Purpose The objective of this study was to build and validate a radiomic signature to predict early a poor outcome using baseline and 2-month evaluation CT and to compare it to the RECIST1·1 and morphological criteria defined by changes in homogeneity and borders.
Methods This study is an ancillary study from the PRODIGE-9 multicentre prospective study for which 491 patients with metastatic colorectal cancer (mCRC) treated by 5-fluorouracil, leucovorin and irinotecan (FOLFIRI) and bevacizumab had been analysed. In 230 patients, computed texture analysis was performed on the dominant liver lesion (DLL) at baseline and 2 months after chemotherapy. RECIST1·1 evaluation was performed at 6 months. A radiomic signature (Survival PrEdiction in patients treated by FOLFIRI and bevacizumab for mCRC using contrast-enhanced CT TextuRe Analysis (SPECTRA) Score) combining the significant predictive features was built using multivariable Cox analysis in 120 patients, then locked, and validated in 110 patients. Overall survival (OS) was estimated with the Kaplan-Meier method and compared between groups with the logrank test. An external validation was performed in another cohort of 40 patients from the PRODIGE 20 Trial.
Results In the training cohort, the significant predictive features for OS were: decrease in sum of the target liver lesions (STL), (adjusted hasard-ratio(aHR)=13·7, p=1·93×10–7), decrease in kurtosis (ssf=4) (aHR=1·08, p=0·001) and high baseline density of DLL, (aHR=0·98, p<0·001). Patients with a SPECTRA Score >0·02 had a lower OS in the training cohort (p<0·0001), in the validation cohort (p<0·0008) and in the external validation cohort (p=0·0027). SPECTRA Score at 2 months had the same prognostic value as RECIST at 6 months, while non-response according to RECIST1·1 at 2 months was not associated with a lower OS in the validation cohort (p=0·238). Morphological response was not associated with OS (p=0·41).
Conclusion A radiomic signature (combining decrease in STL, density and computed texture analysis of the DLL) at baseline and 2-month CT was able to predict OS, and identify good responders better than RECIST1.1 criteria in patients with mCRC treated by FOLFIRI and bevacizumab as a first-line treatment. This tool should now be validated by further prospective studies.
- clinical decision making
- colorectal cancer
- colorectal metastases
- computerised image analysis
Statistics from Altmetric.com
- clinical decision making
- colorectal cancer
- colorectal metastases
- computerised image analysis
Significance of this study
What is already known on this subject?
Assessment of tumour response to chemotherapy is based on size criteria.
Evaluation of tumour response to antivascular endothelial growth factor therapies is limited because changes in the tumour architecture occur earlier than the shrinkage of lesions.
New morphological criteria including homogeneity, borders and density of liver metastases have been proposed for response assessment to antiangiogenic therapies.
What are the new findings?
The proposed radiomic nomogram combining simple computed analysis, baseline density and size criteria is able to predict a poor outcome at 2 months with the same performance as standard evaluation with RECIST1·1 at 6 months.
A simplified score, using only baseline density and decrease in sum of the target liver lesions, showed feasibility to identify early good responders at 6 months.
There was no difference in overall survival between patients with optimal morphological response and those with no response in the present study.
Significance of this study
How might it impact on clinical practice in the foreseeable future?
The Survival PrEdiction in patients treated by 5-fluorouracil, leucovorin and irinotecan (FOLFIRI) and bevacizumab for mCRC using contrast-enhanced CT TextuRe Analysis (SPECTRA) Score was able to predict poor overall survival at the 2-month evaluation CT, earlier than classical RECIST1·1 criteria, allowing for timely go/no-go treatment decisions.
Optimal morphological response should not be considered as able to predict better outcome.
Colorectal cancer (CRC) is the third most common cancer worldwide and one of the leading causes of cancer-related deaths.1 2 Approximately 25% of patients with CRC present with metastases at initial diagnosis and a majority of these patients are not eligible for curative strategy. Treatment of patients with metastatic CRC (mCRC) has undergone major improvements in recent years. First-line treatment for these patients is based on the combination of 5-fluorouracil, leucovorin and irinotecan (FOLFIRI) or oxaliplatin.3 4 The use of targeted therapies such as the antivascular endothelial growth factor (anti-VEGF) (bevacizumab) and antiepidermal growth factor receptor monoclonal antibodies (cetuximab or panitumumab) has resulted in increased survival compared with chemotherapy alone.5 6
Assessment of response to chemotherapy is based on size criteria, the the most common being Response Evaluation Criteria in Solid Tumors (RECIST) 1·1 criteria (.7 However, the evaluation of response to anti-VEGF therapies is limited because changes in the tumour architecture occur earlier than the shrinkage of lesions, which is observed with classical cytotoxic agents.8 9 In 2009, a large study proposed new morphological criteria for response assessment to antiangiogenic therapies in a presurgical setting and validated them in a non-surgical patient cohort.10 In this study, Chun et al concluded that optimal response is obtained when tumours are homogeneous, hypoattenuating and have sharp borders.10 These morphological features were associated with good pathological response.11 However, these morphological criteria are limited because they have only been validated in retrospective studies, and in candidates to surgery receiving chemotherapeutic regimens with or without bevacizumab.12 13 Furthermore, these results have not been incorporated into international guidelines because their usefulness and reproducibility has not been validated in large prospective clinical trials. More recently, depth in response, which defines the nadir of tumour response and early tumour shrinkage (defined as a reduction of at least 20% in tumour size at first reassessment) have been identified as new prognostic biomarkers of survival.12
Our main hypothesis is that reliable quantification of tumour architectural changes may help identify patients with non-responding tumours early, and, as such, allow a personalised treatment strategy with early go or no-go decisions. Recently, several computed analysis algorithms of tumour architecture have shown promising results as biomarkers of response and survival in patients with cancer.14–18 This new paradigm in medical imaging, called radiomics, has gained interest with the development of several tools that are able to extract quantitative features from imaging data that can potentially be used to predict patient outcome.19 20
Other imaging features have been proposed as surrogate biomarkers of response and survival in patients treated by antiangiogenic therapies, including perfusion CT,21 contrast-enhanced ultrasound,22 and diffusion-perfusion MRI.23 However, these techniques have failed widespread implementation primarily because of lack of reproducibility. The ideal scenario would be to develop surrogate markers of response that are reproducible and available using standard imaging techniques currently recommended by international guidelines for the follow-up of these patients, such as the classical portal-venous phase CT.
