ABSTRACT
Objective:To evaluate the value of radiomics analysis based on enhanced MRI in predicting the recurrence of acute pancreatitis (AP).Methods:From January 2017 to December 2020, 201 patients diagnosed with AP were collected retrospectively in the Affiliated Hospital of North Sichuan Medical College. These patients underwent plain and enhanced MRI within 7 days after onset. After clinical follow-up, 102 cases were classified as non-recurrence AP group and 99 cases were classified as recurrent acute pancreatitis (RAP) group. They were divided into training set (140 cases, 71 cases in non-recurrence AP group, 69 cases in RAP group) and validation set (61 cases, 31 cases in non-recurrence AP group, 30 cases in RAP group) using a random number table method. The independent sample t-test, Mann-Whitney U test or χ 2 test were used to compare the clinical characteristics between the two groups, and the clinical characteristics with statistical differences were included in logistic regression to construct the clinical model. The quantitative features of radiomics were extracted based on the late arterial-phase images of contrast-enhanced MRI. The best radiomics features retained after dimensionality reduction were used to construct the radiomics model through logistic regression analysis, and a combined model was constructed by combining the clinical features. The prediction ability of the models was evaluated by the receiver operating characteristic curve, and the area under the curve (AUC) was compared by DeLong test. Results:There were statistical differences in gender, severity, local complications, hyperlipidemia and smoking between non-recurrence AP group and RAP group (all P<0.05). Hyperlipidemia was an independent risk factor for AP recurrence (OR=5.236, 95%CI 2.710-10.101). The 9 best radiomics features by dimensionality reduction were selected to construct a radiomics model. The AUCs of clinical model, radiomics model and combined model in the training set were 0.803, 0.944 and 0.978 respectively, and those in the validation set were 0.678, 0.940 and 0.955 respectively. In the training set and the validation set, the prediction ability of the radiomics model and combined model were higher than those of the clinical model (training set: Z=3.28, 4.83, P=0.001,<0.001; validation set: Z=3.48, 4.05, both P<0.001). Conclusions:The radiomics model based on late arterial-phase enhanced MRI has good quantitative prediction ability for the recurrence of AP, which can provide a reference for the prevention and treatment of RAP.