| Objective:The purpose of this study is to explore the feasibility and efficacy of CT radiomics in aiding the clinical management of endoscopic ultrasound-guided fine-needle aspiration biopsy(EUS-FNA)for the diagnosis of solid pancreatic lesions.Methods:With the approval of the hospital ethics committee,the study retrospectively included 201 patients with resectable pancreatic adenocarcinoma(PDAC)and 54 patients with mass-forming pancreatitis(MFP);21 patients with pancreatic intraepithelial neoplasia(Pan IN)and 50 patients with chronic pancreatitis(CP);184 patients with PDAC and 37 patients with other solid pancreatic tumors who underwent EUS-FNA before surgery at our hospital in different periods from January 2015 to December 2021.The regions of interest were manually segmented on the CT arterial and venous phases and the radiomics features were extracted.The t-test and/or least absolute shrinkage selection operator regression operator(LASSO)and principal component analysis(PCA)were used for dimensionality reduction.Support vector machine(SVM),K-nearest neighbor classification(KNN),and deep neural network algorithm(DNN),etc.were used to construct the radiomics models.The performance of each model was evaluated using receiver operating characteristic(ROC)curves,precision-recall curves(PRC),and decision curve analysis(DCA).The interpretability analysis of the radiomics models was performed based on the feature coefficients,SHapley Additive explanation(SHAP)algorithm or Integrated Gradients algorithm.Results:The CT radiomics was effective in differentiating resectable PDAC from MFP and could further improve the diagnostic efficacy of the models when combined with clinical variables;the clinical-radiomics combined model(PCACli Model)had comparable diagnostic performance with EUS-FNA in the validation cohort(AUC=0.880),and provided higher net benefits than that of EUS-FNA.Both the CT radiomics-based SVM model and the KNN model can effectively differentiate Pan IN from CP in the validation set(AUCSVM=0.769,AUCKNN=0.738)and could provide net benefits to patients.The CT radiomics-based DNN model was effective in distinguishing the pancreatic cancer lesions prone to false-negative EUS-FNA results(AUC=0.745)and providing net benefits in the decision curve analysis.Conclusions:CT radiomics can assist in the management of EUS-FNA for patients with solid pancreatic lesions.The radiomics-clinical model showed comparable performance with EUS-FNA in discriminating resectable PDAC from MFP.CT radiomics can help to identify Pan IN to avoid missing diagnoses which need further investigation such as EUS-FNA.And CT radiomics can quantitatively and objectively alert the endoscopists to which solid pancreatic lesions may be more likely to produce false negative results before EUS-FNA.Part Ⅰ: Is the radiomics-clinical combined model helpful in distinguishing between resectable pancreatic cancer and mass-forming pancreatitis?Objective: To explore the utility of CT radiomics in distinguishing between resectable pancreatic ductal adenocarcinoma(PDAC)and mass-forming pancreatitis(MFP)so as to reduce the need of EUS-FNA in patients with resectable PDAC.Methods: 201 patients with resectable PDAC and 54 patients with MFP from January 2015 to August 2020 in our hospital were included.Development cohort: patients without preoperative EUS-FNA(175 PDAC cases,38 MFP cases),validation cohort: patients with EUS-FNA(26 PDAC cases,16 MFP cases).Based on the radiomics features selected via t-test,two radiomic signatures(LASSOscore,PCAscore)were developed by the LASSO and principal component analysis,respectively.The LASSOCli model and PCACli model were established by combining CT radiomic signatures with clinical features including age,CA19-9 and the double-duct sign.ROC curves,precision-recall curves and decision curve analysis was performed to evaluate the utility of the model versus EUS-FNA in the validation cohort.Results: In the validation cohort,the radiomic signatures(LASSOscore,PCAscore)were both effective in distinguishing between resectable PDAC and MFP(AUCLASSO=0.743,95% CI: 0.590-0.896;AUCPCA=0.788,95% CI: 0.639-0.938),and can improve the diagnostics accuracy of the baseline only