The aim of this ancillary study was to build and validate a score combining size and radiomic criteria in order to predict early the overall survival (OS) and absence of tumour response to FOLFIRI and bevacizumab using CT evaluation at baseline and 2 months, and to compare this score to the RECIST1·1 and morphological criteria proposed by Chun et al.10
Patients and methods
The SPECTRA Study (Survival PrEdiction in patients treated by FOLFIRI and bevacizumab for mCRC using contrast-enhanced CT TextuRe Analysis) is an ancillary analysis of the PRODIGE 9 Study (Partenariat de Recherche en Oncologie DIGEtive) promoted by Fédération Francophone de Cancérologie Digestive (Clinicaltrial.gov identifier: NCT00952029). More than 2 years of follow-up was available for all patients (online supplementary material).24 25 During the initial phase, all patients received first-line treatment with FOLFIRI+bevacizumab for 6 months. Following this initial treatment period, patients were treated with bevacizumab maintenance therapy (Arm A) or observation without further therapy (Arm B) until progression according to randomisation. As CT readings for the primary study were not centralised, CT images were not systematically collected and therefore were not available for all patients at baseline and 2 months (figure 1).
Supplementary file 1
Inclusion criteria for the current SPECTRA Study were as follows (figure 1): (1) Patients with baseline and 2-month CT examinations acquired during the portal-venous phase. (2) CT images were obtained with 3–5 mm slice thickness. (3) Presence of at least one target liver lesion larger than 1 cm. Exclusion criteria were as follows: (1) Unenhanced CT or CT acquired with inadequate timing of iodinated contrast administration. (2) Technical issues regarding Digital Imaging and Communications in Medicine (DICOM) images impairing computed analysis. As an external validation cohort, we performed an ancillary analysis of a randomised phase II trial in patients with mCRC over 75 years treated with first-line chemotherapy with or without bevacizumab (PRODIGE 20 Study).26
Two abdominal radiologists with 7 years (AD) and 22 years (CH) of experience blinded to all clinical data reviewed the baseline CT examinations and those obtained at 2 months. After these independent reviews, discordant readings were resolved in consensus during a third session.
All patients were categorised as optimal responders or non-optimal responders according to the morphological criteria of Chun et al.10 Group 1, homogeneous low attenuation with sharp tumour-liver interface; group 3, heterogeneous attenuation with a thick poorly defined tumour-liver interface; and group 2, intermediate morphology that cannot be rated as group 1 or 3. Morphological response was considered as optimal when metastasis changed from group 2 or group 3 to group 1 as previously published (online supplementary figure 2).10
Supplementary file 2
Contouring of the dominant liver lesion (DLL) in the Arm B cohort was performed during the consensus reading. To assess the reproducibility of quantitative evaluation using a single slice, a third independent abdominal radiologist with 25 years of experience (BG) contoured the DLLs in the Arm A cohort. The response at 2 months was assessed according to RECIST 1.1 and the sum of target liver lesions (STL) at baseline and 2 months was determined for all patients. Response was defined as complete if no target lesion was visible, partial for STL decrease >30%, progression for STL increase >20% and stable disease otherwise.7 Another analysis was performed according to modified RECIST criteria (mRECIST15) with response defined as a decrease in STL>15%. The density of the DLL from among the two target lesions was measured at baseline and at 2 months using manual contouring of the whole tumour on the CT image on which it was the largest.
CT images at baseline and at 2 months were processed using a commercially available, texture computed analysis software programme (TexRAD, University of Sussex. Falmer, Sussex, England)(supplemental data and online supplementary figure 1). This software is integrated in the radiological workflow with any picture archiving and communication system (PACS).
