Identification of Non–Small Cell Lung Cancer Sensitive to Systemic

Published OnlineFirst March 20, 2020; DOI: 10.1158/1078-0432.CCR-19-2942

CLINICAL CANCER RESEARCH | PRECISION MEDICINE AND IMAGING

Identiﬁcation of Non–Small Cell Lung Cancer Sensitive to Systemic Cancer Therapies Using Radiomics Laurent Dercle1,2, Matthew Fronheiser3, Lin Lu1, Shuyan Du3, Wendy Hayes3, David K. Leung3, Amit Roy4, Julia Wilkerson5, Pingzhen Guo1, Antonio T. Fojo6, Lawrence H. Schwartz1, and Binsheng Zhao1

ABSTRACT ◥ Purpose: Using standard-of-care CT images obtained from progression-free survival (NCT01642004, NCT01721759) or sur- patients with a diagnosis of non–small cell lung cancer (NSCLC), gery (NCT00588445). Machine learning was implemented to select we defined radiomics signatures predicting the sensitivity of tumors up to four features to develop a radiomics signature in the training to nivolumab, docetaxel, and gefitinib. datasets and applied to each patient in the validation datasets to Experimental Design: Data were collected prospectively classify treatment sensitivity. and analyzed retrospectively across multicenter clinical trials Results: The radiomics signatures predicted treatment sensitivity [nivolumab, n ¼ 92, CheckMate017 (NCT01642004), Check- in the validation dataset of each study group with AUC (95 Mate063 (NCT01721759); docetaxel, n ¼ 50, CheckMate017; gefi- confidence interval): nivolumab, 0.77 (0.55–1.00); docetaxel, 0.67 tinib, n ¼ 46, (NCT00588445)]. Patients were randomized to (0.37–0.96); and gefitinib, 0.82 (0.53–0.97). Using serial radiograph- training or validation cohorts using either a 4:1 ratio (nivolumab: ic measurements, the magnitude of exponential increase in signa- 72T:20V) or a 2:1 ratio (docetaxel: 32T:18V; gefitinib: 31T:15V) ture features deciphering tumor volume, invasion of tumor bound- to ensure an adequate sample size in the validation set. Radiomics aries, or tumor spatial heterogeneity was associated with shorter signatures were derived from quantitative analysis of early tumor overall survival. changes from baseline to first on-treatment assessment. For Conclusions: Radiomics signatures predicted tumor sensitivity each patient, 1,160 radiomics features were extracted from to treatment in patients with NSCLC, offering an approach that the largest measurable lung lesion. Tumors were classified as could enhance clinical decision-making to continue systemic ther- treatment sensitive or insensitive; reference standard was median apies and forecast overall survival.

Introduction Progress in artificial intelligence (AI) has transformed the field of radiology, especially radiomics. Radiomics depends on the quan- Selecting patients for targeted therapies or immunotherapy is titative transformation of images into comprehensive datasets that crucial to match individuals to the treatment most likely to benefit enables high-throughput data mining and automated analysis of them. In patients with a diagnosis of non–small cell lung cancer patterns present in images. These quantitative imaging biomarkers, (NSCLC), personalization of therapy currently relies on pretreatment defined aprioriusing mathematical formulas, could guide treat- biomarkers acquired in a tumor biopsy taken at baseline. Tumor ment decision. Radiomics features are calculated by algorithmic biopsies are used to perform genomic analyses to find therapeutically analysis of tumor images and have been linked to characteristics of actionable mutations (e.g., EGFR and anaplastic lymphoma kinase, NSCLC. AI can be trained to recognize clinically relevant patterns ALK), as well as the expression of proteins that might help predict on CT images and perform a “digital biopsy” of the imaging sensitivity to immunotherapy (e.g., programmed cell death 1 ligand, phenotype of the entire tumor volume. Quantitative imaging fea- PD-L1). These measures are typically limited to a single biopsy sample, tures have been associated with therapeutically actionable muta- are difficult to perform repeatedly, and thus cannot capture the spatial tions, such as EGFR mutation status (1–15). Early volumetric and temporal heterogeneity of disease. assessment of variation in imaging phenotype on CT scan (16–23) has been shown to predict the biologic activity of targeted therapies suchasanti-EGFRagents(17,23).Finally,theimagingphenotype of NSCLC has been linked to patients' outcome (23, 24–30) by 1Department of Radiology, Columbia University Medical Center/New York predicting metastasis risk, recurrence risk, gross residual disease, Presbyterian Hospital, New York, New York. 2Gustave Roussy, Universite and survival. 3 Paris-Saclay, Villejuif, France. Translational Medicine, Bristol-Myers Squibb, Currently, the main strategy for response assessment is based on a Princeton, New Jersey. 4Clinical Pharmacology and Pharmacometrics, Bristol- 5 radiologist's evaluation of the changes in the size and the appearance of Myers Squibb, Princeton, New Jersey. The National Cancer Institute, NIH, “ ” Bethesda, Maryland. 6Columbia University/New York Presbyterian Hospital and new tumor lesions. In the case of targeted therapies, shrinkage of James J. Peters VA Medical Center, New York, New York. target lesions is considered a hallmark of dependency on the targeted pathway and consequently of treatment sensitivity. With low variation Note: Supplementary data for this article are available at Clinical Cancer Research Online (http://clincancerres.aacrjournals.org/). due to lesion sampling, acquisition protocol, and observer effect, tumor shrinkage was a robust tool in the chemotherapy era to assess Corresponding Author: Laurent Dercle, Columbia University Medical Center, anticancer treatment efficacy, and is widely employed using RECIST 168th Street, New York, NY 10032. Phone: 9172830908; E-mail: – [email protected] 1.1 (31 33). In immune oncology, the utility of RECIST is more limited due to atypical patterns of response and progression on medical Clin Cancer Res 2020;XX:XX–XX imaging. Despite the use of new imaging response assessment criteria, doi: 10.1158/1078-0432.CCR-19-2942 an unmet clinical need for improved assessment remains, which 2020 American Association for Cancer Research. justifies the need for alternative approaches.

AACRJournals.org | OF1

Dercle et al.

