| Objective: In this study,the radiomics was applied to find the texture characteristic parameters of pancreatic CT in patients with type 2 diabetes,and the relationship between RFT-score values of different anatomical regions and clinical indicators was discussed by using spatial clustering method by applying the texture characteristic labeling technique of pancreatic imaging,so as to determine the correlation between different regional heterogeneity and clinical indicators.To obtain new biomarkers for pancreatic bioimaging.Methods: In this retrospective study,which was approved by the Ethics Review Committee,CT images of the upper and middle abdomen of 5669 patients treated in the Department of Internal Medicine of the affiliated Hospital of Yunnan University from July 2017 to December 2021 were collected.Then 196 patients with type 2 diabetes and 62 non-diabetic patients were selected as subjects.The texture features of the whole pancreatic CT images of all subjects were extracted by LIFEx7.2.0 software,and all the subjects were randomly divided into training group and test group according to the proportion of 7:3.After normalization,data dimensionality reduction,feature selection and other processing,non-zero feature parameters of pancreatic texture in CT plain scan sequence were selected.Six machine learning models are established by using support vector machine,linear discriminant analysis,logical regression,LASSO logical regression,random forest and decision tree.The performance of different model is evaluated by ROC curve analysis,and the area under the ROC curve is calculated to select the optimal model.Finally,the texture feature parameters with the highest AUC value in the test group were obtained.Then,these parameters were used to cluster the texture features of pancreatic images of patients with type 2diabetes mellitus and calculate the RTF-score score.The score results were clustered into two groups,and the effects of clinical indicators on the whole pancreas and different parts of the head,body and tail of patients with type 2diabetes mellitus were analyzed by multiple logistic regression.Results: Through the screening of radiomics process,the logical regression model based on seven characteristic parameters can obtain the highest AUC in the test group.After spatial clustering of pancreatic texture features and calculation of RTF-score in patients with type 2 diabetes mellitus,in the logistic regression multivariate analysis of RTF-score and clinical indicators,gender is the influencing factor of the overall high heterogeneity score of pancreas in patients with type 2 diabetes mellitus,and female is the risk factor,with the OR value of 15.51(1.55-439.98,p=0.046).Peripheral neuropathy and total cholesterol were the risk factors for the heterogeneity score of pancreatic head,the OR values were 3.47(1.43-9.60,p=0.010)and1.33(1.06-1.67,p=0.012),respectively.Age,drinking history and cataract are the influencing factors of the pancreatic body high heterogeneity score,and the OR values are 1.04(1.01-1.08,p=0.012),3.14(1.27-7.73,p=0.012)and0.40(0.17-0.89,p=0.029),respectively.Female and peripheral neuropathy are the risk factors of high heterogeneity score in pancreatic tail area,while family history and BMI are the protective factors,with OR values of 3.13(1.47-6.88,p=0.004),5.40(1.98-17.16,p=0.002),0.42(0.19-0.89,p=0.027)and0.89(0.80-0.98,p=0.021),respectively.Conclusion: The pancreatic texture parameters obtained by CT plain scan sequence can be used to distinguish type 2 diabetes mellitus from normal controls.Type 2 diabetic pancreatic CT imaging group texture feature spatial clustering score can capture the correlation between pancreatic tissue heterogeneity and clinical features in different anatomical regions of diabetic patients. |
PDF Full Text Request