Supplementary file 3
OS was the primary end point, defined as the time from randomisation to death (or last news if alive). As both arms did not have the same therapeutic regimen after the first 6 months and were randomised, they were considered as independent sets. The training set used data from Arm B, because it was the set with the largest number of patients, selecting out the strongest radiomic features, and fixing the weights to build the radiomic signature. The model was locked. Then, we tested only this unique radiomic signature on the data from Arm A as an independent validation data set, contoured by an independent reviewer. An external validation was performed with patients from the PRODIGE 20 Trial. Continuous variables were compared using Student’s t-test or Wilcoxon signed-rank test and categorical variables, expressed as numbers and percentages, were compared using χ2test or Fisher’s exact test. Absolute change was calculated for each feature defined as: value at 2 months minus baseline value. Features values were not centred or scaled as they were directly extracted from DICOM images. Because some histogram features are highly correlated, and in order to maintain the parsimony of the model, we excluded highly correlated features (r>0·6) and only included the significant predictive features after Bonferroni correction into the radiomic signature. Univariate Cox regression was used to estimate HRs and their 95% CIs for OS. To consider confounders of survival analysis, a multivariable analysis was performed by using a Cox proportional hazards model with forward stepwise selection of covariates and with entering and removing limits of p<0·05 and p>0·0025. The assumption of risk proportionality and log linearity were verified for each variable. Correlations between all variables were examined. Predictive features that were significantly associated with OS after Bonferroni correction were incorporated in a Cox proportional hazards regression model. A combined radiomic and size score (SPECTRA Score) was built by linear combination of each selected feature that was weighted by its respective coefficient. The performance of the model was assessed by internal validation in the training cohort with 1000 times bootstrap resampling to compute the optimism-corrected estimation of the model concordance index. Calibration of the radiomics nomogram was assessed using calibration plots.
A threshold was identified for the SPECTRA Score, decrease in STL and baseline mean density by calculating the median in 1000 times bootstrap samples. Comparisons between groups (SPECTRA Score, morphological criteria and RECIST1·1 response at 2 months and 6 months) were performed with the logrank test. κ statistics were used to determine interobserver agreement between the two readers regarding morphological criteria. All statistical tests were two-sided, and significance was set at p<0·05 or p<0·0025 for feature selection according to Bonferroni correction. Statistical analysis was performed using SAS V.9·4 and R V.3·4·0 (R Foundation, http://www.r-project.org/) using the ‘rms’ and ‘caret’ packages.
Comparison of the SPECTRA cohort with the PRODIGE 9 Study cohort
Each arm of the final cohort of the SPECTRA Study was not significantly different from the initial randomised cohort from the PRODIGE 9 Study with respect to age, gender, weight, performance status, Kohne criteria, KRAS and BRAF mutation status, resection of primary tumour, location of primary tumour, and number of metastatic sites (online supplementary table 1). Characteristics of SPECTRA study patients were not statistically different between both arms (online supplementary table 2). Our final cohorts included 110 patients in Arm A (78 men, 32 women; mean age 63·97±10·81 (SD); range 26–83 years) and 120 patients in Arm B (81 men, 39 women; mean age 63·66±11·01 (SD); range 35–84 years). Two hundred and sixty-one patients were excluded because patients were never treated (n=3), baseline and/or first evaluation CT examinations were not available for central review (n=158), no target liver lesions were identified during the central review (n=59), CT examinations were unenhanced or acquired with inadequate timing of contrast administration (n=20), patients underwent their follow-up using Positron - emission tomography (PET)/CT or MRI (n=14), or technical issues impaired computed analysis (n=7). The 6-month response according to RECIST1·1 criteria was available for all patients except for one in Arm A and three in Arm B (figure 1). The median number of lesions was 13 in the training cohort and 14 in the validation cohort. In the external validation cohort 40 patients out of the 102 patients included met the inclusion criteria (25 men, 15 women; mean age 81·79±4·00 (SD); range 75–91 years).
Selection of radiomic features and construction of the SPECTRA Score in the training cohort (Arm B)
Among the 114 candidate features, 20 features were kept for further analyses. Univariate analysis was then performed on this 20-feature subset with the logrank test. Considering the 20 tests performed, a p value <0·0025 was considered statistically significant (Bonferroni correction). At multivariable analysis, three features were retained in the model: the decrease in STL (adjusted HR=13·7, 95% CI 5·2 to 36·3, p=1·93×10-7), the density of the dominant lesion on baseline CT (adjusted HR=0·98, 95% CI 0·97 to 0·99, p=0·0004) and the absolute change in kurtosis of pixels in the region of interest with a medium spatial scale filtering (spatial scale filter(ssf)=4) (adjusted HR=1·08, 95% CI 1·03 to 1·13, p=0·0013).
In order to ensure easy use of this score, we presented the results as a nomogram. We built a radiomics nomogram on the basis of multivariable Cox analysis in the training cohort (figure 2). The corrected concordance index of the nomogram was 0·75 (95% CI 0·62 to 0·86). The calibration plot of the radiomics nomogram for the OS demonstrated good agreement between prediction and observation in the training and the validation cohorts (online supplementary figure 3).
OS using the SPECTRA Score in the training and validation cohorts
The median of SPECTRA Score across 1000 bootstrap resampling was 0.02. Figure 3A shows the OS of patients with a SPECTRA Score >0·02 versus patients with a SPECTRA Score ≤0·02 in the Arm B (training cohort) (p<0·0001). Figure 3B shows the OS of patients with a SPECTRA Score >0·02 versus that of patients with a SPECTRA Score ≤0·02 in Arm A (validation cohort) (p<0·0008). Figure 3E shows the OS of patients with a SPECTRA Score >0·02 versus OS of patients with a SPECTRA Score ≤0·02 in the external validation cohort (PRODIGE 20 cohort) (p=0·027).
