performance assessment and comparison of survival models

    Bioinformatics 27:3206–3208.

    https://doi.org/10.1093/bioinformatics/btr511

    • Google Scholar
69. 1. Segal E
    2. Sirlin CB
    3. Ooi C
    4. Adler AS
    5. Gollub J
    6. Chen X
    7. Chan BK
    8. Matcuk GR
    9. Barry CT
    10. Chang HY
    11. Kuo MD

    (2007) Decoding global gene expression programs in liver Cancer by noninvasive imaging

    Nature Biotechnology 25:675–680.

    https://doi.org/10.1038/nbt1306

    • PubMed
    • Google Scholar
70. 1. Sottoriva A
    2. Spiteri I
    3. Piccirillo SG
    4. Touloumis A
    5. Collins VP
    6. Marioni JC
    7. Curtis C
    8. Watts C
    9. Tavaré S

    (2013) Intratumor heterogeneity in human glioblastoma reflects Cancer evolutionary dynamics

    PNAS 110:4009–4014.

    https://doi.org/10.1073/pnas.1219747110

    • PubMed
    • Google Scholar
71. 1. Subramanian A
    2. Tamayo P
    3. Mootha VK
    4. Mukherjee S
    5. Ebert BL
    6. Gillette MA
    7. Paulovich A
    8. Pomeroy SL
    9. Golub TR
    10. Lander ES
    11. Mesirov JP

    (2005) Gene set enrichment analysis: a knowledge-based approach for interpreting genome-wide expression profiles

    PNAS 102:15545–15550.

    https://doi.org/10.1073/pnas.0506580102

    • PubMed
    • Google Scholar
72. Book

    1. Theodoridis S
    2. Koutroumbas K

    (2008)

    Pattern Recognition (Fourth Edition)

    Academic Press.

    • Google Scholar
73. 1. Whitlock MC

    (2005) Combining probability from independent tests: the weighted Z-method is superior to Fisher's approach

    Journal of Evolutionary Biology 18:1368–1373.

    https://doi.org/10.1111/j.1420-9101.2005.00917.x

    • PubMed
    • Google Scholar
74. 1. Win T
    2. Miles KA
    3. Janes SM
    4. Ganeshan B
    5. Shastry M
    6. Endozo R
    7. Meagher M
    8. Shortman RI
    9. Wan S
    10. Kayani I
    11. Ell PJ
    12. Groves AM

    (2013) Tumor heterogeneity and permeability as measured on the CT component of PET/CT predict survival in patients with non-small cell lung Cancer

    Clinical Cancer Research 19:3591–3599.

    https://doi.org/10.1158/1078-0432.CCR-12-1307

    • PubMed
    • Google Scholar
75. 1. Wolchok JD
    2. Saenger Y

    (2008) The mechanism of anti-CTLA-4 activity and the negative regulation of T-cell activation

    The Oncologist 13 Suppl 4:2–9.

    https://doi.org/10.1634/theoncologist.13-S4-2

    • PubMed
    • Google Scholar
76. 1. Yoon HJ
    2. Sohn I
    3. Cho JH
    4. Lee HY
    5. Kim JH
    6. Choi YL
    7. Kim H
    8. Lee G
    9. Lee KS
    10. Kim J

    (2015) Decoding tumor phenotypes for ALK, ROS1, and RET Fusions in lung adenocarcinoma using a Radiomics Approach

    Medicine 94:e1753.

    https://doi.org/10.1097/MD.0000000000001753

    • PubMed
    • Google Scholar
77. 1. Yuan BZ
    2. Jefferson AM
    3. Popescu NC
    4. Reynolds SH

    (2004) Aberrant gene expression in human non small cell lung carcinoma cells exposed to demethylating agent 5-aza-2'-deoxycytidine

    Neoplasia 6:412–419.

    https://doi.org/10.1593/neo.03490

    • PubMed
    • Google Scholar
78. 1. Zhao B
    2. Tan Y
    3. Tsai WY
    4. Qi J
    5. Xie C
    6. Lu L
    7. Schwartz LH

    (2016) Reproducibility of radiomics for deciphering tumor phenotype with imaging

    Scientific Reports 6:23428.

    https://doi.org/10.1038/srep23428

    • PubMed
    • Google Scholar

Author details

1. Patrick Grossmann

   1. Department of Radiation Oncology, Dana-Farber Cancer Institute, Brigham and Women’s Hospital, Harvard Medical School, Boston, United States
   2. Department of Biostatistics and Computational Biology, Dana-Farber Cancer Institute, Boston, United States

   Contribution

   PG, Conceptualization, Resources, Data curation, Software, Formal analysis, Validation, Investigation, Visualization, Methodology, Writing—original draft, Project administration, Writing—review and editing

   Competing interests

   No competing interests declared.

