Objective To investigate the feasibility and value of preoperative prediction of perineural invasion(PNI)in pancreatic cancer based on enhanced CT radiomics features combined with traditional imaging features and clinical informa-tion.Methods 137 patients with pancreatic cancer confirmed by postoperative pathology were retrospectively analyzed.Among them,98 patients with PNI and 39 patients without PNI were randomly divided into training group(n=96)and vali-dation group(n=41).The 3D Slicer was used to manually delineate the tumor on the preoperative enhanced CT arteriove-nous images.Pyradiomics was used to extract the features.The minimum redundancy maximum correlation algorithm(mRMR),minimum absolute contraction and selection operator(LASSO)were used to reduce the dimension and screen the features.The independent radiomics model,clinical-traditional imaging model and fusion radiomics model were con-structed respectively in the training group,and the effectiveness of the model was verified in the validation group.ROC curve was drawn to evaluate the performance of the prediction model.Results Three,two and two radiomics features were screened out for arterial phase,venous phase and arterial phase combined with venous phase,respectively.The fusion ra-diomics model had the highest performance among the independent radiomics model,clinical-traditional imaging model and fusion radiomics model established for the three phases.The AUC values of the arterial phase,venous phase and arterial phase combined with venous phase fusion radiomics model were 0.83,0.85 and 0.80 in the training group,and 0.78,0.76 and 0.80 in the validation group,respectively.Conclusion The fusion radiomics model based on enhanced CT radiomics features combined with vascular invasion can effectively predict the occurrence of perineural invasion in pancreatic cancer before surgery,and the prediction efficiency is better than that of independent radiomics model and clinical-traditional ima-ging model.
NETL
NSTL
维普
万方数据