Interobserver agreement for morphological criteria
The interobserver agreement between the two radiologists for scoring morphological criteria was substantial: κ=0·62 (95% CI 0·51 to 0·73). Discrepancies in scoring morphological criteria occurred in 35/230 patients (15·2%), and were resolved during consensus review.
Comparison of SPECTRA Score to RECIST and morphological criteria
Table 1 shows the prediction of OS achieved using the SPECTRA Score as well as the comparison of survival between non-responder and responder patients according to morphological criteria at 2 months, and RECIST1.1 at 2 months and 6 months (decrease in STL>30%). There was no difference in OS between patients with optimal morphological response and those with no response, both in the training cohort (p=0.37) and in the validation cohort (p=0.87). Online supplementary figure 4 shows Kaplan-Meier curves of OS according to morphological response in the 230 patients (p=0.41).
Supplementary file 4
Among 61 patients with a SPECTRA Score ≤0·02 in the training cohort, 40 showed partial response (40/61; 65·6%), 20 were stable (32·8%) and one progressed (1·6%) at 6 months. Among 65 patients with a SPECTRA Score ≤0·02 in the validation cohort, 41/65 showed partial or complete response (63·1%), 19/65 were stable (29·2%) and 4/65 progressed (6·2%) at 6 months (6 months RECIST was unavailable for one patient, 1·5%). These differences were not statistically significant (p=0·14). Among patients who had response according to RECIST1·1 at 6 months (n=101), only 68 (67·3%) had a 20% response at 6 months while 81 (80·2%) had a good response according to SPECTRA Score (p=0·038).
Poor and good response as a predictor of OS was also assessed using RECIST1·1 and mRECIST15 at 2 months. Results were significant in the training cohort but only mRECIST15 was significant in the validation cohort (p=0·238 and p=0·039, respectively). In addition, mRECIST15 did not allow discriminating between patients with poor survival and those with improved survival at 2 months in the external validation cohort (PRODIGE 20 cohort) (p=0.93). Conversely, the SPECTRA Score was able to stratify survival outcomes at 2 months with the same accuracy as RECIST1·1 at 6 months in both cohorts. The Concordance Index of the nomogram was 0·79 while the Concordance Index of mRECIST15 alone at 2 months was 0·70 and was significantly lower (p=0·009).
Comparison of a simplified evaluation without computed analysis and response at 6 months according to RECIST 1.1
Among the whole cohort of 230 patients, 70/230 patients (30·4%) had a 15% decrease in STL (mRECIST15) at 2 months and a baseline density of the DLL >63 HU (median value obtained from 1000 bootstrap resampling). In this subgroup, 1 patient out of 70 (1·4%) showed progression at 6 months according to RECIST1·1.
In the 63 patients who had a baseline lesion density <63 HU and a decrease in STL<15%, 10/63 (15·9%) showed response at 6 months according to RECIST1·1. Of note, there were 101 responders in the whole study population. Patients who had a 15% decrease in STL at 2 months and a baseline density of the DLL >63 HU had increased OS compared with patients with a 2-month STL decrease <15% and a baseline lesion density <63 HU (median OS 2.63 (95% CI 2.10 to 3.16) vs 1·28 years (95% CI 1.16 to 1.41), p<0.001).
Our study shows that, in patients with mCRC, the use of a radiomic nomogram ‘SPECTRA-score’, combining simple computed analysis, baseline density and size criteria allows prediction of poor OS earlier than the RECIST1·1 criteria alone, in patients with mCRC. A first CT-based evaluation at 2 months with the SPECTRA Score is able to assess OS and response to FOLFIRI+bevacizumab with performances equivalent to those obtained with the classical RECIST1·1 evaluation at 6 months. The SPECTRA Score was validated in an independent cohort and in an external independent cohort from another multicentre prospective trial. Our results suggest that patients with a SPECTRA Score >0·02 at the 2-month CT evaluation may benefit from an early switch in treatment towards other chemotherapy or targeted therapies. Moreover, applying a simplified score, using only baseline density and decrease in STL, showed feasibility to early identify patients who are at low risk of progression at 6 months.
Recent studies have already suggested that size-based criteria such as RECIST1·1 have limited potential to assess tumour response to targeted therapies, particularly anti-VEGF therapies. As such, a number of morphological criteria have thus been proposed to overcome these limitations including changes in density, heterogeneity and thickness of the borders of liver lesions that better correlate with disease-free survival and OS in patients who have undergone surgical resection of liver metastases.10 13 However, these new criteria have not been validated in non-surgical candidates and cannot represent a surrogate biomarker of OS. Moreover, these criteria are qualitative variables and may thus be prone to interobserver variability. On the contrary, computed analysis is likely able to quantify changes in tumour heterogeneity with more accuracy and reproducibility than the radiologist’s naked eyes. In our study, the non-size-based morphological criteria at 2 months did not correlate with outcome.10 This discordance between our study and that of Chun et al may have two reasons.10 First, we assume that morphological criteria are probably more suitable for response assessment in surgical candidates, as these have fewer liver metastases than non-surgical patients do. Second, neither the chemotherapeutic regimens nor the time between the start of chemotherapy and CT were standardised in previous retrospective studies.10 13 Our results also showed that interobserver agreement for the assessment of morphological criteria was not perfect; this result advocates the use of standardised quantitative features such as radiomic. Automatic quantification of volume, density and other features are already available in all radiological evaluation systems and included in almost all PACS systems. Thus, radiological workflow, with the development of new ‘artificial intelligence’ tools, is ready for the use of radiomic signatures such as SPECTRA.