or gefitinib using a quantitative analysis of early tumor changes from Translational Relevance baseline to first on-treatment tumor assessment on serial CT scans. New patterns of response and progression have been observed in Our secondary objective was to test whether features were generaliz- patients treated with immunotherapy, such as pseudoprogression able across treatment arms. and hyperprogression, prompting the need for alternative metrics for response assessment and therapeutic decision-making. Radio- Selection of eligible trials mic signatures, derived from quantitative, artificial intelligence- The following inclusion criteria were used for trial eligibility: based analysis of standard-of-care CT images, offer the potential to completed study; primary tumor type with a large proportion of enhance clinical decision-making as on-treatment markers of measurable, quantifiable disease (NSCLC); ≥40 patients accrued; efficacy. In patients with non–small cell lung cancer treated with FDA-approved drug; and CT images centrally collected and archived. a wide spectrum of systemic cancer therapies (nivolumab, doc- We selected completed trials evaluating three different types of drug etaxel, or gefitinib), radiomic signatures detected early changes classes (an immunotherapy, a cytotoxic chemotherapy, and a molec- from baseline to first on-treatment tumor assessment that were ularly targeted agent) so that imaging metrics could be studied across a associated with sensitivity to treatment. Using serial radiographic range of different therapies. measurements, we observed that an exponential increase in signature features deciphering tumor volume, invasion of tumor Participants boundaries, or tumor spatial heterogeneity was associated with We retrospectively analyzed three NSCLC clinical trials using three treatment insensitivity and shorter overall survival. This indicates treatments (nivolumab, docetaxel, or gefitinib) and 188 patients. radiomic signatures offer an approach that could guide clinical Patients treated with nivolumab (n ¼ 92) had enrolled in the decision-making to continue or modify systemic therapies. NCT01642004 and NCT01721759 multicenter phase II–III trials. Patients who received docetaxel (n ¼ 50) had been treated in the NCT01642004 trial. Patients prescribed gefitinib (n ¼ 46) received the agent on NCT00588445, a single arm phase II trial in NSCLC that The investigation of pretreatment biomarkers to identify patients sought to correlate the radiographic response induced by gefitinib with who might benefit from immunotherapy has gained traction within mutations in the protein-tyrosine kinase domain of the EGFR gene. the research community. The unique challenge is unconventional Patient characteristics are summarized in Table 1 and in Supplemen- patterns of response and progression including hyperprogression and tary SI.1 and SII.1. pseudoprogression. As ancillary studies, given the clear implication of The investigators of these clinical trials obtained written oncologic progression, it was our hypothesis that extending our informed consent from the patients. The studies were conducted imaging evaluation to include pretreatment imaging features as well in accordance with recognized ethical guidelines (Declaration of as serial radiographic measurements might be an untapped source of Helsinki), and the studies were approved by an Institutional Review complementary prognostic information. Board. Our focus was on advanced/metastatic NSCLC because it is a unique Data were collected up to the completion date of the clinical trials. candidate for implementing a radiomics approach. First, there is a Patients were randomly assigned to either training (T) or validation strong clinical need because lung cancer is the second most common (V) sets using either a 4:1 ratio (nivolumab: 72T:20V) or a 2:1 ratio cancer and a leading cause of cancer death for men and women. (docetaxel: 32T:18V; gefitinib: 31T:15V) to ensure an adequate sample Second, the segmentation of lung lesion can be easily implemented in size in the validation set. We estimated that a sample size of a minimum clinical routine because the healthy air-filled lung parenchyma is the of 14 patients was required in the validation set based on the following most hypodense human tissue. Third, from a biological perspective, input and assumption: a type I error of 0.05, a power of 0.8, an AUC of the paradigm of response was developed in cytotoxic chemotherapies, 0.85, and an allocation ratio of 1 (35). which generates a need to explore change in imaging phenotype In NCT01642004 and NCT01721759, CT scan imaging was per- induced by other types of systemic treatments such as targeted therapy formed at baseline and again every 8 weeks until disease progression or and immunotherapy. Finally, radiographic response evaluation is a withdrawal. In NCT00588445, CT scan imaging occurred at baseline standard of care, and the response rate is known to be sufficient to and at 3 weeks, just prior to day-23 surgery. Patients with missing data make the results meaningful (34). were excluded. Additional trial details are included in Supplementary We aimed to explore whether AI techniques may be of clinical SI.1. utility to oncologists as on-treatment markers of efficacy to help decide which patients should continue treatment. To this end, we Quality of CT scan acquisition evaluated the performance of treatment-specificradiomicssigna- We selected 188 patients out of 264 eligible patients (Fig. 1, Table 1) tures measured at baseline and at the first response assessment in based on eligibility criteria ensuring improved imaging quality for this patients with NSCLC receiving treatment with either of three cancer quantitative retrospective analysis: (i) measurable lung lesions accord- therapies: an immunotherapy blocking a negative regulator of T-cell ing to RECIST 1.1 at baseline; (ii) no significant imaging artifacts; (iii) activation and response (nivolumab, a monoclonal IgG4 antibody lung reconstruction kernel; (iv) pixel spacing <1 mm; (v) slice thick- targeting PD-1), a chemotherapy targeting microtubules (doce- ness <10 mm; and (vi) CT scans available at baseline and first response taxel), and a molecular targeted therapy interrupting EGFR signal- evaluation (landmark). ing (gefitinib). The radiomics quality score (RQS) has been proposed as a guideline to evaluate the quality of radiomics studies (36). For the current study, the RQS was estimated to be 28 out of 36 points (78%). More Materials and Methods information about CT scan characteristics, CT scan quality (37–39), Our primary objective was to train and validate three on-treatment and RQS can be found in Supplementary SI.1. A depiction of the signatures to detect NSCLC tumors sensitive to nivolumab, docetaxel, radiomics workflow is shown in Fig. 1.

OF2 Clin Cancer Res; 2020 CLINICAL CANCER RESEARCH

Radiomics Signatures in NSCLC

Table 1. Patients' characteristics and performance of the signatures to detect tumor sensitivity to treatment.

Treatment arm Nivolumab Docetaxel Geﬁtinib

Tumor characteristics Tumor type Squamous cell carcinoma Squamous cell carcinoma Bronchioloalveolar Stage (AJCC) Advanced (IIIB/IV) Advanced (IIIB/IV) Early-stage (resectable I/II) Lesion segmented Lung lesion Lung lesion Primary lung cancer Original clinical trial characteristics Biomarker PD-L1 PD-L1 EGFR mutational status Treatment regimen Immunotherapy Nivolumab None None Chemotherapy None Docetaxel None Anti-EGFR mAbs None None Geﬁtinib Primary endpoint Outcome OS, PFS OS, PFS Surgery at 3 weeks First CT scan assessment Baseline/8-week Baseline/8-week Baseline/3-week Reference standard for treatment sensitivity Above/below median PFS Above/below median PFS Surgery at 3 weeks Clinical trial number NCT01642004, NCT01721759 NCT01642004 NCT00588445 Available n 153 patients 65 patients 46 patients Included n 92 patients 50 patients 46 patients Randomization Ratio 4: 1 2: 1 2: 1 Training n 72 patients 32 patients 31 patients Reference standard Sensitive 28 patients 5 patients 13 patients Insensitive 44 patients 27 patients 18 patients Signature AUC (95 CI) 0.80 (0.69–0.89) 0.68 (0.38–0.98) 0.81 (0.61–0.92) Delta features 1. Volume (burden) 1. Volume (burden) 1. Shape SI4 (boundaries) 2. GLCM IMC1 (heterogeneity) 2. LoG Z Entropy (heterogeneity) 3. DWT1 (heterogeneity) 3. GTDM Contrast (heterogeneity) 4. Sigmoid slope (boundaries) 4. LoG X Entropy (heterogeneity) Algorithm (42) Random Forest Random Forest Random Forest Validation n 20 patients 18 patients 15 patients Reference standard Sensitive 5 patients 6 patients 7 patients Insensitive 15 patients 12 patients 8 patients Signature AUC (95 CI) 0.77 (0.55–1.00) 0.67 (0.37–0.96) 0.82 (0.53–0.97) Sensitivity 0.80 0.92 0.83 Speciﬁcity 0.53 0.45 0.88

Note: The EGFR signature was designed in a cohort of patients with metastatic colorectal cancer treated with anti-EGFR using CT scans acquired at baseline and 8 weeks (44). The signature was then transferred and validated in Gefitinib patients with NSCLC treated with gefitinib. Signature features are all delta features measuring the change in the imaging feature between baseline and first response assessment.

Lesion segmentation and feature extraction Signature building in each treatment arm The largest measurable lung tumor present at baseline was In the training set of each cohort, we developed a multivariable segmented in the baseline and in the first on-treatment response prediction model, i.e., the signature, to predict treatment sensitivity. assessment CT scan (nivolumab, 8 weeks; docetaxel, 8 weeks; Using machine learning, quantitative imaging features were combined gefitinib, 3 weeks) in all patients that met the inclusion criteria. by high-throughput mining to build the signature. In keeping with our Segmentation was performed using an algorithm developed in- ultimate aim to generate simple to use, noninvasive clinical decision house that enables semiautomatic creation of contours on all tools using standard of care CT scan images, the radiomics signature available CT scans (40). Imaging features were extracted from ranging from 0 (highest treatment sensitivity) to 1 (highest treatment lung tumors using aprioridefinitions of radiomics features (38). insensitivity) for each patient was based on the analysis of the change of In total, 1,160 quantitative image features were extracted from the the largest measurable lung lesion identified at baseline on CT scan. images of each lesion from both the baseline and the first on- In the implementation, a “coarse” to “fine” strategy was developed treatment CT scan. Delta radiomics features were calculated to to select optimal features from the large number of the extracted characterize the early changes in the features. Full details of lesion quantitative image features to build the signature. The coarse selection segmentation and feature extraction can be found in Supplementary approach consists of reproducibility analysis, redundancy analysis, SI.2 and SI.3. and feature ranking that eliminate those nonreproducible, redundant,

AACRJournals.org Clin Cancer Res; 2020 OF3

Dercle et al.

