Study On The Value Of A Multimodal Radiomics Model Based On Magnetic Resonance Imaging In Staging Of Hepatic Fibrosis

Objective:To develop a radiomics-based model for staging liver fibrosis by using T2WI fat suppressed(T2WI-FS)and diffusion-weighted imaging(DWI)combined with clinical serological features,and explore the predictive efficiency of the model for the staging of hepatic fibrosis.Methods:(1)1038 patients with hepatic fibrosis who underwent segmental hepatectomy in the Affiliated Hospital of Qingdao University from February 2016 to December 2019were collected retrospectively.All patients underwent T2WI-FS and DWI scan within two weeks before operation,laboratory examination within two weeks,and liver biopsy after operation.According to the inclusion and exclusion criteria of this study,a total of 249patients met the requirements,including 179 males and 70 females.127 patients with S0-S2 stage liver fibrosis and 122 patients with S3-S4 stage liver fibrosis.(2)The T2WI-FS and DWI images of MRI were imported into ITK-snap software from PACS system.After drawing the region of interest(ROI)manually,the image features were extracted by Pyradiomics software.Then,according to the proportion of 7:3,249 patients were randomly divided into development cohort and verification cohort.The development cohort was used for feature screening and modeling,and the verification cohort was used to verify the model.(3)the clinical and serological indexes were statistically analyzed by IBM SPSS26 statistical software.Mann-Whitney U test was used for quantitative data,χ~2 test was used for qualitative data.Receiver operating curve(ROC)and area under curve(AUC)were used to evaluate the value of FIB-4、APRI、GPR、AAR.(4)R language statistical software was used to analyze the histological data,and multivariate logistic regression was used to screen out meaningful clinical-serological features.(5)Then inter-and intra-crass correlation coefficients(ICC)were used to exclude the inter-observer and intra-observer difference of feature extraction.Wilcoxon test and LASSO regression were used to select image features.Three imaging tags of T2WI-score,DWI-score and T2-DWI-score were established by Logistic regression analysis,and the best diagnostic tags were selected by ROC curve.Using Logistic regression,the best imaging labels and selected clinical-serological indicators were used to build a combined model and draw a histological nomogram.The accuracy,sensitivity,specificity,positive predictive value and negative predictive value of the variables were evaluated by ROC.The net benefit of nomograph in the diagnosis of different stages of fibrosis was analyzed by decision curve analysis(DCA).Results:(1)The results of multivariate Logistic regression showed that history of hepatitis B,PLT,GGT and LDH were independent predictors of fibrosis in S3-S4.1561 parameter features were extracted both from T2WI-FS and DWI image.After feature dimensionality reduction screening,26 and 10 features were retained in T2WI-FS and DWI,respectively.The combinatorial tag T2-DWI-score calculated from the above 36 combinatorial features had a better prediction effect than the single combinatorial tag of T2WI-FS or DWI.In the tag of T2WI-FS and DWI,the AUC of the development cohort was 0.968(95%CI:0.923-0.971)and the AUC of the verification cohort was 0.877(95%CI:0.839-0.902).(2)The nomograph established by 4 clinical-serum features and 36 radiomics tags had a good predictive effect on liver fibrosis in S3-S4 stage.In the development cohort,the final nomograph prediction performance was better than that of the clinical-serum nomograph(AUC of the final nomograph:0.981,95%CI:0.872-0.993).In the verification cohort,the final nomograph prediction performance was also higher than that of the clinical-serum nomograph(AUC of the final nomograph:0.905,95%CI:0.843-0.937).The results of DCA showed that the clinical practicability of nmograph was higher than that of clinical-serum nomograph.(3)Compared with the four serum diagnostic models of FIB-4,APRI,GPR and AAR established by serological indexes,the final nomograph had higher diagnostic performance,and the AUC was 0.895(95%CI:0.850-0.924),Delong test results showed that P<0.01.The final nomograph had high diagnostic efficacy in the diagnosis of fibrosis(≥S1),significant hepatic fibrosis(≥S2),advanced hepatic fibrosis(≥S3)and early liver cirrhosis(S4).The AUC was 0.973(95%CI:0.931-0.980),0.964(95%CI:0.925-0.971),0.895(95%CI:0.850-0.924)and 0.931(95%CI:0.905-0.951),respectively.Conclusion:The final nomograph based on T2WI-FS and DWI radiomics features combined with clinical serological features is a non-invasive diagnostic tool and has high diagnostic efficacy in the diagnosis of liver fibrosis staging.