The SPECTRA Score combines one size criterion (ie, decrease in STL) and two new non-size-based criteria. Depth of response has been identified as a surrogate marker of survival for different types of primary tumours although decrease in size is delayed, especially in patients treated with bevacizumab.9 In the present study, the most discriminating cut-off value for decrease in STL at 2-month CT was a decrease of 15% in target lesions in the present study. This criterion alone was more suitable to predict poor OS than RECIST1·1 in both cohorts. This finding confirms that size criteria remain a powerful tool that should be included in any new model development. However, the Concordance Index of mRECIST15 alone at 2 months was significantly lower than that of SPECTRA (p=0.009). Moreover, in the external validation cohort, the SPECTRA Score was definitively superior to mRECIST15. Baseline density of DLL was also incorporated in the SPECTRA Score as a good predictor of OS. Our findings are consistent with those of Choi et al who combined a 10% threshold for decrease in STL size and a 15% decrease in tumour density at 2 months in patients with GI stromal tumours treated by targeted therapy.8 These findings suggest that tumours with high degrees of enhancement (ie,>63 HU) are most likely to respond to antiangiogenic treatments, while poorly enhancing tumours, probably containing necrotic tissue, are not.27 Recent studies have highlighted the different proliferation growth patterns in liver metastases.28 29 Metastases presenting with replacement growth pattern, where cancer cells infiltrate the hepatic plates of the liver parenchyma, have been shown to respond poorly to antiangiogenic treatments.28 We assume that baseline features of tumours on CT may correlate with the growth pattern and biology of the tumour, and that quantitative computed analysis is able to quantify the complexity of the different patterns of enhancement inside the tumour.19 30 We thus incorporated the absolute decrease in kurtosis in the SPECTRA Score. Indeed, kurtosis reflects the changes in the enhancement pattern of the tumour as well as tumorous changes in homogeneity and border definition to some degrees.31 Regions of high kurtosis are consistent with a larger distribution of tissue types, while a decrease in kurtosis reflects the tendency of the tumour to become more homogeneous and with less neovascularisation.32 Such a decrease in tumorous heterogeneity has been shown to correlate with pathological response and survival in previous studies.11 28
The major interest of our study is that our cohort of patients is homogeneous in terms of treatment during the first 6 months. As chemotherapies used in patients with mCRC usually have the same efficacy although toxicity profiles may be different, we assume that changes in imaging features related to cytotoxic agents rather than to antiangiogenic agents should be similar.
Moreover the reproducibility of the SPECTRA Score in the external validation cohort in patients that have received different types of chemotherapy suggest that the SPECTRA Score may also be effectively treated with other front-line chemotherapies than FOLFIRI+bevacizumab. Our study has several limitations. First, we were not able to include all patients randomised in the PRODIGE 9 Study. However, the SPECTRA Study arm remained similar to the PRODIGE 9 arm in terms of clinical, biological and oncological criteria. Second, CT examinations were performed in different institutions using different imaging parameters. However, we assumed that building a score validated on multicentrally acquired portal venous phase CT examinations would give the SPECTRA Score greater reproducibility, robustness and potentially enable its wide diffusion. Third, contouring of the lesions was not performed independently by the two radiologists in the training cohort and was manual. However, contouring was performed by an independent radiologist in the validation cohort to ensure reproducibility and robustness of the SPECTRA Score. Moreover, we made the choice to roughly include the interface between normal liver parenchyma and the metastasis with the ulterior motive of widespread applicability and routine use of the technique and of the score. We indeed assumed that the border of the lesions would sometimes be difficult to determine with precision and reproducibility if it implied meticulous and cumbersome contouring. Fourth, we only included three features in the SPECTRA Score. Some additional texture features may have further contributed to improve the model, but, as we said, we tried to keep our model parsimonious to avoid problems of overfitting.33 The final limitation is that patients did not have the same treatment, depending on the arm to which they pertained. However, initial treatment was the same in all patients. Moreover, we decided to consider both arms as independent cohorts in order to construct and validate the score on two different homogeneous cohorts, each cohort including only patients that had exactly the same therapeutic protocol. In conclusion, our study is the first to propose a radiomic signature for predicting OS in non-surgical candidates with mCRC treated by FOLFIRI and bevacizumab, and to validate it in a large prospective study. A radiomic signature (combining decrease in STL, density and computed texture analysis of the DLL) at baseline and 2-month CT was able to predict OS at the 2-month evaluation CT, and identify very good responders better than the more classical RECIST1.1 criteria in patients with mCRC treated by FOLFIRI and bevacizumab as a first-line treatment. This tool should now be validated by further prospective studies. Computed analysis with its ability to quantify changes in tumour morphology will soon play a pivotal role in the prediction and evaluation of treatments in oncology.
Supplementary file 5
The authors thank all participating patients and their families, and the study groups and investigators from the participating countries (appendix). The authors also thank the team from the FFCD data centre. The FFCD also thanks the Ligue Nationale Contre le Cancer for its grant. AD thanks the Société Française de Radiologie for its support.