Figure 1. Disposition of study patients. Patients could be excluded for multiple reasons. The withdrawal boxes show the number of patients excluded at each step. CT scans acquired at sites are transferred to our academic core. Image selection and quality check using a computer-aided algorithm designed by machine learning. Step 1. Segmentation of the largest measurable lung tumor on CT scan by an expert radiologist at baseline in all patient (inclusion criteria), as well as all available radiographics measurement. Steps 2–3. Tumor imaging phenotype in each patient based on imaging features extraction in the largest measurable segmented lung lesion (1,160 imaging features characterizing changes between baseline and first CT assessment). Step 4. Dimension reduction using machine learning. Identification of reproducible, nonredundant, and informative candidate imaging features for model building. Step 5. Signature building in the training set to enhance strategic decision-making and predict treatment sensitivity. Step 6. Signature validation. Step 7. Transfer of the signature features for evaluation of g and d values using serial radiographic measurements. Step 8. A subset of imaging biomarker is identified.

and noninformative features. The fine selection approach was com- (n ¼ 100 patients) a four-feature treatment sensitivity signature posed of “forward” search and feature combination, aiming to select (Random Forest algorithm; ref. 41) in a cohort of patients with the most significant features to build the best predictive model. To metastatic colorectal cancer treated with anti-EGFR monoclonal prevent overfitting, up to the top four features (in terms of prediction antibodies. To prevent overfitting, the four features identified in the importance outputted by the machine-learning algorithm; ref. 41) of colorectal cancer signature were transferred to be calibrated to predict the identified image features in the best predictive model developed in treatment sensitivity in the training set of the gefitinib cohort. the training set were integrated in the signature. Full details of the The radiomics signature was trained to predict the tumor sensitivity model building can be found in Supplementary SI.4. to systemic anticancer treatment. All patients with cancer were divided Due to the limited number of patients available in the gefitinib into two groups: sensitive and insensitive to treatment. In patients with cohort (n ¼ 46 patients), we defined an alternative strategy NSCLC treated with nivolumab and docetaxel, the reference standard for signature building. We trained (n ¼ 202 patients) and validated to determine treatment sensitivity was median progression-free

OF4 Clin Cancer Res; 2020 CLINICAL CANCER RESEARCH

Radiomics Signatures in NSCLC

survival (PFS). The reference standard for gefitinib-treated patients Statistical analysis was derived from the analysis of the surgical specimen 2 days after Statistical analysis was conducted using Matlab2016a and SPSS23.0. stopping 21 days of gefitinib therapy. The independent validation The reported P values were two-sided, with the level for statistical dataset, consisting of unseen data that were not used for training, was significance set at a ¼ 0.05. The performance of the model was used to evaluate the performance of the signature. Further details for evaluated by computing the area under the ROC curve. the reference standards are provided in Supplementary SI.5.1.

Generalizability of the features Results Our primary objective was to evaluate one single lesion at two Performance of radiomics signature: nivolumab timepoints (baseline and first on-treatment assessment). As an ancil- In the training set (n ¼ 72), the tumors of 28 patients were classified lary study, we evaluated the generalizability of imaging features and as sensitive to nivolumab. The nivolumab radiomics signature compared them with alternative outcome measures. To this end, we included four delta-radiomics features characterizing the change used all measurable lesions when we studied the clinical value of in tumor volume, heterogeneity, and margin sharpness: delta- imaging features using baseline measurement (one timepoint) or serial Volume, delta-GLCM IMC1 (Gray-Level Co-Occurrence Matrix), radiographic measurement (≥two timepoints). delta-DWT1 (Discrete Wavelet Transform), and delta-sigmoid slope (Table 1). The nivolumab signature achieved an AUC [95 Estimating rates of tumors with exponential changes in confidence interval (CI)] of 0.80 (0.69–0.89; P < 10 4) in its ability radiomics features using serial radiographic measurements to identify the sensitivity of a patient's tumor to nivolumab. We attempted to reduce the risk of type I error/overfitting and to In the validation set (n ¼ 20), the tumors of 5 patients were classified generalize the features identified in the three radiomics signatures as sensitive to nivolumab. The performance of the nivolumab signature (nivolumab, docetaxel, and gefitinib). To this end, we modeled the in the validation set was AUC (95 CI) of 0.77 (0.55–1.00). Using evolution of these features across time using serial radiographic Kaplan–Meier plots, the estimated median PFS of the overall popu- measurements of tumors. We evaluated if these models could be used lation (95 CI) was 2.1 (1.3–2.9) months (Fig. 2A). The estimated to understand disease behavior during treatment, compare study median PFS (95 CI) was 2.0 (1.8–2.2) versus 6.3 (4.0–8.6) months for interventions, and forecast overall survival (OS; ref. 42). patients with (n ¼ 57 patients) or without (n ¼ 35 patients) a high-risk Previous studies have demonstrated the simultaneous occurrence of nivolumab signature (signature > 0.5) respectively (P < 10 4). two processes in the overwhelming majority of tumors: exponential growth of the treatment-insensitive fraction of the tumor at a rate Performance of radiomics signature: docetaxel described by a growth rate constant designated g for growth, and In the training set (n ¼ 32), the tumors of 5 patients were classified as exponential regression of the treatment-sensitive portion of the tumor sensitive to docetaxel. The radiomics signature for the docetaxel cohort at a rate described by a regression rate constant designated d for decay. was a single delta-radiomics feature, delta-Volume (Table 1). The Both processes occur exponentially, and the rates of growth and performance of the docetaxel signature was AUC (95 CI) of 0.68 (0.38– regression can be estimated using simple mathematical equations. 0.98). Using the same equations, we were able to estimate the rates of In the validation set (n ¼ 18), the tumors of 6 patients were classified exponential change in the radiomics features over time and based on as sensitive to docetaxel, and the performance of the docetaxel radio- the rate of change for each radiomic feature assigned tumors into one mics signature was AUC (95 CI) of 0.67 (0.37–0.96). Using Kaplan– of three categories: (i) those in which the designated feature was Meier analyses, we observed that compared with tumors with a high- observed to only exponentially grow/increase or become more prom- risk docetaxel signature (signature > 0.5, n ¼ 39 patients) those without inent during treatment (gx), (ii) those in which the analyzed feature the high-risk signature (signature 0:5; n ¼ 35 patients) had a longer only decreased or disappeared exponentially in quantity during ther- estimated median PFS (95 CI) of 6.2 (5.5–7.0) months as compared apy (dx); and (iii) those in which the data were best described by an with 2.1 (2.0–2.3) months (P < 0.001; Fig. 2B). equation that considered that there had occurred concurrent exponential increase and disappearance of the radiomic feature exam- Performance of radiomics signature: gefitinib ined,asportionsofthetumorsensitive to the therapy disappeared In the training set (n ¼ 31), the tumors of 13 patients were classified while those resistant to the treatment increased in abundance (gd; as sensitive to gefitinib. The gefitinib signature was composed of four ref. 42). Rates of growth or decay of each feature were obtained by delta-radiomics features characterizing the change in tumor shape and using all available radiographic measurements from baseline to a heterogeneity: delta-Shape-SI4, delta-LOG-X-Entropy, delta-LOG-Z- landmark analysis at 8 months. The differences in OS between Entropy, and delta-GTDM-Contrast (Table 1). The performance of patients with g above median and below median were analyzed the gefitinib signature in the training set was AUC (95 CI) of 0.81 using a cox proportional hazards model and log-rank test (Kaplan– (0.61–0.92). Meier analysis) in which landmark analysis divided the follow-up In the validation set (n ¼ 15), the tumors of 7 patients were classified time at 8-month time point and a P value less than 0.05 was as sensitive to gefitinib. The performance of the gefitinib radiomics considered significant. signature was AUC (95 CI) of 0.82 (0.53–0.97). Table 1 includes a summary of the patient information, treatment-specific radiomics Baseline imaging predictors of radiographic progression signatures, and performance metrics. We evaluated whether baseline radiomics features could predict the best overall response of individual lung lesions. To this end, we Estimating rates of tumors with exponential changes in classified all measurable lung lesions in the nivolumab cohort into radiomics features using serial radiographic measurements two categories: progressive versus nonprogressive per iRECIST In addition to assessing the radiomic features at baseline and in the criteria (43). Supplementary SII.3 contains additional methodology fi rst on-treatment tumor assessment, we were also interested in details. examining the kinetics of change in these same radiomic features as

AACRJournals.org Clin Cancer Res; 2020 OF5

Dercle et al.

