Preoperative prediction of lymphovascular and visceral pleural invasion of lung adenocarcinoma based on 18F-FDG PET radiomics
Objective To evaluate the predictive value of 18F-FDG PET-based radiomics models for lymphovascular invasion(LVI)and visceral pleural invasion(VPI)in lung adenocarcinoma(LAC)pri-or to surgery.Methods Eighty-seven patients with LAC(42 males,45 females,age:(64.6±9.0)years;90 lesions)pathologically confirmed in the Affiliated Taizhou People's Hospital of Nanjing Medical Univer-sity between August 2018 and August 2022 were retrospectively included.Based on the radiomics features extracted from PET images,the machine learning models were constructed by using the support vector ma-chine(SVM),logical regression(LR),decision tree(DT),and K-nearest neighbor(KNN)algorithm.Stratified sampling(Python's StratifiedkFold function)was employed to divide the data into training set and test set at a ratio of 8∶2.The model stability was assessed using the 50％discount cross-validation.The ROC curve was drawn,and the AUC was calculated to evaluate the value of radiomics models in predicting LVI and VPI in LAC.Delong test was used to compare AUCs of different models.Results The radiomics mod-els(SVM,LR,DT,KNN)based on PET images showed good predictive value for LVI and VPI in patients with LAC.For LVI,the AUCs were 0.91,0.90,0.91,0.91 in the training set,and were 0.85,0.87,0.77,0.78 in the test set;for VPI,the AUCs were 0.86,0.86,0.84,0.81 in the training set,and were 0.82,0.80,0.69,0.78 in the test set.The F1 scores of the SVM model were the best(0.59 and 0.66 for predicting LVI and VPI respectively).The Delong test showed that there were no significant differences in AUCs among the four models(z values:from-1.46 to 1.71,all P＞0.05).Conclusions The machine learning models based on 18 F-FDG PET radiomics features are effective in predicting LVI and VPI in patients with LAC prior to surgery.These models can assist clinicians in stratifying the risk of LAC and making informed clinical de-cisions.The SVM model has the best performance in predicting LVI and VPI.
NETL
NSTL
万方数据