Cli Model(AUConly Cli=0.760,95% CI:0.614-0.960)after combining with variables including age,CA19-9,and the double-duct sign(AUCPCACli=0.880,95%CI: 0.776-0.983;AUCLASSOCli=0.825,95% CI:0.694-0.955).The PCACli Model showed comparable performance to FNA(AUCFNA=0.810,95% CI:0.685-0.935).In DCA,the net benefit of PCACli Model was superior to EUS-FNA,which can avoid biopsies in 70 per 1000 men at a risk threshold of 35%.Conclusions: The PCACli Model showed comparable performance with EUS-FNA,which may provide a non-invasive alternative for preoperative diagnosis and help to avoid unnecessary biopsies for resectable pancreatic masses.Part Ⅱ: The utility of CT radiomics with machine learning in distinguishing Pan IN from chronic pancreatitisObjective: To explore the value of CT radiomics with machine learning to differentiate CP from Pan IN which requires further EUS-FNA.Methods: A total of 21 cases of Pan IN and 50 cases of CP with equivocal CT diagnoses from January 2015 to December 2021 at our institution were retrospectively included and randomly divided into training and validation sets with the ratio of 3:1.The radiomics features were extracted from ROIs of the whole pancreases at AP and VP.After dimensionality reduction by t-test and LASSO,the modeling was developed by SVM,KNN and RF,respectively.The clinical model incorporating only clinical information were developed via univariate and multivariate logistic regression analysis.The efficacy of the CT radiomics-based machine learning models was evaluated using receiver operating curves,precision-recall curves and decision curve analysis.And SHapley Additive explanation(SHAP)was performed for the model interpretability.Results: In this study,the univariate analysis showed that there were no clinical variables significant in the differential diagnosis of Pan IN and CP,and the clinical model performed worse than a random classifier with an AUC<0.5 in the validation set.Both the SVM and KNN models based on CT radiomics were effective in differentiating Pan IN from CP(Training set: AUCSVM=0.860,AUCKNN=0.880;Validation set: AUCSVM=0.769,AUCKNN=0.738).The decision curve analysis showed that both models could provide net benefits for patients in the diagnosis.SHAP analysis showed that the radiomics of shape and texture contributed the most to the model diagnosis.Conclusions: The machine learning-based CT radiomics model can effectively differentiate the diagnosis of Pan IN from CP,which helps to avoid the missed diagnosis of Pan IN without further examination such as EUS.Part Ⅲ: Avoid nondiagnostic EUS-FNA: A DNN model as a possible gatekeeper to distinguish pancreatic lesions prone to inconclusive biopsyObjective: To explore the utility of CT radiomics in distinguishing the pancreatic lesions prone to nondiagnostic EUS-FNA.Methods: A total of 498 cases with PDAC who had EUS-FNA from January 2015 to October 2020 at our institution were retrospectively reviewed(Development cohort: 147 PDAC;Validation cohort: 37 PDAC).Pancreatic lesions not PDAC were also tested exploratively.Radiomics features extracted from contrast-enhanced CT was integrated with deep neural networks(DNN)after dimension reduction.The ROC curves and decision curve analysis were performed for model evaluation.And the explainability of the DNN model was analyzed by Integrated Gradients.Results: The DNN model was effective in distinguishing PDAC lesions prone to nondiagnostic EUS-FNA(Development cohort: AUC=0.821,95% CI:0.742-0.900;Validation cohort: AUC=0.745,95% CI:0.534-0.956).In all cohorts,the DNN model showed better utility than the logistic model based on traditional lesion characteristics with NRI >0(P<0.05).And the DNN model had net benefits of 21.6% at the risk threshold of 0.60 in the validation cohort.As for the model explainability,gray-level co-occurrence matrix(GLCM)features contributed the most averagely and the first-order features were the most important in the sum attribution.Conclusions: The CT radiomics-based DNN model can be a useful auxiliary tool for distinguishing the pancreatic lesions prone to nondiagnostic EUS-FNA and provide alerts for endoscopists preoperatively to reduce unnecessary EUS-FNA. |
PDF Full Text Request