Figure 2. Probability of PFS and OS over time as a function of signature score and signature features. Prolonged PFS was observed in patients with low-risk/treatment- sensitive signatures (≤0.5) in both treatment study groups (Fig. 2) using baseline and 8-week CT scans. In the nivolumab cohort (A), median PFS (95% CI) was 2.0 months (1.8–2.2) for patients whose tumors had a signature score > 0.5 (predicted insensitivity, n ¼ 57) and 6.3 months (4.0–8.6) for patients with a signature score ≤ 0.5 (predicted sensitivity, n ¼ 35; P < 104). In the docetaxel cohort (B), median PFS (95% CI) was 2.1 months (2.0–2.3) for patients whose tumors had a signature score > 0.5 (predicted insensitivity, n ¼ 39) and 6.2 months (5.5–7.0) for signature score ≤ 0.5 (predicted sensitivity, n ¼ 11; P < 104). Using serial radiographic measurements and a landmark at 8-month after drug initiation, we observed that the rate of exponential increase (g) of the radiomic features included in the signatures was associated with OS in patients from both treatment groups (C–E, pooled groups). The magnitude of exponential increase in tumor volume

(gVolume, C), tumor spatial heterogeneity (gGLCM-IMC1, D), or boundary irregularity (gShape-SI4, E) was associated with shorter OS.

treatment was administered. To do this, we used the values of the whereas the rates for docetaxel were dx 21% (14–35), gx 22% (18–30), eight radiomic features mentioned above from serial scans to assess and gd 11% (11–12). For boundaries features, the nivolumab rates were which equation previously shown to describe the rates of change in dx 18% (13–24), gd 11% (4–18), and gx 40% [39–41]; whereas the rates tumor volume over time could best describe the kinetics of change for docetaxel were dx 11% (8–15), gx 36% (23–48), and gd 13% (4–21). in the radiomic features with therapy. Table 2 shows that in the Figure 2C–E show shorter OS for patients whose kinetics majority of the 224 patients with data available for analysis, simple were such that the rate of exponential increase in radiomics mathematical equations could be used to describe three categories feature (g) was above the median for gVolume (P ¼ 0.005), P ¼ P < for each radiomics feature in the analyzed tumors: those with only gGLCM-IMC1 ( 0.02), and gShape-SI4 ( 0.001). Therefore, an exponential increase in the designated radiomic feature (gx), those exponential increase in either tumor volume, or tumor heteroge- with only exponential decrease in the analyzed feature (dx), and neity, or shape irregularity can forecast shorter OS. The imaging those tumors in which the kinetic change in the studied radiomics feature Shape-SI4 provided clinically useful information across four feature was best described by an equation that included concurrent treatment arms. Shape-SI4, originally identified in patients with exponential increase and reduction (gd). Figure 3 shows the colorectal cancer (44), was transferred to predict sensitivity to distribution (i.e., bimodal, trimodal) of g and d values for each gefitinib in patients with NSCLC. Shape-SI4 was associated with radiomics feature. OS in NSCLC treated with docetaxel and nivolumab. See additional To simplify the analysis and its interpretation, the eight features informationinSupplementarySI.6.2andSII.4. mentioned above were divided into three categories: tumor burden (volume), tumor heterogeneity (LOG-X-Entropy, LOG-Z-Entropy, Baseline imaging predictors of radiographic progression GTDM-Contrast, GLCM-IMC1, GTDM-contrast), and boundaries On a per-lesion analysis, the best overall response was objective (Shape-SI4, sigmoid-slope). Tumor burden was established with either radiographical progression in 136 (36.4%) lesions treated with unidimensional or bidimensional measurements. The average per- nivolumab and evaluated in at least two timepoints (baseline and centage (min–max) of tumors with estimable rates of decrease (dx), 8 weeks). The best overall response in other lesions (63.6%) was increase (gx), or concurrent increase and decrease (gd) in the various pseudoprogression (n ¼ 4, 1.1%), response (n ¼ 26,7.0%),and radiomic features was computed for the three categories of radiomics stability (n ¼ 207, 55.5%). An analysis of baseline radiomics features features (Table 2). For the tumor burden features, the rates for of these lung lesions demonstrated that the best baseline predictors nivolumab were dx 7% (6–7), gx 49% (43–55), and gd 24% (21– of progression were features associated with tumor heterogeneity 25); whereas the rates for docetaxel were dx 4% (3–6), gx 43% (35–49), [RUN PLU, AUC (95 CI) ¼ 0.82 (0.72–0.92)], shape [Shape index 3, and gd 26% (23–27). For the heterogeneity features, the rates for AUC (95 CI) ¼ 0.82 (0.89–67)], and volume [AUC (95 CI) ¼ 0.78 nivolumab were dx 23% (19–31), gx 22% (19–25), and gd 12% (6–15); (0.89–0.67)]. These findings suggest that larger infiltrative lung

OF6 Clin Cancer Res; 2020 CLINICAL CANCER RESEARCH

Downloaded from clincancerres.aacrjournals.org on September 24, 2021. © 2020 American Association for Cancer Research. AACRJournals.org Table 2. Summary of the percentage of patients ﬁtting g and d models for the eight radiomics features discovered in the three signatures.

Downloaded from Tumor burden Heterogeneity Boundaries Log X Log Z DWT1 GLCM GTDM Shape Sigmoid Averagea Uni Bi Volume Averagea Entropy Entropy HHL IMC1 contrast Averagea SI4 Slope

NIVOLUMAB Patients analyzed, Total included 129 (81.6) 122 (77.2) 128 (81.0) 137 (86.7) 85 (54.1) 62 (39.2) 88 (55.7) 93 (58.9) 87 (55.1) 97 (61.4) 112 (70.6) 114 (72.2) 109 (69.0)

n (%) dx 10 (6.6) 11 (7) 9 (5.7) 11 (7) 37 (23.4) 33 (20.9) 34 (21.5) 39 (24.7) 30 (19) 49 (31) 29 (18.4) 20 (12.7) 38 (24.1) Published OnlineFirstMarch20,2020;DOI:10.1158/1078-0432.CCR-19-2942 gx 77 (48.5) 68 (43) 75 (47.5) 87 (55.1) 34 (21.7) 39 (24.7) 35 (22.2) 31 (19.6) 30 (19) 36 (22.8) 63 (39.6) 61 (38.6) 64 (40.5) clincancerres.aacrjournals.org gd 37 (23.6) 40 (25.3) 39 (24.7) 33 (20.9) 18 (11.7) 23 (14.6) 15 (9.5) 21 (13.3) 23 (14.6) 10 (6.3) 18 (11.1) 29 (18.4) 6 (3.8)

Patients not Total excluded 29 (18.4) 36 (22.8) 30 (19.0) 21 (13.3) 73 (45.9) 96 (60.8) 70 (44.3) 65 (41.1) 71 (44.9) 61 (38.6) 47 (29.4) 44 (27.8) 49 (31.0) analyzed, n (%) Erroneous data ————1 (0.5) — 4 (2.5) —————— No measurement 1 (0.6) 1 (0.6) 1 (0.6) 1 (0.6) 1 (0.6) 1 (0.6) 1 (0.6) 1 (0.6) 1 (0.6) 1 (0.6) 4 (2.2) 1 (0.6) 6 (3.8) Two evaluations ≤20% 17 (10.8) 24 (15.2) 17 (10.8) 10 (6.3) 39 (24.8) 42 (26.6) 42 (26.6) 39 (24.7) 45 (28.5) 28 (17.7) 23 (14.6) 36 (22.8) 10 (6.3) difference Not ﬁt 11 (7.0) 11 (7) 12 (7.6) 10 (6.3) 25 (15.7) 19 (12) 23 (14.6) 25 (15.8) 25 (15.8) 32 (20.3) 20 (12.7) 7 (4.4) 33 (20.9)

