Pretreatment CT-Based Radiomics Signature as a Potential Imaging Biomarker for Predicting the Expression of PD-L1 and CD8+TILs in ESCC
Qiang Wen,Zhe Yang,Jian Zhu,Qingtao Qiu,Honghai Dai,Alei Feng,Ligang Xing
DOI: https://doi.org/10.2147/OTT.S261068
IF: 4
2020-11-20
OncoTargets and Therapy
Abstract:Qiang Wen, 1 Zhe Yang, 1 Jian Zhu, 2 Qingtao Qiu, 2 Honghai Dai, 1 Alei Feng, 1 Ligang Xing 3 1 Department of Oncology, Shandong Provincial Hospital Affiliated to Shandong University, Shandong University, Jinan 250021, People's Republic of China; 2 Department of Radiation Physics, Shandong Cancer Hospital and Institute, Shandong First Medical University, Jinan 250117, People's Republic of China; 3 Department of Radiation Oncology, Shandong Cancer Hospital and Institute, Shandong First Medical University, Jinan 250117, People's Republic of China Correspondence: Zhe Yang Department of Oncology, Shandong Provincial Hospital Affiliated to Shandong University, Shandong University, Jinan, Shandong 250021, People's Republic of China Tel +86-18560222576 Email sdslyyyz@126.com Background: The present study constructed and validated models to predict PD-L1 and CD8+TILs expression levels in esophageal squamous cell carcinoma (ESCC) patients using radiomics features and clinical factors. Patients and Methods: This retrospective study randomly assigned 220 ESCC patients to a discovery dataset (n= 160) and validation dataset (n= 60). A total of 462 radiomics features were extracted from the segmentation of regions of interest (ROIs) based on pretreatment CT images of each patient. The LASSO algorithm was applied to reduce the dimensionality of the data and select features. A multivariable logistic regression analysis was adopted to build radiomics signatures. Receiver operating characteristic (ROC) curve analysis was performed to evaluate the predictive accuracy of these models. Results: There was no significant difference between the training and validation datasets for any clinical factors in patients with ESCC. The PD-L1 expression level correlated with the differentiation degree ( p = 0.011) and tumor stage ( p = 0.032). Smoking status ( p = 0.043) and differentiation degree ( p = 0.025) were associated with CD8+TILs expression levels. The radiomics signatures achieved good performance in predicting PD-L1 and CD8+TILs with AUCs= 0.784 and 0.764, respectively. The combined model showed a favorable predictive ability compared to radiomics signatures or clinical factors alone and improved the AUCs from 0.669 to 0.871 for PD-L1 and from 0.672 to 0.832 for CD8+TILs. These results were verified in the validation dataset with the AUCs of 0.817 and 0.795, respectively. Conclusion: CT-based radiomics features have a potential value for classifying patients according to PD-L1 and CD8+TILs expression levels. The combination of clinical factors and radiomics signatures significantly improved the predictive performance in ESCC. Keywords: radiomics features, PD-L1, CD8+TILs, esophageal squamous cell carcinoma, computed tomography
oncology,biotechnology & applied microbiology