Study On CT Radiomics-based Prediction About The Efficacy Of Anti-PD1 Immunotherapy For Non-small Cell Lung Cancer

Objective: Significant challenges remain in the decision-making for the individual treatment of non-small cell lung cancer(NSCLC)patients during the programmed cell death 1(PD1)inhibitor therapy.This study aims to explore the values of computed tomography(CT)-based Delta radiomics and subregional radiomics in predicting therapeutic response to anti-programmed cell death immunotherapy in NSCLC patients.Methods: This study was conducted in two parts.The first part of the study analyzed 44 NSCLC patients(97 lesions)who underwent pre-contrast and contrast-enhanced CT scans before and after the anti-PD1 immunotherapy between February 2017 and December 2019.The radiomics features were extracted from pre-and post-treatment CT images,and the Delta radiomics features were calculated to reflect the changes in CT images before and after treatment.Then features were selected with Mann-whitney U test,least absolute shrinkage and selection operator(LASSO)and logistic regression with akaike information criterion to build the pretreatment radiomics signature,post-treatment radiomics signature and Delta radiomics signature(Rad-Delta).The second part of the study enrolled 66 NSCLC patients from November 2017 to July 2020.The lesions identified on pretreatment CT scans were subdivided into phenotypically consistent subregions on the patient-level and population-level clustering by intratumoral segmentation.Handcrafted and deep learning-based radiomics features were extracted separately from the entire tumor region and subregions,then selected using Mann-whitney U test,LASSO and logistic regression with akaike information criterion.Based on the selected features and correlation coefficients,the handcrafted radiomics signatures(Handcrafted-RSs),deep learning radiomics signatures(Deep-RSs)and fusion radiomics signatures(Fusion-RSs)integrating these two features were constructed for each region.The area under the receiver operating characteristic(ROC)curve(AUC)was calculated to assess the RS.Results: The results of the first part of the study showed that the Rad-Delta exhibited higher AUCs in both the training(AUC = 0.907,specificity = 0.900,sensibility = 0.771)and validation(AUC = 0.827,specificity = 0.867,sensibility = 0.706)cohorts compared with the radiomics signature derived from the CT data acquired before or after the treatment separately.The results of the second part of the study showed that the lesion was finally divided into two subregions,denoted as marginal subregion 1(S1)and inner subregion 2(S2).Radiomics signatures derived from the S1 outperformed those from S2 and whole tumor area for both the handcrafted and deep learning features.The Fusion-RS derived from S1 achieved the best prediction performance in the training(AUC = 0.947,specificity = 0.895,sensibility = 0.878)and validation(AUC = 0.875,specificity = 0.724,sensibility = 0.952)cohorts.Conclusion: The first part of the study revealed the association between CT-based Delta radiomics and responses to anti-PD1 immunotherapy.The predictive performance of the Delta radiomics signature was better than the radiomics signatures using only pre-or posttreatment images.The second part of the study demonstrated that the subregional radiomics analysis based on CT images can be useful for predicting the therapeutic response to anti-PD1 immunotherapy in NSCLC.The Fusion radiomics signature established by intratumoral segmentation algorithm considering intratumoral heterogeneity achieved better predictive performance than traditional radiomics and Delta radiomics.The developed Fusion-RS may be considered as a potential non-invasive tool for individual treatment managements.