DOCETAXEL Patients analyzed, Total included 49 (74.7) 46 (69.7) 53 (80.3) 49 (74.2) 36 (55.2) 33 (50.0) 33 (50.0) 42 (63.6) 30 (45.5) 44 (66.7) 41 (62.1) 36 (54.5) 46 (69.7)

Cancer Research. n (%) dx 3 (4.0) 4 (6.1) 2 (3) 2 (3) 14 (21.2) 10 (15.2) 13 (19.7) 15 (22.7) 9 (13.6) 23 (34.8) 8 (11.4) 5 (7.6) 10 (15.2) gx 28 (42.9) 23 (34.8) 32 (48.5) 30 (45.5) 15 (22.4) 15 (22.7) 12 (18.2) 20 (30.3) 14 (21.2) 13 (19.7) 24 (35.6) 15 (22.7) 32 (48.5)

Patients not Total excluded 17 (25.3) 20 (30.3) 13 (19.7) 17 (25.8) 30 (44.8) 33 (50.0) 33 (50.0) 24 (36.4) 36 (54.5) 22 (33.3) 25 (37.9) 30 (45.5) 20 (30.3) analyzed, n (%) Erroneous data ————————————— No measurement ————————————— Two evaluations ≤20% 12 (18.7) 19 (28.8) 8 (12.1) 10 (15.2) 20 (30) 24 (36.4) 23 (34.8) 18 (27.3) 24 (36.4) 10 (15.2) 14 (20.5) 22 (33.3) 5 (7.6) difference Not ﬁt 4 (6.6) 1 (1.5) 5 (7.6) 7 (10.6) 10 (14.9) 9 (13.6) 10 (15.2) 6 (9.1) 12 (18.2) 12 (18.2) 12 (17.4) 8 (12.1) 15 (22.7)

GEFITINIB Patients analyzed, Total included 5 (10.9) 4 (8.7) 5 (10.9) 6 (13.0) 3 (6.5) 2 (4.3) 3 (6.5) 3 (6.5) 3 (6.5) 4 (8.7) 4 (8.7) 4 (8.7) 4 (8.7) n (%) dx 0.7 (1.5) 1 (2.2) 1 (2.2) — 0.2 (0.4) ——1 (2.2) ——1 (2.2) — 2 (4.3) gx 0.3 (0.7) ——1 (2.2) 1.8 (3.9) — 3 (6.5) 2 (4.3) 2 (4.3) 2 (4.3) 1 (2.2) 1 (2.2) 1 (2.2) gd 4 (8.7) 3 (6.5) 4 (8.7) 5 (10.9) 1 (2.2) 2 (4.3) ——1 (2.2) 2 (4.3) 2 (4.3) 3 (6.5) 1 (2.2)

Patients not Total excluded 41 (89.1) 42 (91.3) 41 (89.1) 40 (87.0) 43 (93.5) 44 (95.6) 43 (93.5) 43 (93.5) 43 (93.5) 42 (91.3) 42 (91.3) 42 (91.3) 42 (91.3) analyzed, n (%) Erroneous data ————————————— No measurement ——————————0.5 (1.1) — 1 (2.2) NSCLC in Signatures Radiomics Two evaluations ≤20% 0.7 (1.5) 1 (2.2) 1 (2.2) — 1 (2.2) 1 (2.2) 1 (2.2) 1 (2.2) 1 (2.2) 1 (2.2) ——— difference Not ﬁt 0.3 (0.7) 1 (2.2) ——2 (4.3) 3 (6.5) 2 (4.3) 2 (4.3) 2 (4.3) 1 (2.2) 2 (4.3) 2 (4.3) 2 (4.3) lnCne e;2020 Res; Cancer Clin

aAverage: average value for the features included in the categories tumor burden, heterogeneity, and boundaries. Uni (Unidimensional), Bi (Bidimensional). Computation of the rate of decay (d) and growth (g) of radiomics features using serial radiographics measurement. The eight features discovered in the three signatures (nivolumab, docetaxel, and geﬁtinib) are generalized and applied to the three cohorts. In the cohort geﬁtinib, patients had only two evaluations, hence d and g values were not assessable in most patients. OF7 Published OnlineFirst March 20, 2020; DOI: 10.1158/1078-0432.CCR-19-2942

Dercle et al.

Figure 3. Distribution of the rates of patients with an exponential increase (g) or decrease (d) in the eight features included in the Radiomic signatures. Using serial radiographic measurements, the eight features discovered in the three signatures (nivolumab, docetaxel, geﬁtinib) (A) were generalized and applied to the all cohorts. Patients with exponential increase (g values) or decrease (d values) in Radiomic features are displayed using tumor burden (B), heterogeneity (C), and boundary features (D). This is a proof of concept that AI can be trained to differentiate the simultaneous occurrence of two processes in the overwhelming majority of tumors: exponential growth of the treatment-insensitive fraction of the tumor at a rate described by a growth rate constant designated g for growth, and exponential regression of the treatment-sensitive portion of the tumor, at a rate described by a regression rate constant designated d for decay. Strikingly, the distribution is bimodal in the Geﬁtinib cohort suggesting a wider variability between sensitive and insensitive tumors.

lesions are more likely to progress per iRECIST criteria. Additional complex task: identifying a pattern of baseline and treatment-induced details can be found in Supplementary SII.3. changes on CT images associated with sensitivity to systemic nivolu- Baseline imaging predictors of radiographic progression in lung mab, docetaxel, and gefitinib therapy in patients with a diagnosis of lesions treated with gefitinib were surrogates of tumor heterogeneity NSCLC (Fig. 4). and were reported previously (39, 45). Using baseline and first on-treatment assessment (nivolumab: 8 weeks, docetaxel: 8 weeks, gefitinib: 3 weeks), the radiomics signatures output a probability ranging from 0 to 1, corresponding respec- Discussion tively to the highest treatment sensitivity and treatment insensitivity. Using standard-of-care CT images acquired in multicenter clinical The innovation of our work compared with the existing literature was trials, machine-learning techniques successfully performed a specific that the signatures were dynamic, used a limited subset of features to

OF8 Clin Cancer Res; 2020 CLINICAL CANCER RESEARCH

Radiomics Signatures in NSCLC

Figure 4. Distribution of the rates of patients with an exponential increase (g) or decrease (d) in the eight features included in the Radiomic signatures. Visual representation of the imaging features included in the signature. The changes in tumor imaging phenotype of the “most sensitive” patient treated with nivolumab is displayed below. Tumor was segmented, and its shape and volume are represented using volume rendering (A). As demonstrated, CT scan images are transformed to other mathematical spaces for feature extraction, e.g., CT image is transformed to LOG space for computing the entropy value (spatial heterogeneity), and tumor pixels within segmentation contour are transformed to GLCM matrix (B). Using this information, a radiomic signature predicts treatment sensitivity which is associated with patients' OS (C). reduce overﬁtting, and signature features were generalized to three routinely acquired CT scans. Such an analysis could also be incorpo- treatments evaluated in multicenter studies (46, 47). The eight signa- rated into hybrid imaging modalities such as on the CT portion of a ture features were reproducible across image reconstruction settings noncontrast PET/CT. Therefore, they could be leveraged—once ﬁne- and were robust across tumor sites (Supplementary SII.6). Once tuned and optimized in larger cohorts—to guide clinical decisions such designed using machine learning, these signatures can be computed as changing systemic therapies at an appropriate time. for a given patient using a laptop based on the segmentation of the These signatures can be understood by both clinical oncologists and largest measurable lung lesion by an experienced radiologist on radiologists as noninvasive in vivo surrogates of biological changes

AACRJournals.org Clin Cancer Res; 2020 OF9

Dercle et al.

