Abstract Imaging can capture phenotypic intra- and inter-tumoral heterogeneities. Radiomics provides a radiophenotypic quantification but has mostly focused on single tumors, letting apart metastatic patients. Our aim was to develop inter-site heterogeneity indices (ISHIs) using radiomics and to investigate their predictive value regarding treatment response and overall survival (OS) in lung adenocarcinoma patients. All measurable target lesions per RECIST with volume>1cm3 from consecutive adult patients with newly-diagnosed metastatic lung adenocarcinoma were segmented on contrast-enhanced CT-scan to extract 121 3D-radiomics features (RFs). Four methods were developed to quantify the dissimilarity (Ds) between the lesion profiles from a same patient in the radiomics space, namely: Centroid-to-Lesion (CL), Lesion-to-Lesion (LL), Primitive-to-Lesion (PL) - followed by mean, range and standard deviation (SD) of those distances -, and using clustering to define lesion groupings and calculating grouping diversity with Shannon Entropy (SE). Associations between the 10 resulting ISHIs and disease control rate (DCR), objective response rate (ORR) to first-line treatment and OS were investigated in univariate and stepwise multivariate analyses including cofounding variables (performance status, sex, age, staging, number of metastatic sites, liver and bone metastases). 167 patients (62 women, median age: 62.3 years) were included between 2016 and 2019, and randomly separated into one experimental (N=117 [70%], 546 lesions) and one validation (N=50 [30%], 232 lesions) cohorts. Higher ISHIs were systematically found in patients without disease control, without objective response and with worse OS. Range(Ds-PL) was independently associated with DCR in both cohorts (odds ratio [OR]=1.30, P=0.02 and OR=1.50, P=0.02, respectively). Range(Ds-PL) was independently associated with ORR in the validation cohort (OR=1.46, P=0.049 versus OR=1.23, P=0.055 in the experimental cohort). SE was independently correlated with OS in the experimental cohort (hazard ratio [HR]=2.39, P=0.005) whereas Range(Ds-PL) was selected in the last regression step in the validation cohort (HR=1.16, P=0.08). None of the ISHIs was associated with PD-L1 status; however, trend towards lower ISHI values were observed in EGFR, ALK and ROS1 altered tumors versus the others. Radiomics-based ISHIs were associated with ORR, DCR and OS and appeared as promising metrics to deepen our understanding of the clinical implication of intra-patient inter-tumor heterogeneity. Correlation assessment between radiomics-based ISHIs and tumor heterogeneity evaluated by a baseline ctDNA profiling (FondationOne® Liquid CDx) before treatment in metastatic lung adenocarcinoma patients is ongoing and will be presented. Citation Format: Mathilde Lafon, Sophie Cousin, Mélissa Alamé, Michèle Kind, Antoine italiano, Amandine Crombé. Predictive value of baseline inter-sites heterogeneity measured by radiomics in metastatic lung adenocarcinomas [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2023; Part 1 (Regular and Invited Abstracts); 2023 Apr 14-19; Orlando, FL. Philadelphia (PA): AACR; Cancer Res 2023;83(7_Suppl):Abstract nr 863.
Read full abstract