Establishment Of An Imaging-tumor Marker Model To Predict The Short-term Efficacy Of Bronchial Arterial Chemoembolization Combined With ¹²⁵I Seed Implantation In The Treatment Of Non-small Cell Lung Cancer

Objective The main purpose of this study was to investigate whether serological or histological tumor markers and radiomic features of lesions in patients with non-small cell lung cancer（NSCLC）can be used as influential factors of short-term efficacy of bronchial arterial chemoembolization（BACE）combined with ¹²⁵I seed implantation therapy,and the predictive value for the teratment efficacy was further confirmed by establishing an imaging-tumor marker model.Methods Preoperative analysis of clinical,image and tumor marker data of 80 patients with NSCLC who were pathologically confirmed and treated with BACE combined with¹²⁵I seed implantation in our hospital from January 2015 to December 2021 was underwent.The patients were divided into progression group and non-progression group according to the efficacy response after 3 months of teratment.Firstly,the radiomic features of lesions with region of interest（ROI）were extracted based on the arterial phase images of chest-enhanced CT in the modeling group（60 cases）,and the predictive radiomic features were screened by LASSO regression combined with 5-fold cross-validation.The radiomics score were built by such predictive radiomic features,and then multivariate Logistic regression was used to screen independent effect factor from the clinical characteristics and tumor markers.The radiomics score,tumor markers,and clinical characteristics,as efficacy predictors,were incorporated into multivariate Logistic regression to establish an imaging-tumor marker model,and a nomogram prediction tool that was convenient for clinical using was produced.In the validation group（20 cases）,the receiver operating characteristic curve（ROC）,AUC value,and the calibration curve were used to evaluate the efficacy and stability of the model in predicting the teratment efficacy of BACE combined with ¹²⁵I seed implantation.Results After 3 months of BACE combined with ¹²⁵I seed implantation in 80NSCLC patients,the ORR was 42.5%,and the DCR was 73.8%.With LASSO r egression analysis,in the modeling group cohort,3 predictive features were scre ened from 234 radiomic features that could be extracted stably and repeatedly（E xponential＿Glszm＿Size Zone Non Uniformity,Logarithm＿Firstorder＿Minimum,Wavel et-HLH＿Firstorder＿Mean）and the radiomics score was built.After multivariate L ogistic regression analysis,Ki-67 and CYFRA21-1 could be used as independent effect factors,but clinical characteristics were not correlated with the effect.The established imaging-tumor marker models including radiomics score,Ki-67 and C YFRA21-1 had AUCs of 0.94(sensitivity:87.5%,specificity:88.6 in the modeli ng group and 0.88（sensitivity:80.0%,specificity:93.3%in the validation group）for predicting the short-term efficacy of treatment,and its calibration curve can s how that the model has a good and stable goodness of fit.Conclusion Through this study,we found that related radiomics features,Ki-67 and CYFRA21-1 can be used as independent short-term efficacy influencing factors of teratment for NSCLC patients receiving BACE combined with ¹²⁵I seed implantation.And the imaging-tumor marker model has a good prediction ability and well clinical value for treatment efficacy prediction.