The Clinical Application Of Gd-EOB-DTPA Enhanced MRI In Quantitative Liver Function Assessment In Patients With Hepatocellular Carcinoma

Objective:1.To develop a nomogram model based on preoperative Gadolinium ethoxybenzyl-diethylenetriaminepentaacetic acid(Gd-EOB-DTPA)enhanced MRI for predicting posthepatectomy liver failure(PHLF)and to explore its clinical application value.2.To establish and internally validate the radiomics model based on Gd-EOB-DTPA enhanced MRI in predicting the risk of PHLF in HCC patients.3.An unsupervised machine learning algorithm was used to identify HCC subgroups with different liver function reserves from the preoperative Gd-EOB-DTPA enhanced MRI radiomics features,compare the differences between the subgroups in traditional liver function assessment indicators,and evaluate the correlation with PHLF risk,postoperative complications,length of hospital stay and other events.Materials and Methods:According to different purposes of the study,three groups of study populations were included,and different research methods were adopted to explore the application of Gd-EOB-DTPA enhanced MRI in liver function evaluation.The study populations and research methods included were as follows:1.The clinical data of 117 patients with liver cancer who underwent liver resection in the Department of Hepatobiliary Surgery of our hospital from September 2019 to November 2020 were retrospectively analyzed(study population 1).The portal vein bifurcation plane was selected on the axial image of hepatobiliary phase of preoperative Gd-EOB-DTPA enhanced MRI,and the region of interest was delineated at the position of the remaining liver and bilateral paraspinal muscles.Liver-to-muscle ratio(LMR)was calculated based on the signal intensity of the region of interest,and LMR was used to quantitatively evaluate Liver function.LMR and common clinicopathological factors were included in univariate logistic regression analysis,and statistically significant variables in univariate analysis were included in multivariate logistic regression analysis to screen independent predictors.A prediction nomogram model was constructed with the predictors selected through uni-and multivariable logistic regression analysis.Receiver operating characteristic curve and calibration curve were adopted to evaluate the model’s performance.2.The clinicopathological data of 276 patients who underwent liver resection(open or laparoscopic-assisted)in the Department of Hepatobiliary Surgery of our hospital from January 2017 to March 2019 were retrospectively collected(study population 2).According to the ratio of 7:3,they were randomly divided into training set and test set.The training set was used for the development of clinical prediction model,and the test set was used to verify the prediction performance of the model.Clinicopathological variables were evaluated to identify significant risk factors for PHLF prediction.The hepatobiliary phase images of Gd-EOB-DTPA enhanced MRI were obtained,and the normal liver tissue was segmented and the radiomics features were extracted.The features were evaluated for consistency,relevance,and LASSO regression to obtain reproducible,robust,and non-redundant features for modeling.The PHLF prediction models based on clinicopathological variables(clin-model),radiomics features(rad-model)and their combinations were established.The best PHLF prediction model was determined based on model performance,and the model was visualized in the form of a nomogram.3.The clinical data of 276 HCC patients who underwent hepatectomy from January2017 to March 2019 were retrospectively collected(study population 3).The radiomics features of non-tumor liver tissues on Gd-EOB-DTPA enhanced hepatobiliary phase MRI were extracted,and the reproducible and non-redundant features were selected for consensus clustering analysis to detect different subgroups.The liver function reserve of each subgroup was compared,and the correlations between the subgroups and PHLF,postoperative complications and length of hospital stay was evaluated.Results:1.In study population 1,Model for end-stage liver disease(MELD)score,surgical approach and LMR were independent risk factors for predicting PHLF by univariate and multivariate Logistic regression analysis(P < 0.05),with odds ratio(OR)of 1.33,5.39 and0.44,respectively and were enrolled in the nomogram model.The area under the ROC curve(AUC)of the model was 0.83,with sensitivity of 91.4% and specificity of 64.6%,and the accuracy was 72.7%.The calibration curve showed that the model had good consistency.The model fit was good.DCA curve showed that the model had high clinical practicability.2.In study population 2,The incidence of PHLF in all patients was 24%.The combined model consisting of albumin-bilirubin(ALBI)grade,indocyanine green retention test at 15 minutes(ICG-R15)and radiomics score(derived from 16 radiomics features)was superior to the clinical model and radiomics model alone.It can also improve the predictive performance of common clinical liver function evaluation indexes.In the training set,the AUC value of the combined model was 0.84(95%CI: 0.77-0.90),the sensitivity was 0.78,and the specificity was 0.81.In the test cohort,the AUC value was 0.82(95%CI: 0.72-0.91),the sensitivity was 0.93,and the specificity was 0.67.The model demonstrated a good consistency by the Hosmer–Lemeshow test and the calibration curve.The combined model was visualized as a nomogram for estimating individual risk of PHLF.3.In study population 3,A total of 107 radiomics features were extracted and 37 features were selected for unsupervised clustering analysis.Consensus clustering analysis identified two distinct subgroups(138 patients in each subgroup)with liver function reserve.Subgroup 1 had significantly more patients with younger age,albumin-bilirubin grade 1,lower indocyanine green retention rate,and higher indocyanine green plasma disappearance rate than that of subgroup 2(all P< 0.05).Compared with subgroup 1,subgroup 2 was associated with a higher risk of PHLF,postoperative complications,and longer hospital stay(> 18 days),with an odds ratio of 2.83(95%CI: 1.58-5.23),2.41(95%CI: 1.15-5.35),and 2.14(95%CI: 1.32-3.47),respectively.Conclusions:1.A nomogram model based on MELD score,surgical approach,and Gd-EOB-DTPA enhanced MRI quantitative liver function index LMR was constructed for predicting PHLF.The area under the ROC curve was 0.83,and the sensitivity was 91.4%,which had good discrimination and prediction efficiency.This study demonstrated the potential clinical value of Gd-EOB-DTPA-enhanced MRI in quantitative liver function assessment.2.A comprehensive radiomics model based on Gd-EOB-DTPA enhanced MRI,which combines two clinicopathological variables(ALBI score and ICG-R15)and one radiomics variable(radiomics score),was constructed and internally validated.The AUC of the prediction model was 0.84 in the training set and 0.82 in the test set.The prediction performance of the model was higher than that of the clinical index model and radiomics model in both training and testing datasets.In conclusion,radiomics features can reflect the heterogeneity of liver parenchyma,and radiomics has a good potential application prospect in the assessment of liver function.3.Unsupervised consensus clustering analysis based on Gd-EOB-DTPA enhanced MRI identified two distinct subgroups of HCC patients.The comparison showed that the liver function reserve(ALBI grade,ICG-R15,ICG-PDR were significantly different),the risk of PHLF,postoperative complications,and the length of hospital stay were also different between the two groups,indicating that radiomics features based on Gd-EOB-DTPA enhanced MRI can reflect liver function.Unsupervised clustering analysis algorithms have the potential to automatically identify these features and classify different liver function groups.