TA and CH contributed equally.
Contributors AD is the guarantor of the present work. AD designed the study, collected the data and drafted the manuscript. BG, CR, PS supervised the study and drafted the manuscript. BG, VB, JT, OB collected the data and drafted the manuscript. KLM conducted the statistical analysis and drafted the manuscript. JB, FG, EF, J-MP, CB, RF, J-FS, FK-A, DG, JLJ, YR, SJ, MBA, FD, PT, ES, CL collected the data, and edited and approved the manuscript. EB conducted the statistical analysis. TA collected the data, conducted the original studies, supervised the present study and drafted the manuscript. CH supervised the present study and drafted the manuscript.
Funding Fédération Francophone de Cancérologie Digestive, Ligue Nationale Contre le Cancer, ROCHE. The study was funded by Fédération Francophone de Cancérologie Digestive (FFCD). The Liguecontre le Cancer provided financial support to FFCD. ROCHE provided financial support for study management.
Competing interests VB reports consultingor/and advisory boards for Merck Serono, Ipsen, Prestizia, Eisai, and Bayer. DrBoige has received honoraria for Merck Serono, Sanofi, Amgen, MSD and Bayer. CL reports personal grants for boardmembership for AAA, grants from Novartis and travel grants from Ipsen Pharma,Amgen and Bayer. OB reports personalgrants for consultancy for A mgen, Merck, Roche, Bayer; payment forlectures including service onspeakers bureau for Servier, Pierre Fabre ; and travel grants from Roche,Lilly. CB reports consultingfor BMS ; payment for lectures for Merck Serono et BMS ; Paymentsfor development of educational presentations for Aztra-Zenecaand BMS; travel grants from Merck Serono and Aztra-Zeneca; Advisory Boardmembership for BMS and Aztra-Zeneca. EF reports personal grants for boardmembership for Roche, Merck and Sanofi. JB reports personal grants for boardmembership for Roche, Boehringer-Ingelheim, Astra-Zeneca, Servier and BMS;payment for lectures including service on speaker bureau for Roche,Astra-Zeneca and travel grants from Roche and BMS. JT reports personal grantsfor consultancy for Abbvie, Amgen, Baxalta, Celgene, Lilly, Merck, Roche,Servier, MSD, Pierre-Fabre and Sanofi; payment for lectures including serviceon speakers bureau for ABBVIE, Amgen, Celgene, Lilly, Merck, Roche; paymentsfor development of educational presentations for Roche and travel grants fromRoche and Merck. J-MP reports personal grants for board membership for Roche,Sanofi, Lilly Merck and Amgen; consultancy for Lilly and Roche; payments fordevelopment of educational presentations for Roche, Sanofi, Lilly Merck andAmgen; grants for his institution for board membership for Roche and Merck ;payments of grants form Merck and Roche. J-FS reports personal grants forconsultancy for Celgene, Lilly, Merck, Novartis Oncology, Pfizer, Sanofi,Roche; grants from Roche; payments for development of educational presentationsfor Amgen and Lilly; travel grants from Ipsen Pharma and Merck. F-KAreports consultancy for Sanofi as Board member and travelgrants from Roche, Bayer and Ipsen. M-BA reports consulting and travel supportfrom Amgen, Bayer, Merck, Roche, Sanofi, and Ipsen. T A reports personal grants Consultancy for Pierre Fabre; Grantsfrom Roche and Amgen; Payments for development of educational presentations forNovartis Oncology, Pfizer, Sanofi, Roche; Travel grants from Ipsen Pharma,Novartis Oncology, Sanofi, Roche.
Provenance and peer review Not commissioned; externally peer reviewed.
Collaborators Dr Julien Volet (Centre Hospitalier Universitaire Robert Debré, Reims), Dr Sylvie Kirscher (Institut Sainte Catherine, Avignon), Dr Laurence Delique (Centre Hospitalier Universitaire Nancy-Brabois, Vandoeuvre les Nancy), Pr Philippe Rougier (Hôpital Européen Georges Pompidou, Paris), Dr Céline Lepere (Hôpital Européen Georges Pompidou, Paris), Dr Olivier Dubreuil (Hôpital Européen Georges Pompidou, Paris), Dr Aziz Zaanan (Hôpital Européen Georges Pompidou, Paris), Dr Guillaume Medinger (Centre Hospitalier Départemental Vendée, La Roche-Sur-Yon), Pr Antoine Adenis (Centre Oscar Lambret, Lille), Dr Farid El Hajbi (Centre Oscar Lambret, Lille), Dr Gilles Gatineau