Feasibility study of differentiating autoimmune pancreatitis from pancreatic cancer based on imaging omics-machine learning
Objective To study the value of imaging omics-machine learning in differentiating autoimmune pancreatitis (AIP) from pancreatic cancer (PC). Methods This study selected 60 patients with AIP and 60 patients with PC as the research subjects. Magnetic resonance imaging (MRI) was performed in all patients,and enhanced scans were performed after intravenous injection of the contrast agent gadopentetate dimeglumine injection into the elbow,including arterial,portal,and delayed phase scanning. Then,according to the requirement of 7:3 random ratio,the cases were divided into the training group (84 cases) and the verification group (36 cases),and 5 machine learning models using Logistic regression (LR),support vector machine (SVM),decision tree (DT),random forest (RF) and extreme gradient boosting (XGBoost) were established. Area under receiver operating characteristic (ROC) curve (AUC),accuracy,precision,sensitivity,specificity,F1 value and Brier score were used to evaluate the performance of each model. Results Among the 3 stages of arterial phase,portal phase and delayed phase,the training group and verification group had the highest AUC,sensitivity and accuracy than those of arterial phase and delayed phase,but the specificity and Youden's index of arterial phase was the highest. Among the LR,SVM,DT,RF and XGBoost machine learning models,the AUC (0.924),accuracy (82.5％),specificity (76.4％) and F1 value (0.806) of RF in the training group were the highest,and the Brier score was the lowest (0.46);DT has the highest accuracy (88.6％). In the verification group,XGBoost had the highest AUC (0.764),accuracy (76.4％),accuracy (82.2％),specificity (70.5％) and F1 value (0.712),while DT had the lowest Brier score (0.21). Conclusion In the differentiation of AIP and PC,MRI can make a more accurate differential diagnosis from the signs and signals of the two,but it is difficult to distinguish the superficial signs and signals. The imaging omics model can better reflect the difference between the two diseases,and it is more significant in the portal phase,which is helpful to improve the differential diagnosis effect of the two diseases.
Magnetic resonance imagingImaging omicsAutoimmune pancreatitisPancreatic cancerDifferential diagnosis
万方数据
维普