following treatment (Table 1). The eight imaging biomarkers included Early changes in tumor-parenchyma boundaries were strongly in the signatures fall into three categories: (i) indicators of change in associated with tumor sensitivity in patients treated with nivolumab tumor burden, (ii) indicators of change in tumor spatial heterogeneity, and gefitinib. Although these features are influenced by segmenta- and (iii) characterization of tumor-parenchyma boundaries. There- tion (39) and the possibility of change in the shape of lung tumors fore, we can assume that we identified three imaging hallmarks that during respiration, they might capture macroscopically the growth appear to be prognostic and generalizable quantitative CT radiomics pattern at tumor–lung parenchyma interfaces in NSCLC. Complex response biomarkers predicting sensitivity to therapies: a decrease in lung interfaces are indeed associated with aggressive malignant tumors tumor volume, heterogeneity, and tumor-parenchyma invasiveness. and poorer survival (52, 53). The combination of these biomarkers into a signature successfully In an attempt to generalize the features included in the three identified tumors sensitive to cancer treatments. signatures, we succeeded in the majority of patients in describing Using serial radiographic measurement and a landmark analysis the kinetics of change of the eight radiomics features. We estimated at 8 months, we demonstrated that a substantial percentage of rate constants for both the regression (decrease, d)andgrowth tumors exhibit an exponential increase in either tumor volume (increase, g) of the radiomic features using serial radiographic fi (gVolume), tumor spatial heterogeneity of contrast-enhanced images measurements from multiple timepoints. This is the rst demon- (gGLCM-IMC1), or shape irregularity (gShape-SI4). We demonstrated stration that the effect of a treatment on the rates of growth and/or that the magnitude of this exponential increase can be leveraged to decay in radiomics features can be estimated. Because OS remains forecast shorter OS. In sensitive tumor, we observed tumor shrink- the gold standard for efficacy in oncology, we performed a landmark age, an increase in tumor homogeneity (decrease in heterogeneity), analysis for OS and could demonstrate that increased gvolume (tumor and a progressive regularization of tumor contours. This is the burden), gGLCM IMC1 (heterogeneity), or gShape SI4 (boundaries first demonstration that the evolution of radiomics features deci- invasiveness) were associated with shorter OS. This is the first phering tumor phenotype under systemic therapy selection pressure demonstration that the rate of growth of radiomics features can be follows exponential kinetics and coincides with previous work used to predict OS (42). demonstrating that the quantity of tumor increases and decreases In an ancillary lesion-based analysis conducted on baseline lung exponentially. lesions, we observed that small invasive lung tumors with high Changes in tumor volume predicted treatment sensitivity in all heterogeneity (low RUN—Primitive length uniformity) are more cohorts. However, it played a more important role in the chemo- likely to progress per iRECIST criteria and further corroborate the therapy signature (docetaxel) than in the immunotherapy signature clinical significance of the hallmarks of treatment sensitivity used in (nivolumab), and was not of importance in the anti-EGFR signature the nivolumab signature. (gefitinib). This is interesting because size-based response criteria Ourstudyhaslimitations.Thisisaproof-of-conceptstudywith were originally developed to assess response of tumors to cytotoxic relatively small sample size. The sample was representative of a chemotherapies, with many questioning their general value in population of NSCLC in a large multicenter clinical trial. Using a assessing targeted molecular agents which are often said to be multicenter clinical trial and randomization of patients in training cytostatic and immunotherapy agents which have been reported and validation sets reduced the risk of overfitting. There was no to lead to unconventional patterns of response and progression. We apparent selection bias with the covariates balanced between demonstrated that temporal changes in intratumoral spatial het- included and excluded patients, and between training and valida- erogeneity were associated with sensitivity of NSCLC to anti–PD-1 tion cohorts for each treatment. However, there was a selection (nivolumab)aswellasanti-EGFR(gefitinib) therapies. Because CT based on the presence of a measurable lung lesion. The high scans are standard of care, noninvasive, informative of the entire frequency of measurable lung lesions in NSCLC in this series made tumor burden, and can be performed serially, they are well suited to our model applicable to 61% of patients with refractory NSCLC address spatial and temporal heterogeneity. Tumor heterogeneity is treated with nivolumab and 100% of patients with early stage a hallmark of cancer, and some have argued can emerge from the NSCLC treated with gefitinib. The overall classification perfor- inherent dynamic evolution and adaptation of clones in the pres- mance in docetaxel was likely underfitted. This was because the ence of drug selection pressure, albeit more likely over prolonged dataset was unbalanced with a minority (16%) of tumor sensitive to periodsoftimeandnotoverweeks(48).However,contrast- treatment. Hence, the final machine-learning model included one enhanced CT scans may capture macroscopic patterns of accumu- single feature to accurately predict treatment sensitivity because lation of iodine linked to tumor neovascularization occurring over more complex models would have been overfitted. The radiomics shorter time periods. This neovasculature is marked by heteroge- pipeline is complex, which makes the selection and identification neous and excessive blood flow, reduced drug delivery, hypoxia, of imaging biomarkers difficult to be widely adopted in nonaca- immune evasion, tumor progression, and metastasis (48). Our demic institutions. Nonetheless, we extensively described our meth- modeling framework using the largest measurable lung lesion is odology and clinically relevant features. In addition, we have supported by studies demonstrating that under drug selection identified a limited subset of imaging biomarkers that could be pressure, a similar dynamic is observed in the majority of individual easily implemented and computed on any laptop. The study was not lesions within the same tumor site as well as in lesions with different designed to evaluate how various time intervals alter feature selec- anatomic locations (49). The incremental value of imaging surro- tion and classification. However, delta features identified at 3 weeks gatesoftumorheterogeneityandthekineticsorratesoftheir (NCT00588445) were generalizable at 8 weeks (NCT01642004, evolution should therefore be investigated prospectively as potential NCT01721759). candidates to guide precision medicine approaches in systemic The radiomics signatures were applicable in selected patients with therapies with unconventional patterns of response and progres- measurable lung tumors reaching predefined clinical and imaging sion. Pilot results support our findings in the field of immune quality criteria at baseline and at first CT evaluation (Fig. 1). The therapies because lung tumor heterogeneity on contrast-enhanced output to be predicted by radiomics signatures was PFS rather than OS CT scan has been linked to OS (50) and immune contexture (51). because OS suffered from limitations such as unbalanced number of