Sailliant (Centre Hospitalier, Meaux)Dr Christophe Locher (Centre Hospitalier, Meaux), Dr Pierre-Luc Etienne (Centre Cario-HPCA, Plérin), Dr Sandrine Lavau Denes (Centre Hospitalier Universitaire Dupuytren, Limoges), Pr Thierry Lecomte (Hôpital Trousseau, Tours), Dr Vincent Hautefeuille (Hôpital Nord, Amiens), Pr Romain Coriat (Hôpital Cochin, Paris), Pr David Tougeron (Hôpital de la Milétrie, Poitiers), Dr Mathieu Baconnier (Centre Hospitalier Annecy-Genevois, Pringy), Pr Pierre Michel (Hôpital Charles Nicolle, Rouen), Dr Valérie Blondin (Hôpital Charles Nicolle, Rouen), Dr Olivier Joseph capitain (Institut de Cancérologie de l’Ouest Site Paul Papin, Angers), Dr Anne-Laure Villing (Centre Hospitalier, Auxerre), Dr Olivier Boulat (Hôpital Henri Duffaut, Avignon), Dr Anne-Marie Queuniet (Centre Hospitalier Intercommunal Elbeuf-Louviers-Val de Reuil, Elbeuf), Dr Véronique Jestin Le Tallec (Hôpital Morvan, Brest), Dr Robert Herve (Clinique Clairval, Marseille), Dr Virginie Sebbagh (Centre Hospitalier, Compiègne), Dr Isabelle Baumgaertner (Hôpital Henri Mondor, Créteil), Dr Gilles Boschetti (Hôpital Edouard Herriot, Lyon), Dr Amar Aouakli (Centre Hospitalier de l’Arrondissement de Montreuil, Rang du Fliers), Pr Emmanuel Mitry (Institut Curie, Saint Cloud), Dr Elisabeth carola (Centre Hospitalier, Senlis), Dr Ahmed Bedjaoui (Hôpitaux du Léman, Thonon Les Bains), Dr Hélène castanie (Institut de Cancérologie de l’Ouest – Site René Gauducheau, Saint Herblain), Pr Jean-Yves Douillard (Institut de Cancérologie de l’Ouest – Site René Gauducheau, Saint Herblain), Dr Hélène Senellart (Institut de Cancérologie de l’Ouest – Site René Gauducheau, Saint Herblain), Dr Sandrine Hiret (Institut de Cancérologie de l’Ouest – Site René Gauducheau, Saint Herblain), Dr Véronique Lorgis (Centre Georges François Leclerc, Dijon), Dr Aude Vincent (Centre Georges François Leclerc, Dijon)Pr Michel Ducreux (Institut Gustave Roussy, Villejuif), Dr Pascal Burtin (Institut Gustave Roussy, Villejuif), Dr David Malka (Institut Gustave Roussy, Villejuif), Dr Elie Zrihen (Institut Gustave Roussy, Villejuif), Dr Antoine Hollebecque (Institut Gustave Roussy, Villejuif), Dr Bruno Landi (Hôpital Européen Georges Pompidou, Paris), Dr Isabelle Trouilloud (Hôpital Européen Georges Pompidou, Paris), Dr Philippe Follana (Centre Antoine Lacassagne, Nice), Dr Xavier Coulaud (Centre Hospitalier de Saint Etienne – Hôpital Nord, Saint Priest en Jarez), Dr Christian Borel (Centre Paul Strauss, Strasbourg), Dr Christine Belletier (Centre Paul Strauss, Strasbourg), Dr François Dewaele (Centre Hospitalier Départemental Vendée, La Roche-Sur-Yon), Dr Thierry Chatellier (Centre Hospitalier Départemental Vendée, La Roche-Sur-Yon) ET (Clinique Mutualiste de l’Estuaire, Saint Nazaire), Pr Laétitia Dahan (Centre Hospitalier Universitaire La Timone, Marseille), Dr Muriel duluc (Centre Hospitalier Universitaire La Timone, Marseille), Dr Emmanuelle Norguet monnereau (Centre Hospitalier Universitaire La Timone, Marseille), Dr Catherine Bechi (Centre de Cancérologie du Grand Montpellier), Dr Béatrice Lafforgue (Centre de Cancérologie du Grand Montpellier), Dr Eric Janssen (Centre de Cancérologie du Grand Montpellier), Pr Laurent Bedenne (Centre Hospitalier Universitaire François Mitterrand, Dijon), Pr Côme Lepage (Centre Hospitalier Universitaire François Mitterrand, Dijon), Dr Paul Arthur Haineaux (Centre Hospitalier Universitaire François Mitterrand, Dijon), Dr Franck Audemar (Centre Hospitalier Côte Basque, Bayonne) ET (Centre Hospitalier Saint Jean, Perpignan), Dr Christelle De La Fouchardiere (Centre Léon Bérard, Lyon), Dr Jérôme Dauba (Hôpital Layné, Mont de Marsan), Dr Yves Bechade (Institut Bergonié, Bordeaux), Pr Dominique Bechade (Institut Bergonié, Bordeaux), Dr Laurent Cany (Clinique Francheville, Périgueux), Dr Charles-Briac Levache (Clinique Francheville, Périgueux), Dr Nicolas Barriere (Hôpital Européen, Marseille), Pr Jean-Baptiste Bachet (Hôpital La Pitié Salpetrière, Paris), Dr Touraj Mansourbakht (Hôpital La Pitié Salpetrière, Paris), Dr Lynn Rob (Institut de Cancérologie de Lorraine - Alexis Vautrin, Vandoeuvre Les Nancy), Pr Thierry Conroy (Institut de Cancérologie