Validation of a radiomic approach to decipher NSCLC immune microenvironment in surgically resected patients

Background: Radiomics has emerged as a noninvasive tool endowed with the potential to intercept tumor characteristics thereby predicting clinical outcome. In a recent study on resected non-small cell lung cancer (NSCLC), we identified highly prognostic computed tomography (CT)–derived radiomic features (RFs), which in turn were able to discriminate hot from cold tumor immune microenvironment (TIME). We aimed at validating a radiomic model capable of dissecting specific TIME profiles bearing prognostic power in resected NSCLC. Methods: The validation cohort included 31 radically resected NSCLCs clinicopathologically matched with the training set (n = 69). TIME was classified in hot and cold according to a multiparametric immunohistochemical analysis involving PD-L1 score and incidence of immune effector phenotypes among tumor infiltrating lymphocytes (TILs). High-throughput radiomic features (n = 841) extracted from CT images were correlated to TIME parameters to ultimately define prognostic classes. Results: We confirmed PD-1 to CD8 ratio as best predictor of clinical outcome among TIME characteristics. Significantly prolonged overall survival (OS) was observed in patients carrying hot (median OS not reached) vs cold (median OS 22 months; hazard ratio 0.28, 95% confidence interval 0.09–0.82; p = 0.015) immune background, thus validating the prognostic impact of these two TIME categories in resected NSCLC. Importantly, in the validation setting, three out of eight previously identified RFs sharply distinguishing hot from cold TIME were endorsed. Among signature-related RFs, Wavelet-HHH_gldm_HighGrayLevelEmphasis highly performed as descriptor of hot immune contexture (area under the receiver operating characteristic curve 0.94, 95% confidence interval 0.81–1.00; p = 0.01). Conclusion: Radiomics may decipher specific TIME profiles providing a noninvasive prognostic approach in resected NSCLC and an exploitable predictive strategy in advanced cases