Experimental Study On Identification Of NASH In Rats By Combination Of MRS And Radiomics

Purpose:To explore the application value of combining magnetic resonance spectroscopy(MRS)and multimodality imaging data to construct a nomogram in early identification of non-alcoholic steatohepatitis(NASH)in rats with non-alcoholic fatty liver disease(NAFLD).Materials and Methods: A total of 68 male SD rats were fed with high-fat and highcholesterol diet in a clean environment with free access to food and water.In order to obtain SD rats with non-alcoholic fatty liver disease of different pathological grades,eight rats were fed with high-fat and high-cholesterol diet for six weeks and 15 rats were fed for 8,10,12 and 14 weeks,respectively.At the end of the feeding cycle,CT,MRI and MRS were performed.The corresponding traditional imaging data and quantitative data of liver fat fraction of rats were obtained,and then the rats were killed immediately.The liver tissues of rats were obtained for HE staining for pathological evaluation.The DICOM format data of rats were extracted from Picture Archiving and Communication System(PACS)of our hospital.Under the supervision and guidance of another senior deputy chief physician,a radiologist imported the data into 3D-slicer software to complete the manual delineation of the region of interest(ROI)and extract the radiomics features.Single factor analysis was used to preliminarily screen the radiomics features.Pearson or Spearman correlation analysis was performed to remove the internal redundancy of the radiomics feature data.Subsequently,10-fold cross-validation and forward stepwise logistic analysis were used to further screen the radiomics features and construct single-sequence and multiple-sequence radiomics models.Single-factor and multi-factor Logistic regression analysis was used to screen the independent risk factors for the clinical data.The clinical independent risk factors and the radiomics features were combined to construct a nomogram of radiomics.At the same time,R language programming software was used to draw the calibration curve to reflect the goodness of fit of the nomogram,and Hosmer-Lemehome test was performed.The receiver operating characteristic curve(ROC)was drawn and the sensitivity,specificity and accuracy of each model as well as the area under the ROC curve(AUC)were calculated to evaluate the efficacy of each prediction model.The Delong test was used to evaluate whether the differences in AUC between different models were statistically significant.The Decision curve analysis(DCA curve)was used to evaluate the clinical application value of each prediction model.Results:(1)univariate Logistic regression analysis showed that MRS＿PDFF of rats in NASH group was higher than that of rats in non-NASH group,and the difference was statistically significant(P < 0.001).MRS＿PDFF was included as an independent risk factor in the construction of clinical prediction model,and the clinical model had moderate prediction efficiency,with [AUC 0.859(95% CI 0.806-0.911)] in the training set and[AUC 0.857(95% CI 0.800-0.910)] in the cross-validation set.(2)851 radiomics features were extracted from each sequence of each sample.After univariate analysis,correlation analysis,10-fold cross-validation and forward stepwise logistic analysis,the remaining 3,1,2,2,2 features of each sequence(CT,T2WI-FS,T2 WI,T1WI-IP,T1WI-OP)were finally screened for the construction of single sequence radiomics model.The above 10single-sequence radiomics features were fused,and finally five key radiomics features were screened out to construct a multi-sequence radiomics model and calculate the Rad score of each sample.The prediction efficiency of multi-sequence radiomics models is better than that of single-sequence radiomics models in both training set [AUC 0.929(95% CI 0.868-0.945)] and cross-validation set [AUC 0.930(95% CI 0.868-0.945)].The T2WI-FS radiomics model had the worst prediction performance,with only [AUC 0.709(95% CI0.650-0.733)in the training set and only [AUC 0.714(95% CI 0.578-0.839)] in the cross-validation set.(3)The radiomics nomogram was constructed jointly with MRS＿PDFF and Rad score.The radiomics nomogram obtained the best prediction performance with the training set of [AUC 0.944(95% CI 0.899-0.973)] and crossvalidation set of [AUC 0.943(95% CI 0.902-0.972)].The P value tested by HosmerLemehome was 0.687,and P > 0.05 indicated that the model had good fitting degree.The decision curve(DCA)showed that when the threshold range was 0.08～1,the net benefit of nomogram was the highest.Conclusions:(1)The NAFLD rat model was successfully established after male SD rats were fed with high-fat and high-cholesterol diet for 8–14 weeks.(2)As a noninvasive tool,the nomogram constructed by MRS and radiomics based on multimodality imaging data is helpful to early identify the occurrence of NASH in NAFLD rats.(3)The predictive efficiency of radiomics nomogram is better than that of clinical model and multimodality radiomics model.