Abstract
242587
Introduction: The malignancy degree of pancreatic IPMN determines its surgical methods and whether to continue other treatments after surgery.The purpose of this study is to explore the value of 18F-FDG PET/MR radiomics features in predicting malignant degree of pancreatic IPMN,thereby providing guidance for clinical treatment.
Methods: The clinical and PET/MR imaging data of 189 patients with IPMN were collected, including 76 cases of Benign, 55 cases of borderline and 58 cases of malignant. Pathological and clinical diagnosis results serve as the gold standard for diagnosis. We used AK software to extract the most relevant imageomics features for tumor classification, and randomly divided the two groups of images into training set (70%) and test set (30%). The maximum correlation and minimum redundancy (mRMR) and minimum absolute shrinkage and selection operator (LASSO) methods were used to select features from 1800 features extracted from MR and PET, and finally 9 best features were retained. Multivariate logistic regression analysis was performed using the radiomics features and clinical variables to establish the prediction model. The receiver operating characteristic (ROC) analysis is used to evaluate the prediction model.
Results: The established PET/MR imaging features have good prediction efficiency for distinguishing malignant degree of pancreatic IPMN(P<0.05). The AUC of the training group and the validation group were 0.945 (95% CI: 0.787-0.956), 0.934 (95% CI: 0.776 – 0.945).The calibration curve showed that the nomogram of radiomics had goodness of fit, and DCA proved that the nomogram of radiomics was useful in clinical practice.
Conclusions: The prediction model of PET/MR radiomics features can be used as an auxiliary method to predict the malignant degree of pancreatic IPMN. It can also provide objective basis for clinical diagnosis and individualized treatment, and may has guiding significance for clinical treatment.