Evaluation Of FDG PET Sum Of Standardized Uptake Value And Its Prognostic Value By CT Radiomics

Objective:Previous studies have shown that CT radiomics is highly related to the metabolic parameters of FDG PET.Therefore,this study attempts to es tablish a functional relationship between CT radiomics and PET parameters,and evaluate the prognostic value of parameters.Methods:The study cohort included lung cancer?SMU LC,n=206?,esophageal cancer?SMU EC,n=61?,and lung cancer?RIDER LC,n=45?in the image archive center.All patients were pathologically confirmed a nd underwent ¹⁸F-FDG PET/CT scan.The lymph nodes of SMU LC were the training group,the primary focus of SMU LC,the lymph nodes of SMU EC and the lymph nodes of RIDER LC were the verification group.Each lymph node and primary focus of CT image were segmented,and the characteristics of radiology were extracted.In the training group,ln?SUVsum?was used as the dependent variable,CT features as the independent variable,and the gradient enhanced region trees?GBRT?and partial least square method?PLS?were used to screen variables and regression model coefficients,and the influence of relevant factors on the model estimates was discus sed.In addition,three validation groups can evaluate the estimation ability of the same and different time models of acquisition equipment.Finally,the prognosis value of the estimated value was evaluated by the follow-up data of SMU LC patients.Results:In the training group,141 image features?1683 variables?can be extracted from each volume of interest area.GBRT method is used to rank the contribution of each image feature.The top 20 features?n=6?between the same machine and thin-layer CT are selected.Finally,three features are selected by PLS method to build the prediction model,and then PLS and Passing&Bablok step-by-step regression model coefficients are used.There was no significant difference between the predicted value and the measured value?P=0.873?.The mean value of prediction deviation in the lymph node group of SMU EC and the primary focus of SMU LC was 0.059±0.476?r=0.896,P=0.033?and 0.479±0.557?r=0.944,P=0.000?,The system deviation of RIDER LC group increased slightly,but it could be corrected by Passing&Bablok regression.The related factors of the model have little influence on the accuracy of the model,such as:the benign and malignant lymph nodes,injection dose,collection time,lymph node volume and pathological type.Finally,to analyze the patient's survival information?n=124?,sum SUV sum of all individual lesions,and calculate the total amount of glycol ysis.Cox regression analysis showed that predicted ln?TLG?and evaluation ln?TLG?were independent prognostic factors,except for stage and treatment.HR scores were 1.78?95%CI:1.18-2.68?and 1.91?95%CI:1.27-2.87?.Conclusion:The natural logarithm of lymph node SUVum can be estimated by three CT radiomics.The model is rarely affected by the evaluation factors including metastasis or not,and the predicted value of the model has a significant predictive value for the survival of patients.