The Value Of ¹⁸F-FDG PET In Predicting Lymphovascular,Visceral Pleural Invasion And Ki-67 Expression In Lung Adenocarcinoma Before Operation

Objective:To explore the value of clinical parameters,maximum standardized uptake value（SUVmax）of ¹⁸F-FDG PET traditional metabolic parameters and ¹⁸F-FDG PET radiomics in predicting lymphovascular invasion（LVI）,visceral pleural invasion（VPI）and Ki-67 expression in patients with lung adenocarcinoma before treatment,in order to provide reference for individualized precision therapy.Methods:This study retrospectively analyzed the patients with LAC diagnosed by operation and pathology in Taizhou people’s Hospital from August 2018 to August 2022,including 87 patients with 90 lesions.The main results were as follows:（1）The basic clinical and PET parameters were collected,LVI,VPI and Ki-67 were analyzed by univariate and multivariate analysis,the independent influencing factors were determined,the receiver operating characteristic（ROC）analysis was performed.,the area under the curve（AUC）was calculated and the best cutoff value was obtained.（2）Randomly divide the data into training set and verification set according to 8:2,draw the region of interest（ROI）on the PET image,and extract the radiomics features.The maximum correlation minimum redundancy（mRMR）method is used to screen out the features with the highest correlation with LVI,VPI and Ki-67,respectively,and further build the radiomics models,including logical regression（LR）,K-nearest neighbor（KNN）,decision tree（DT）and support vector machine（SVM）model.The 5-fold-cross-validation method was used to verify the stability of the model performance in the training set and verification,draw the ROC curve,calculate the AUC,evaluate the predictive efficiency of the radiomics models for lung adenocarcinoma LVI,VPI and Ki-67,and compare the AUC differences between the models by Delong test.Results:1.Correlation analysis of clinical and PET basic parameters with LVI,VPI and Ki-67:（1）In univariate analysis,PET/CT tumor size and SUVmax were correlated with LVI（both P<0.005）,while in multivariate analysis,only SUVmax（P=0.005）was correlated with LVI,with AUC of 0.80 and the cutoff value of 9.85.（2）In univariate analysis,gender and SUVmax were related to the expression of Ki-67（both P<0.005）,while in multivariate analysis,only SUVmax（P=0.005）was related to the expression of Ki-67,and the AUC was 0.59,the cutoff value was 4.95.（3）Age,sex,tumor location（left or right lung lobes）,PET/CT tumor size and SUVmax were not significantly correlated with VPI.2.PET radiomics were related to LVI,VPI and Ki-67:from the 855 radiomic features extracted,10 radiomics features most related to LVI,VPI and Ki-67 were screened,and four common machine learning classification models（SVM,LR,DT,KNN）were established to evaluate the status of LVI,VPI and Ki-67respectively.（1）For LVI,the AUC of training set is 0.91,0.90,0.91,0.91 respectively,and the verification set is 0.85,0.87,0.77,0.78 respectively;the F1 score of SVM model is the best,and its AUC is higher than that of SUVmax.（2）For VPI,the AUC of training set is 0.86,0.86,0.84,0.81,and the verification set is 0.82,0.80,0.69,0.78respectively;the F1 score of SVM model is the best,but there was no significant correlation between SUVmax and VPI.（3）For Ki-67,the AUC of the training set is0.85,0.85,0.90,0.90 respectively,and the verification set is 0.79,0.80,0.74,0.75respectively;the F1 score of LR model is the best,and its AUC was significantly higher than that of SUVmax.（4）There was no significant difference in the AUC of predicting LVI,VPI and Ki-67 status of LAC among the four models by Delong test.Conclusion:1.PET/CT tumor size and SUVmax were correlated with LVI,gender and SUVmax were correlated with Ki-67,and SUVmax was an independent predictor of LVI and Ki-67.Age,sex,tumor location（left or right lung lobes）,PET/CT tumor size and SUVmax were not significantly correlated with VPI.2.The model based on¹⁸F-FDG PET radiomics features shows good predictive efficiency in predicting the state of LVI,VPI and Ki-67,and can assist clinical decision-making,and the effectiveness of the radiomics model is better than the traditional metabolic parameter SUVmax.