1982PDIdentification of a radio-immune signature with high prognostic value in surgically resected NSCLC

• Published in: Annals of oncology Vol. 30; no. Supplement_5
• Main Authors: Mazzaschi, G, Quaini, F, Milanese, G, Gnetti, L, Bocchialini, G, Ampollini, L, Minari, R, Silva, M, Sverzellati, N, Roti, G, Tiseo, M
• Format: Journal Article
• Language: English
• Published: Oxford University Press 01-10-2019
• ISSN: 0923-7534; 1569-8041
• DOI: 10.1093/annonc/mdz269.001

Abstract
Background
The role of radiomics against clinical and histologic standard has been repeatedly demonstrated, although the integration of high-throughput imaging into a multidimensional prediction model is still in its early dawn. Thus, the aim of the present study was to determine whether CT-derived radiomic features (CT-RFs) might intercept the landscaped arrangement of the tumor immune microenvironment (TIME), offering a novel non-invasive assessment of prognostic factors in NSCLC.
Methods
A cohort of 100 (70 training and 30 validation) surgically resected NSCLC was investigated. TIME was classified according to PD-L1 and Tumor Infiltrating Lymphocytes (TILs) levels and further defined as hot, intermediate or cold by the relative contribution of effector and suppressor phenotypes. Extracted CT-RFs were correlated to TIME profiles and ROC curves were used to test the accuracy of the radiomic predictor. The impact of integrated radio-immune parameters on clinical outcome was estimated by Kaplan Meier method.
Results
Patient-specific tissue immune profiles were reflected by a strong correlation between CT-RFs and TIME parameters (U-Mann Whitney Test). Cluster Tendency and GrayLevelNonUniformity were upregulated in PD-L1high and TILs rich tumors (p < 0.01), while Skewness and Coarseness were correlated to PD-L1low and TILs poor (p < 005) samples, respectively. Among 13 CT extracted features distinctive of hot TIME (hCT-RFs), the most significantly upregulated were Energy, Busyness, and Entropy (p < 0.01), underlining the heterogeneous nature of inflamed NSCLC. Conversely, cold TIME-related features (cCT-RFs) showed a restricted variability being essentially confined to LowGrayLevelEmphasis (p < 0.001). When hCT-RFs and cCT-RFs were entered together in a multivariate logistic regression model, a highly specific and sensitive radiomic predictor of hot (1.00 AUC, p < 0.001) and cold (0.92 AUC, p = 0.001) TIME was revealed. Strikingly, a progressive decrease in OS and DFS (Kaplan Meier, p < 0.001) was documented, respectively, in hot, intermediate and cold NSCLC as defined by the radio-immune model.
Conclusions
Specific TIME profiles inscribe radiomic features resulting in a radio-immune signature with prognostic impact on NSCLC.
Legal entity responsible for the study
University of Parma.
Funding
University of Parma.
Disclosure
All authors have declared no conflicts of interest.