Noninvasive Prediction And Clinical Application Of Extrapancreatic Perineural Invasion In Pancreatic Cancer

Part 1: Application of CT-based perineural invasion score in pancreatic ductal adenocarcinomaObjective: To investigate the relationship between the perineural invasion score based on multidetector computed tomography(CT)and extrapancreatic perineural invasion(EPNI)in pancreatic ductal adenocarcinoma(PDAC).Methods: The clinical,radiological,and pathological data of 374 patients pathologically diagnosed as pancreatic cancer who underwent radical resection in the our hospital from March 2018 to May 2020 were analyzed retrospectively.Patients were divided into EPNI negative group(n = 111)and EPNI positive group(n = 263)based on the pathological presence of EPNI.The perineural invasion score was performed for each patient based on radiological images.Univariate and multivariate logistic regression models were used to analyze the association between the perineural invasion score based on CT and EPNI in PDAC.Finally,the area under the receiver operating characteristic curve(AUC),sensitivity,specificity,positive predictive value,negative predictive value and accuracy were used to evaluate the diagnostic efficacy of CT-based perineural invasion score in predicting EPNI.Results: There were significant statistical differences between EPNI negative group and positive group on both pathological characteristics(T stage,N stage,invasion of common bile duct,and positive surgical margin)and radiological characteristics(tumor size,vascular invasion,lymph node metastasis,perineural invasion score based on CT,pancreatic border,parenchymal atrophy,invasion of duodenum,invasion of spleen and splenic vein and invasion of common bile duct)(all P value <0.05).Univariate analysis revealed that the tumor size,vascular invasion,lymph node metastasis,perineural invasion score based on CT,pancreatic border,pancreatic atrophy,invasion of duodenum,invasion of spleen and splenic vein and invasion of common bile duct were independently associated with EPNI.Multivariate analyses revealed that the perineural invasion based on CT was an independent risk factor for EPNI in pancreatic cancer(score = 1,OR = 2.93,95% CI 1.61~5.32,P <0.001;score = 2,OR = 5.92,95% CI 2.68~13.10,P<0.001).The AUC of perineural invasion score in the diagnosis of EPNI was 0.67(95%CI 0.60 ~ 0.73),and the sensitivity,specificity,positive predictive value,negative predictive value and accuracy were 69.96%,63.06%,81.78%,46.98% and 0.68,respectively.Conclusion: The perineural invasion score based on CT was an independent risk factor for EPNI in pancreatic cancer and can be used as an evaluation indicator for preoperative prediction of EPNI in PDAC.Part2: The application of multi-layer perception network classifier based on CT in the diagnosis of extrapancreatic perineural invasion in pancreatic ductal adenocarcinomaObjective: Extrapancreatic perineural invasion(EPNI)is an important cause of positive surgical margin and early recurrence after resection of pancreatic ductal adenocarcinoma(PDAC).However,the existing imaging scores are greatly affected by subjectivity.In this study,a new scoring method,tumor-vascular contact(TVC)score,was established for the diagnosis of EPNI.Finally,based on clinical features,TVC score,deep learning and radiomics features(DLR),the DLR multi layer perception classifier was formed to diagnose EPNI before surgery.Methods: The clinical,imaging and pathological data of 690 patients with PDAC who were admitted to the our hospital from March 2016 to December 2020 were collected.According to the results of pathological EPNI,the patients were divided into EPNI positive group and EPNI negative group.The nn U-Net model was used for automatic segmentation of pancreatic tumor and surrounding blood vascular,and then the radiomics features of intratumoral and perivascular space and deep learning features of intratumoral were extracted.The least absolute shrinkage and selection operator(LASSO)logistic regression algorithm was used for feature selection.Finally,the multi-layer perceptual network classifier was used to build the clinical model and DLR model respectively.The area under the receiver operating characteristic curve(AUC),sensitivity,specificity,positive predictive value,negative predictive value and accuracy were used to evaluate the performance of the two models.Results: A total of 690 patients were enrolled,including 506 patients in the EPNI positive group and 184 patients in the EPNI negative group.The DLR prediction model showed good diagnostic performance in the training set and validation set,with AUC of0.89 and 0.88,respectively.The sensitivity,specificity,positive predictive value,negative predictive value and accuracy of the training set and the validation set were 81.10%,81.53%,92.79%,59.54%,0.81 and 76.74%,86.65%,92.59%,67.35%,0.86,respectively.Conclusion: The CT-based DLR multi-layer sensing network classifier is an accurate non-invasive tool for the diagnosis of EPNI in PDAC patients.