   "This ORCID iD identifies the author of this article:" 0000-0003-4918-6902

2. Olya Stringfield

   Department of Cancer Imaging and Metabolism, H. Lee Moffitt Cancer Center and Research Institute, Tampa, United States

   Contribution

   OS, Resources, Data curation, Supervision, Investigation, Writing—original draft, Project administration, Writing—review and editing

   Competing interests

   No competing interests declared.

3. Nehme El-Hachem

   Integrative systems biology, Institut de recherches cliniques de Montreal, Montreal, Canada.

   Contribution

   NE-H, Formal analysis, Investigation, Methodology, Writing—review and editing

   Competing interests

   No competing interests declared.

4. Marilyn M Bui

   Department of Anatomic Pathology, H. Lee Moffitt Cancer Center and Research Institute, Tampa, United States

   Contribution

   MMB, Resources, Data curation, Software, Formal analysis, Validation, Investigation, Visualization, Methodology, Writing—review and editing

   Competing interests

   No competing interests declared.

5. Emmanuel Rios Velazquez

   Department of Radiation Oncology, Dana-Farber Cancer Institute, Brigham and Women’s Hospital, Harvard Medical School, Boston, United States

   Contribution

   ERV, Resources, Software, Writing—review and editing

   Competing interests

   No competing interests declared.

6. Chintan Parmar

   1. Department of Radiation Oncology, Dana-Farber Cancer Institute, Brigham and Women’s Hospital, Harvard Medical School, Boston, United States
   2. Department of Radiation Oncology, Research Institute GROW, Maastricht University, Maastricht, Netherlands

   Contribution

   CP, Resources, Software, Writing—review and editing

   Competing interests

   No competing interests declared.

7. Ralph TH Leijenaar

   Department of Radiation Oncology, Research Institute GROW, Maastricht University, Maastricht, Netherlands

   Contribution

   RTHL, Resources, Software, Writing—review and editing

   Competing interests

   No competing interests declared.

8. Benjamin Haibe-Kains

   1. Princess Margaret Cancer Centre, University Health Network, University of Toronto, Toronto, Canada
   2. Medical Biophysics Department, University of Toronto, Toronto, Canada

   Contribution

   BH-K, Conceptualization, Software, Formal analysis, Supervision, Writing—review and editing

   Competing interests

   No competing interests declared.

9. Philippe Lambin

   Department of Radiation Oncology, Research Institute GROW, Maastricht University, Maastricht, Netherlands

   Contribution

   PL, Resources, Data curation, Software, Funding acquisition, Project administration, Writing—review and editing

   Competing interests

   No competing interests declared.

10. Robert J Gillies

    Department of Cancer Imaging and Metabolism, H. Lee Moffitt Cancer Center and Research Institute, Tampa, United States

    Contribution

    RJG, Conceptualization, Resources, Data curation, Supervision, Funding acquisition, Investigation, Writing—original draft, Project administration, Writing—review and editing

    Competing interests

    RJG: declares a collaboration with HealthMyne.

    "This ORCID iD identifies the author of this article:" 0000-0002-8888-7747

11. Hugo JWL Aerts

    1. Department of Radiation Oncology, Dana-Farber Cancer Institute, Brigham and Women’s Hospital, Harvard Medical School, Boston, United States
    2. Department of Biostatistics and Computational Biology, Dana-Farber Cancer Institute, Boston, United States
    3. Department of Radiology, Brigham and Women’s Hospital, Harvard Medical School, Boston, United States

    Contribution

    HJWLA, Conceptualization, Resources, Data curation, Software, Supervision, Funding acquisition, Investigation, Methodology, Writing—original draft, Project administration, Writing—review and editing

    For correspondence

    hugo_aerts@dfci.harvard.edu

    Competing interests

    No competing interests declared.

    "This ORCID iD identifies the author of this article:" 0000-0002-2122-2003

• 6,590

  views

• 1,275

  downloads

• 276

  citations

Views, downloads and citations are aggregated across all versions of this paper published by eLife.