OF10 Clin Cancer Res; 2020 CLINICAL CANCER RESEARCH

Radiomics Signatures in NSCLC

events (nivolumab: 33 deaths occurring with a median follow-up of Myers Squibb. A. Roy is an employee for and holds ownership interest (including 8.6 months), the occurrence of crossovers with new treatment lines in patents) in Bristol-Myers Squibb. L.H. Schwartz is a paid advisory board member for Roche and Novartis, and reports receiving commercial research grants from Merck patients experiencing disease progression (nivolumab and docetaxel), fl fi and Boehringer Ingelheim. No potential con icts of interest were disclosed by the and the excellent outcome of resectable early stage NSCLC (ge tinib). other authors. Perhaps more or different information would be obtained with evaluations earlier than 8 weeks (nivolumab and docetaxel), as demonstrated by the value of a 3-week evaluation (gefitinib). Authors’ Contributions In conclusion, this study is a proof of concept that AI support Conception and design: L. Dercle, M. Fronheiser, S. Du, W. Hayes, D.K. Leung, could provide clinicians an early indication of the likelihood of A. Roy, A.T. Fojo, L.H. Schwartz, B. Zhao Development of methodology: L. Dercle, M. Fronheiser, L. Lu, S. Du, W. Hayes, success of treatment with the new generation of systemic anticancer D.K. Leung, A. Roy, A.T. Fojo, L.H. Schwartz, B. Zhao therapies using conventional imaging techniques. Computers Acquisition of data (provided animals, acquired and managed patients, provided excelled in mining and integrating large amounts of data from a facilities, etc.): L.H. Schwartz, B. Zhao quantitative CT analysis of a single lung lesion segmented by an Analysis and interpretation of data (e.g., statistical analysis, biostatistics, expert radiologist. Using this data, early changes in tumor imaging computational analysis): L. Dercle, M. Fronheiser, L. Lu, S. Du, D.K. Leung, phenotype on standard-of-care CT scan were translated into a A. Roy, J. Wilkerson, A.T. Fojo, L.H. Schwartz, B. Zhao Writing, review, and/or revision of the manuscript: L. Dercle, M. Fronheiser, L. Lu, quantitative and synthetic signature to predict treatment sensitivity. S. Du, W. Hayes, D.K. Leung, A. Roy, A.T. Fojo, L.H. Schwartz, B. Zhao Treatment sensitivity was associated with changes in the interface Administrative, technical, or material support (i.e., reporting or organizing data, between lung tumor and normal lung parenchyma, as well as constructing databases): D.K. Leung, P. Guo, L.H. Schwartz, B. Zhao heterogeneity of the lung tumor. Once further prospectively vali- Study supervision: L. Dercle, D.K. Leung, A.T. Fojo, L.H. Schwartz, B. Zhao dated, these signatures could be used clinically to enhance the strategic decision-making of a practicing clinical oncologist opti- Acknowledgments fi mizing precision treatment. Consequently, AI-generated on- Authors acknowledge nancial support from the NIH (U01 CA225431) and Bristol-Myers Squibb. L. Dercle's work was partially funded by grants from treatment signatures could allow for more accurate treatment Fondation Philanthropia and Fondation Nuovo-Soldati. The content is solely decision-making which could constitute the basis for the imple- the responsibility of the authors and does not necessarily represent the funding mentation of adapted treatment guided by quantitative CT scan sources. interpretation in patients with NSCLC treated with systemic therapies. The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance Disclosure of Potential Conflicts of Interest with 18 U.S.C. Section 1734 solely to indicate this fact. M. Fronheiser is an employee for Bristol-Myers Squibb. S. Du is an employee for Bristol-Myers Squibb. W. Hayes is an employee for and holds ownership interest Received September 7, 2019; revised November 27, 2019; accepted January 22, 2020; (including patents) in Bristol-Myers Squibb. D.K. Leung is an employee for Bristol- published first March 20, 2020.

References 1. Yamamoto S, Korn RL, Oklu R, Migdal C, Gotway MB, Weiss GJ, et al. ALK 10. Shi Z, Zheng X, Shi R, Song C, Yang R, Zhang Q, et al. Radiological and clinical molecular phenotype in non-small cell lung cancer: CT radiogenomic charac- features associated with epidermal growth factor receptor mutation status of terization. Radiology 2014;272:568–76. exon 19 and 21 in lung adenocarcinoma. Sci Rep 2017;7:364. 2. Kadota K, Yeh YC, D'Angelo SP, Moreira AL, Kuk D, Sima CS, et al. Associations 11. Hsu JS, Huang MS, Chen CY, Liu GC, Liu TC, Chong IW, et al. Correlation between mutations and histologic patterns of mucin in lung adenocarcinoma: between EGFR mutation status and computed tomography features in patients invasive mucinous pattern and extracellular mucin are associated with KRAS with advanced pulmonary adenocarcinoma. J Thorac Imaging 2014;29:357–63. mutation. Am J Surg Pathol 2014;38:1118–27. 12. Ozkan E, West A, Dedelow JA, Chu BF, Zhao W, Yildiz VO, et al. CT gray-level 3. Kim TJ, Lee CT, Jheon SH, Park JS, Chung JH. Radiologic characteristics of texture analysis as a quantitative imaging biomarker of epidermal growth factor surgically resected non-small cell lung cancer with ALK rearrangement or EGFR receptor mutation status in adenocarcinoma of the lung. AJR Am J Roentgenol mutations. Ann Thorac Surg 2016;101:473–80. 2015;205:1016–25. 4. Zhou JY, Zheng J, Yu ZF, Xiao WB, Zhao J, Sun K, et al. Comparative analysis of 13. Liu Y, Kim J, Balagurunathan Y, Li Q, Garcia AL, Stringﬁeld O, et al. Radiomic clinicoradiologic characteristics of lung adenocarcinomas with ALK rearrange- features are associated with EGFR mutation status in lung adenocarcinomas. ments or EGFR mutations. Eur Radiol 2015;25:1257–66. Clin Lung Cancer 2016;17:441–8.e6. 5. Yang Y, Yang Y, Zhou X, Song X, Liu M, He W, et al. EGFR L858R mutation is 14. Yoon HJ, Sohn I, Cho JH, Lee HY, Kim JH, Choi YL, et al. Decoding tumor associated with lung adenocarcinoma patients with dominant ground-glass phenotypes for ALK, ROS1, and RET fusions in lung adenocarcinoma using a opacity. Lung Cancer 2015;87:272–7. radiomics approach. Medicine (Baltimore) 2015;94:e1753. 6. Liu Y, Kim J, Qu F, Liu S, Wang H, Balagurunathan Y, et al. CT features 15. Wu W, Parmar C, Grossmann P, Quackenbush J, Lambin P, Bussink J, et al. associated with epidermal growth factor receptor mutation status in patients Exploratory study to identify radiomics classiﬁers for lung cancer histology. with lung adenocarcinoma. Radiology 2016;280:271–80. Front Oncol 2016;6:71. 7. Rizzo S, Petrella F, Buscarino V, De Maria F, Raimondi S, Barberis M, et al. CT 16. Liu F, Zhao B, Krug LM, Ishill NM, Lim RC, Guo P, et al. Assessment of therapy radiogenomic characterization of EGFR, K-RAS, and ALK mutations in non- responses and prediction of survival in malignant pleural mesothelioma through small cell lung cancer. Eur Radiol 2016;26:32–42. computer-aided volumetric measurement on computed tomography scans. 8. Lee HJ, Kim YT, Kang CH, Zhao B, Tan Y, Schwartz LH, et al. J Thorac Oncol 2010;5:879–84. Epidermal growth factor receptor mutation in lung adenocarcinomas: 17. Zhao B, Oxnard GR, Moskowitz CS, Kris MG, Pao W, Guo P, et al. A pilot study relationship with CT characteristics and histologic subtypes. Radiology of volume measurement as a method of tumor response evaluation to aid 2013;268:254–64. biomarker development. Clin Cancer Res 2010;16:4647–53. 9. Choi CM, Kim MY, Hwang HJ, Lee JB, Kim WS. Advanced adenocarcinoma of 18. Chow DS, Qi J, Guo X, Miloushev VZ, Iwamoto FM, Bruce JN, et al. Semi- the lung: comparison of CT characteristics of patients with anaplastic lymphoma automated volumetric measurement on postcontrast MR imaging for analysis of kinase gene rearrangement and those with epidermal growth factor receptor recurrent and residual disease in glioblastoma multiforme. AJNR Am J Neuror- mutation. Radiology 2015;275:272–9. adiol 2014;35:498–503.

AACRJournals.org Clin Cancer Res; 2020 OF11

Dercle et al.