de Lorraine - Alexis Vautrin, Vandoeuvre Les Nancy), Dr Marie-Christine Kaminsky (Institut de Cancérologie de Lorraine - Alexis Vautrin, Vandoeuvre Les Nancy), Dr Isabelle Cumin (Centre Hospitalier de Bretagne Sud-Hôpital du Scorff, Lorient), Dr Joëlle Egreteau (Centre Hospitalier de Bretagne Sud-Hôpital du Scorff, Lorient), Dr Régine Lamy (Centre Hospitalier de Bretagne Sud-Hôpital du Scorff, Lorient), Dr Jean-Marc Gornet (Hôpital Saint Louis, Paris), Dr Oana Cojocarasu (Centre Hospitalier, Le Mans), Dr Mohamed Gasmi (Centre Hospitalier Universitaire – Hôpital Nord, Marseille), Dr Dominique Luet (Centre Hospitalier Universitaire, Angers), Pr François-Xavier Caroli-Bosc (Centre Hospitalier Universitaire, Angers), Dr Guillaume Roquin (Centre Hospitalier Universitaire, Angers), Dr Caroline Couffon (Centre Hospitalier Universitaire, Angers), Dr Benjamin Linot (Institut de Cancérologie de l’Ouest – Site Paul Papin, Angers), Dr Véronique Guerin-meyer (Institut de Cancérologie de l’Ouest – Site Paul Papin, Angers), Dr Suzanne Nguyen (Centre Hospitalier, Beauvais), Dr Nathalie Gerardin (Centre Hospitalier, Montauban), Dr Miguel Carreiro (Centre Hospitalier, Montauban), Dr Corinne Sarda (Centre Hospitalier Intercommunal Castres-Mazamet, Castres), Dr Nathalie Hess laurens (Centre Hospitalier Intercommunal Castres-Mazamet, Castres), Dr Yann Molin (Hôpital Edouard Herriot, Lyon), Dr Catherine Lombard-bohas (Hôpital Edouard Herriot, Lyon), Pr Thomas Walter (Hôpital Edouard Herriot, Lyon), Dr Julien Forestier (Hôpital Edouard Herriot, Lyon), Dr Hatem Salloum (Centre Hospitalier, Meaux), Dr Farah Zerouala Boussaha (Centre Hospitalier, Meaux), Dr Jérôme Meunier (Centre Hospitalier Départemental Vendée, Orléans), Dr Corina Cornila (Centre Hospitalier Départemental Vendée, Orléans), Dr Laetitia Stefani (Centre Hospitalier Annecy Genevois, Pringy), Dr Christian Maillard (Centre Hospitalier Annecy-Genevois, Pringy), Dr Marc Porneuf (Hôpital Yves Le Foll, Saint Brieuc), Dr Vincent Quentin (Hôpital Yves Le Foll, Saint Brieuc), Dr Claire Giraud (Centre Léonard de Vinci, Dechy), Dr Jean-Marie Vantelon (Centre Léonard de Vinci, Dechy), Dr Anne Cadier Lagnes (Centre Hospitalier Côte Basque, Bayonne), Dr Mohammed Ramdani (Centre Hospitalier, Béziers), Dr Yann Le Bricquir (Centre Hospitalier, Béziers), Dr Florence Kikolski (Hôpital Robert Boulin, Libourne), Dr Gaël Goujon (Hôpital Bichat, Paris), Dr Eric Terrebonne (Hôpital du Haut Lévèque, Pessac), Dr Julien Vergniol (Hôpital du Haut Lévèque, Pessac), Dr Sophie Pesque-Penaud (Hôpital du Haut Lévèque, Pessac), Dr Marie-Claude Gouttebel (Hôpitaux Drôme Nord – Site de Romans, Romans sur Isère), Dr Laurent Gasnault (Centre Joliot Curie, Saint Martin Les Boulogne), Dr Annunziato Alessio (Clinique Trenel, Sainte Colombe les Vienne), Dr Ivan graber (Clinique Trenel, Sainte Colombe les Vienne), Pr Etienne Dorval Danquechin (Hôpital Trousseau, Tours), Dr Dany Gargot (Centre Hospitalier, Blois), Dr Vincent Bourgeois (Hôpital Duchenne, Boulogne sur Mer), Dr Anne-Marie Queuniet (Centre Hospitalier Intercommunal Elbeuf-Louviers-Val de Reuil, Elbeuf), Dr Dominique Sevin Robiche (Polyclinique Sainte Marguerite, Auxerre), Dr Stéfanie Oddou Lagraniere (Hôpital Général, Gap), Dr Joël Ezenfis (Centre Hospitalier, Longjumeau), Dr Denis Pere Verge (Hôpital Saint Joseph Saint Luc, Lyon), Souquet Jean-Christophe (Hôpital de la Croix Rousse, Lyon), Dr Lionel Wander (Hôpital de la Croix Rousse, Lyon), Dr Abakar Abakar Mahamat (Hôpital l’Archet II, Nice), Dr Jacques Cretin (Institut de Cancérologie du Gard, Nîmes), Dr Anne-Claire Dupont gossart (Hôpital Jean Minjoz, Besançon), Dr Philippe Martin (Centre Hospitalier Intercommunal, Créteil), Dr Geneviève Boilleau-Jolimoy (Institut de cancérologie de Bourgogne – GRReCC, Dijon), Dr Anne Thirot Bidault (Hôpital de Bicêtre, Kremlin-Bicêtre), Dr Ahmed Azzedine (Centre Hospitalier, Montélimar),Dr Zied Ladhib (Centre Hospitalier, Valence).
Correction notice This article has been corrected since it published Online First. The author Thomas Aparicio’s name has been corrected.
Patient consent for publication Not required.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.