| Objective:To study the the diagnostic value of 1.5T MRI FS-T2 WI and ADC sequence image radiomics features in stage III type 2 diabetic nephropathy(DN):to build FS-T2 WI,ADC and combined image radiomics diagnostic models of type 2 diabetic patients with stage III type 2 diabetic nephropathy,aim to make early diagnosis of stage III type 2 diabetes nephropathy with radiomics.Materials and methods:A total of 88 subjects were included,including 55 patients with stage III type 2 diabetes nephropathy confirmed by puncture pathology,aged 30-68 years;There are 33 normal healthy volunteers,ranging in age from 31 to 74 years old.The general clinical data of the subjects included five indicators including age,sex,creatinine,urea and body mass index(BMI).The data were analyzed using spss 26 software,the classification data were analyzed using chi-square test,and the continuous variables were analyzed using independent sample t-test;Perform 1.5T magnetic resonance examination on all subjects and obtain axial FS-T2 WI and ADC images.Use ITK-SNAP software to complete the delineation of the right kidney of the subjects,including the renal cortex and medulla.Use the Python-based radiologic library to preliminarily extract and screen the FS-T2 WI and ADC texture features of the right kidney of the subjects in the training set and test set groups,and extract 1409 imaging features from the original FS-T2 WI and ADC images through wavelet and Gaussian Laplace filter.The images of 50 random subjects were sketched again by the same doctor at different times,and the intra-group consistency test(ICC)was performed,and the features with ICC>0.75 were retained for the next stage of analysis,and the R software and LASSO algorithm were used to finally select the high-value features with a coefficient of not 0 through multi-fold crossvalidation,The data set was divided into training set group(50 people)and test set group(38 people)by 6:4 using random stratified sampling method,and FS-T2 WI,ADC and combined image logical regression(LR)models were established and trained respectively,and then verified in the validation set.Finally,the area under the curve(AUC)of each diagnostic model was evaluated by Delong test,and the corresponding accuracy(ACC),sensitivity(SEN)and specificity(SPE)were calculated.result:In general clinical data,the three clinical characteristics of urea,creatinine and gender between diabetes nephropathy patients and normal volunteers were statistically significant(p<0.05),of which urea and creatinine were considered to have significant statistical differences(p<0.001);The imageomics model finally selected 8 features for FST2 WI modeling through multiple iterative filtering,including 1 firstorder feature and 7 texture features,and selected 4 features for ADC modeling,including 2 first-order features and 2 texture features.Finally,after fusion filtering,11 features were selected for FS-T2WI+ADC joint model modeling,including 3 first-order features and 8 texture features:FS-T2 WI,ADC The AUC values of the training set of FS-T2WI+ADC combined image logistic regression model were 0.96,0.91 and 0.98;The AUC values of the test set are 0.91,0.89 and 0.93,and the three models have good classification and diagnosis ability.The FS-T2WI+ADC combined image model has the highest diagnostic efficiency,and its AUC value is the highest in the training set and the test set.There is no significant statistical difference in the AUC values of the three models by Delong test(p>0.05).The specificity,accuracy and sensitivity of the combined model test set are 90.9 and 89.5,88.9 respectively.Conclusion:It is feasible to diagnose stage III type 2 diabetic nephropathy based on 1.5T magnetic resonance FS-T2 WI and ADC image radiomics diagnosis model,and FS-T2WI+ADC combined image radiomics diagnosis model has higher diagnostic efficiency. |
PDF Full Text Request