AUTHOR=Tong Haipeng , Sun Jinju , Fang Jingqin , Zhang Mi , Liu Huan , Xia Renxiang , Zhou Weicheng , Liu Kaijun , Chen Xiao 

TITLE=A Machine Learning Model Based on PET/CT Radiomics and Clinical Characteristics Predicts Tumor Immune Profiles in Non-Small Cell Lung Cancer: A Retrospective Multicohort Study

JOURNAL=Frontiers in Immunology

VOLUME=Volume 13 - 2022

YEAR=2022

URL=https://www.frontiersin.org/journals/immunology/articles/10.3389/fimmu.2022.859323

DOI=10.3389/fimmu.2022.859323

ISSN=1664-3224

ABSTRACT=Background The tumor immune microenvironment (TIME) phenotypes have been reported to mainly impact the efficacy of immunotherapy. Given the increasing use of immunotherapy in cancers, knowing the individual’s TIME phenotypes could help in identifying patients more likely to benefit from immunotherapy. This study aims to develop and validate a clinically practical model to predict TIME profiles using 18F-FDG PET/CT radiomics and clinical characteristics in non-small cell lung cancer (NSCLC).
Methods The RNA-seq data of 1137 NSCLC patients from TCGA cohort was analyzed. Then, 221 NSCLC patients from Daping Hospital (DPH) cohort underwent pre-therapy 18F-FDG PET/CT scans and were tested for CD8 expression. The radiomic features of PET/CT were extracted and radiomics signature was developed. We compared the predictive performance of models established by radiomics signature, clinical features, and their combination using receiver operating curves (ROCs). In addition, a nomogram based on radiomics signature score and clinical features was developed. Finally, we applied radiomics-clinical combined model to predict TIME of NSCLC patients in TCIA cohort (n = 39).
Results TCGA data showed CD8 expression could represent the TIME profiles in NSCLC. In DPH cohort, ROC analysis showed predictive performance for PET/CT radiomics model (AUC = 0.907), better than CT model (AUC = 0.861, P = 0.0314). Also, PET/CT radiomics-clinical combined model performed better (AUC = 0.932) to predict CD8 expression than PET/CT radiomics model (AUC = 0.907, P = 0.0326) or clinical model (AUC = 0.868, P = 0.0036). In TCIA cohort, the predicted CD8-high group had significantly higher immune scores and more activated immune pathways than CD8-low group (P = 0.0421).
Conclusion This study suggests the feasibility of non-invasively detecting tumor immune status in NSCLCs using a machine learning model based on combined 18F-FDG PET/CT radiomic features and clinical characteristics.