19. Chang K, Zhang B, Guo X, Zong M, Rahman R, Sanchez D, et al. Multimodal 36. Lambin P, Leijenaar RT, Deist TM, Peerlings J, de Jong EE, van Timmeren J, et al. imaging patterns predict survival in recurrent glioblastoma patients treated with Radiomics: the bridge between medical imaging and personalized medicine. bevacizumab. Neuro-oncol 2016;18:1680–7. Nat Rev Clin Oncol 2017;14:749–62. 20. Ha R, Mema E, Guo X, Mango V, Desperito E, Ha J, et al. Three-dimensional 37. Dercle L, Lu L, Lichtenstein P, Yang H, Wang D, Zhu J, et al. Impact of variability quantitative validation of breast magnetic resonance imaging background in portal venous phase acquisition timing in tumor density measurement and parenchymal enhancement assessments. Curr Probl Diagn Radiol 2016;45: treatment response assessment: metastatic colorectal cancer as a paradigm. 297–303. JCO Clin Cancer Inform 2017:1–8. 21. HaR,MemaE,GuoX,MangoV,DesperitoE,HaJ,etal.Quantitative3D 38. Zhao B, Tan Y, Tsai W-Y, Qi J, Xie C, Lu L, et al. Reproducibility of radiomics for breast magnetic resonance imaging fibroglandular tissue analysis and cor- deciphering tumor phenotype with imaging. Sci Rep 2016;6:23428. relation with qualitative assessments: a feasibility study. Quant Imaging Med 39. Huang Q, Lu L, Dercle L, Lichtenstein P, Li Y, Yin Q, et al. Interobserver Surg 2016;6:144–50. variability in tumor contouring affects the use of radiomics to predict mutational 22. Koshkin VS, Bolejack V, Schwartz LH, Wahl RL, Chugh R, Reinke DK, et al. status. J Med Imaging 2018;5:011005. Assessment of imaging modalities and response metrics in Ewing sarcoma: 40. Tan Y, Schwartz LH, Zhao B. Segmentation of lung lesions on CT scans using correlation with survival. J Clin Oncol 2016;34:3680–5. watershed, active contours, and Markov random field. Med Phys 2013;40: 23. Aerts HJ, Grossmann P, Tan Y, Oxnard GG, Rizvi N, Schwartz LH, et al. Defining 043502. a radiomic response phenotype: a pilot study using targeted therapy in NSCLC. 41. Breiman L. Random forests. Mach learn 2001;45:5–32. Sci Rep 2016;6:33860. 42. Wilkerson J, Abdallah K, Hugh-Jones C, Curt G, Rothenberg M, Simantov R, 24. Coroller TP, Agrawal V, Narayan V, Hou Y, Grossmann P, Lee SW, et al. et al. Estimation of tumour regression and growth rates during treatment in Radiomic phenotype features predict pathological response in non-small cell patients with advanced prostate cancer: a retrospective analysis. Lancet Oncol lung cancer. Radiother Oncol 2016;119:480–6. 2017;18:143–54. 25. Mattonen SA, Palma DA, Johnson C, Louie AV, Landis M, Rodrigues G, et al. 43. Seymour L, Bogaerts J, Perrone A, Ford R, Schwartz LH, Mandrekar S, et al. Detection of local cancer recurrence after stereotactic ablative radiation therapy iRECIST: guidelines for response criteria for use in trials testing immunother- for lung cancer: physician performance versus radiomic assessment. Int J Radiat apeutics. Lancet Oncol 2017;18:e143–e52. Oncol Biol Phys 2016;94:1121–8. 44. Dercle L, Lu L, Schwartz LH, Qian M, Tejpar S, Eggleton P, et al. Radiomics 26. Cunliffe A, Armato SG 3rd, Castillo R, Pham N, Guerrero T, Al-Hallaq HA. Lung response signature for identification of metastatic colorectal cancer sensitive to texture in serial thoracic computed tomography scans: correlation of radiomics- therapies targeting EGFR pathway. JNCI: Journal of the National Cancer based features with radiation therapy dose and radiation pneumonitis devel- Institute, 2020. opment. Int J Radiat Oncol Biol Phys 2015;91:1048–56. 45. Li Y, Lu L, Xiao M, Dercle L, Huang Y, Zhang Z, et al. CT slice thickness and 27. Fried DV, Tucker SL, Zhou S, Liao Z, Mawlawi O, Ibbott G, et al. Prognostic value convolution kernel affect performance of a radiomic model for predicting EGFR and reproducibility of pretreatment CT texture features in stage III non-small status in non-small cell lung cancer: a preliminary study. Sci Rep 2018;8:17913. cell lung cancer. Int J Radiat Oncol Biol Phys 2014;90:834–42. 46. Sun R, Limkin EJ, Vakalopoulou M, Dercle L, Champiat S, Han SR, et al. A 28. Coroller TP, Grossmann P, Hou Y, Rios Velazquez E, Leijenaar RT, Hermann G, radiomics approach to assess tumour-infiltrating CD8 cells and response to anti- et al. CT-based radiomic signature predicts distant metastasis in lung adeno- PD-1 or anti-PD-L1 immunotherapy: an imaging biomarker, retrospective carcinoma. Radiother Oncol 2015;114:345–50. multicohort study. Lancet Oncol 2018;19:1180–91. 29. Huang Y, Liu Z, He L, Chen X, Pan D, Ma Z, et al. Radiomics signature: a 47. Limkin E, Sun R, Dercle L, Zacharaki E, Robert C, Reuze S, et al. Promises and potential biomarker for the prediction of disease-free survival in early-stage (I or challenges for the implementation of computational medical imaging (radio- II) non-small cell lung cancer. Radiology 2016;281:947–57. mics) in oncology. Ann Oncol 2017;28:1191–206. 30. Emaminejad N, Qian W, Guan Y, Tan M, Qiu Y, Liu H, et al. Fusion of 48. Hanahan D, Weinberg RA. Hallmarks of cancer: the next generation. Cell 2011; quantitative image and genomic biomarkers to improve prognosis assessment of 144:646–74. early stage lung cancer patients. IEEE Trans Biomed Eng 2016;63:1034–43. 49. Terranova N, Girard P, Ioannou K, Klinkhardt U, Munafo A. Assessing similarity 31. Oxnard GR, Zhao B, Sima CS, Ginsberg MS, James LP, Lefkowitz RA, et al. among individual tumor size lesion dynamics: the CICIL methodology. Variability of lung tumor measurements on repeat computed tomography scans CPT Pharmacometrics Syst Pharmacol 2018;7:228–36. taken within 15 minutes. J Clin Oncol 2011;29:3114–9. 50. Trebeschi S, Kurilova I, Calin AM, Lambregts DMJ, Smit EF, Aerts H, et al. 32. Zhao B, James LP, Moskowitz CS, Guo P, Ginsberg MS, Lefkowitz RA, Radiomic biomarkers for the prediction of immunotherapy outcome in patients et al. Evaluating variability in tumor measurements from same-day repeat with metastatic non-small cell lung cancer. J Clin Oncol 2017;35:e14520. CT scans of patients with non-small cell lung cancer. Radiology 2009;252: 51. Grossmann P, Stringfield O, El-Hachem N, Bui MM, Rios Velazquez E, Parmar 263–72. C, et al. Defining the biological basis of radiomic phenotypes in lung cancer. Elife 33. Eisenhauer EA, Therasse P, Bogaerts J, Schwartz LH, Sargent D, Ford R, et al. 2017;6. New response evaluation criteria in solid tumours: revised RECIST guideline 52. Grove O, Berglund AE, Schabath MB, Aerts HJ, Dekker A, Wang H, et al. (version 1.1). Eur J Cancer 2009;45:228–47. Quantitative computed tomographic descriptors associate tumor shape com- 34. Planchard D, Popat S, Kerr K, Novello S, Smit EF, Faivre-Finn C, et al. Metastatic plexity and intratumor heterogeneity with prognosis in lung adenocarcinoma. non-small cell lung cancer: ESMO Clinical Practice Guidelines for diagnosis, PLoS One 2015;10:e0118261. treatment and follow-up. Ann Oncol 2019;30:863–70. 53. Al-Kadi OS, Watson D. Texture analysis of aggressive and nonaggressive lung 35. Obuchowski NA. ROC analysis. Am J Roentgenol 2005;184:364–72. tumor CE CT images. IEEE Trans Biomed Eng 2008;55:1822–30.

OF12 Clin Cancer Res; 2020 CLINICAL CANCER RESEARCH

Identification of Non−Small Cell Lung Cancer Sensitive to Systemic Cancer Therapies Using Radiomics

Laurent Dercle, Matthew Fronheiser, Lin Lu, et al.

Clin Cancer Res Published OnlineFirst March 20, 2020.

Updated version Access the most recent version of this article at: doi:10.1158/1078-0432.CCR-19-2942

Supplementary Access the most recent supplemental material at: Material http://clincancerres.aacrjournals.org/content/suppl/2020/07/07/1078-0432.CCR-19-2942.DC1

E-mail alerts Sign up to receive free email-alerts related to this article or journal.

Reprints and To order reprints of this article or to subscribe to the journal, contact the AACR Publications Subscriptions Department at [email protected].

Permissions To request permission to re-use all or part of this article, use this link http://clincancerres.aacrjournals.org/content/early/2020/03/16/1078-0432.CCR-19-2942. Click on "Request Permissions" which will take you to the Copyright Clearance Center's (CCC) Rightslink site.