Correlation Study Between The Spatial Clustering Score Of Pancreatic MRI Texture Characteristics In Type 2diabetes Mellitus And Clinical Indicators

Objective This study intends to obtain texture feature parameters of routine MRI sequences to distinguish the pancreas of type 2 diabetes group and normal control group by using Radiomics process,select the texture feature parameters of the optimal sequence for spatial labeling,and obtain RTF-score values of different anatomical parts of the pancreas of type 2 diabetes by spatial clustering method,so as to judge the correlation between the heterogeneity of different regions of the pancreas and clinical indicators,to obtain new pancreatic biomarkers.Methods A retrospective analysis was made of 1540 patients admitted to the Department of Endocrinology of the Affiliated Hospital of Yunnan University from January1,2019 to December 30,2021.Among them,50 patients with type 2 diabetes who underwent upper abdominal MRI due to non-pancreatic related diseases and 55 non diabetes controls who underwent upper abdominal MRI due to non-pancreatic related diseases were selected as the study subjects.The texture features of the whole pancreas MRI image of all the subjects were extracted by LIFEx7.3.0 software,and all patients were randomly divided into training group and testing group at 7:3 ratio.After normalization,dimensionality reduction and feature selection,six machine learning models were established,including support vector machine,linear discriminant analysis,Logistic regression,Lasso Logistic regression,random forest,and decision tree.The receiver operating characteristic(ROC)curve analysis was used to evaluate the performance of different models,calculate the area under the ROC curve(AUC)to select the optimal model,use the selected imaging parameters to carry out spatial clustering of pancreatic image texture features and calculate the RTF-score values,and then cluster the score values into two groups: high and low.Logistic regression analysis was used to analyze the correlation between the RTF-score values of different parts of the pancreas head,body,and tail in patients with type 2 diabetes and clinical indicators.Results Through the screening of radiomics process,DWI sequence had relatively high AUC value in T1 WI in-phase,T2 WI Pressure-grease and DWI sequence,and its AUC value in training group and testing group were 0.804 and 0.808 respectively.After spatial clustering of pancreatic image texture features of DWI sequence parameters of patients with type 2 diabetes and calculation of RTF-score values,the OR value of pancreatic head high heterogeneity score and drinking history was 17.92(90% CI: 2.25-239.27,p=0.013)in logistic regression multi-factor analysis of RTF-score values and clinical indicators;The OR value of pancreatic body high heterogeneity score and HDL-C was 443.63(95% CI: 6.61-120143.92,p=0.013);The OR value of pancreatic body high heterogeneity score and female was 0.08(95%:0.01-0.57,p=0.022).Conclusion The pancreatic texture feature parameters obtained from MRI commonly used sequences have certain differentiation ability between type 2diabetes group and normal control group,and DWI sequences is the best one;The spatial clustering of pancreatic MRI texture features and RTF-score in type 2 diabetes patients can capture the correlation between the heterogeneous changes of pancreatic tissue in different anatomical regions and clinical characteristics,and provide a visual method for further understanding the pancreatic pathological changes.