Construction Of Radiomics Model For Preoperative Subtype Identification And Prognostic Prediction Of Hepatocellular Carcinoma

Hepatocellular carcinoma(HCC)is one of the most common malignancies worldwide.Therefore,effective preoperative risk stratification and accurate prognostic assessment are of great significance for clinical decision making and the surveillance of this disease.Radiomics,a modern technology that involves the transformation of conventional medical images into analyzable,quantitative high-dimensional imaging data,has attracted extensive attention from biomedical researchers.Through the technology of radiomics,the intrinsic information about tumors and their heterogeneity can be captured noninvasively.In addition,combined with genomics,transcriptomics,proteomics and clinicopathological information,radiomics can comprehensively and effectively assist in the evaluation of the biological characteristics of tumors from various aspects.The present study aimed to explore the value of radiomics for HCC patients.In the first part,we reported on our exploration of the performance of ultrasound-based radiomics,for the preoperative identification of HCC subtypes and the subtype differences.Furthermore,the efficacy of radiomics for the preoperative evaluation of clinicopathological features was proposed,such as the status of cancer embolus,microvascular invasion,and early recurrence.In the second part,we provide a predictive model for the effective prognosis evaluation of HCC patients by integrating radiomics based on contrast enhanced computed tomography(CECT),transcriptomics and clinicopathological parameters.In summary,the model proposed in this study has potential to provide effective clinical decision-making strategies for precision medicine for HCC.Part One Preoperative Subtype Identification of Hepatocellular Carcinoma Based on Radiomics DataPurpose: The current study aimed to identify HCC subtype based on radiomics data extracted from medical ultrasound imaging characteristics and to analyze the alterations in the clinicopathological characteristics of different subtypes.Furthermore,we attempted to develop radiomics scores based on the ultrasound characteristics to assess the noninvasive preoperative prediction value of the radiomics scores for HCC recurrence.Methods: The medical ultrasound images,and clinical and pathological features of 182 patients with primary single-focus HCC were obtained from the First Affiliated Hospital of Guangxi Medical University.The largest imaging slice of the ultrasound images of the Digital Imaging and Communications in Medicine(DICOM)was used to extract relative features.Regions of interest(ROI)were manually sketched using ITK-SNAP(Version 3.8.0),image segmentation software.Ultrosomics software,Version 1.1(GE Healthcare)was used to extract the quantitative features of the radiomics based on the selection of the ROIs.Radiomics features with a variance of zero were removed,and the rest were submitted to z-score standardized processing and subsequent steps.Using Consensus Cluster Plus(Version 1.46.0),we divided the HCC patients into several subtypes and compared the alterations of the clinicopathological features among patients with different subtypes.Furthermore,the HCC patients were further divided into various labeled groups according to the prediction tasks of the status of cancer embolus,microvascular invasion,and early recurrence.A hypothesis test screened the histological characteristics of the images for the differences between the groups,and a Least Absolute Shrinkage and Selection Operator(LASSO)regression model was further used to establish the prediction models for each group based on the radiomics features;the results were assessed using 10-fold cross-validation.The predictive powers of the models were examined by the area under the curve(AUC)of the receiver operating characteristic curve(ROC).Results: According to the exclusion criteria of this study,a total of 182 patients with HCC were all included.Their medical ultrasound images were examined within two weeks of their surgery.By segmenting of the largest sections of the tumor images,a total of 1076 radiomics features were obtained in four categories: first-order statistical features,shape-size features,texture features and wavelet features.After radiomics features with zero variance were removed,968 image features were analyzed.Four HCC subtypes were identified by using unsupervised clustering of the radiomics features.There were significant alterations in HCC patients between different subtypes for Barcelona Clinic Liver Cancer and AFP levels.Hypothesis tests screened 135,87,and 103 ultrasound radiomics features that were differentially displayed in the cancer embolus groups(positive/negative),the microvascular invasion groups(positive/negative),and the early recurrence groups(positive/ negative).According to the LASSO algorithm,5,8,and 6 features were identified for the construction of the radiomics models to predict the formation of tumor thrombus,microvascular invasion,and early recurrence,respectively.For the prediction of cancer embolus formation,the AUC value,corresponding 95% confidence interval,sensitivity,and specificity were 0.7675(95%CI: 0.6978-0.8372),0.7155 and 0.7121,respectively.For the prediction of microvascular invasion,the AUC value,the corresponding 95% confidence interval,sensitivity,and specificity were 0.8306(95%CI: 0.7305-0.9306),0.7750 and 0.8148,respectively.For the prediction of early postoperative recurrence,AUC value,the corresponding 95% confidence interval,sensitivity,and specificity were0.8303(95%CI: 0.7404-0.9203),0.6400 and 0.9310,respectively.Conclusions: The hierarchical clustering of radiomics proposed in the present study provides an in-depth understanding of the tumor biology of HCC and provides a new perspective for the personalized treatment of HCC.In addition,the radiomics prediction models,based on ultrasound features,are capable of effectively predicting tumor thrombosis,microvascular invasion,and early recurrence in HCC patients preoperatively.Part Two The Clinical Outcome Prediction of Hepatocellular Carcinoma Using Combined Radiomics and Transcriptomics AnalysisPurpose: To construct prognostic models of HCC patients based on radiomics and transcriptomics respectively,and to explore the integrative value of radiomics,transcriptomics and clinicopathological data for the survival prediction of HCC patients.Methods: RNA sequencing data from 374 HCC patients were acquired from The Cancer Genome Atlas database.Genes with marked clinical significance for HCC were identified by differential expression analysis and survival analysis.One of the genes,polo like kinase 1(PLK1),was verified by immunohistochemistry with clinical samples collected from our institute.A prediction model based on transcriptomics data was constructed by multivariate Cox regression analysis.Meanwhile,the maximum cross-sectional images of the portal phase from CECT of 38 HCC patients were downloaded from The Cancer Imaging Archive database.The ROIs were delineated using ITK-SNAP(Version3.8.0)software for image segmentation.Ultrosomics software(GE healthcare,Version 1.1)was used to extract quantitative features from the radiomics features in the ROIs.The features with zero variance were removed and a z-score normalized process was conducted.Survival associated-radiomics features were identified by univariate COX analysis.Subsequently,a prognostic model based on radiomics was constructed via multivariate COX analysis.Finally,a nomogram plot was proposed to integrate the radiomics,genomics and clinical parameters.Results: A total of 77 differentially expressed long non-coding RNAs,29 micro RNAs and 1014 messenger RNAs were found to be closely related to the survival of HCC patients.Moreover,a prognostic index for HCC based on nine indicators was proposed,which exhibited moderate prognostic evaluation efficacy(AUC=0.776).Immunohistochemical experiments confirmed that PLK1 was significantly highly expressed in HCC tissues as compared to non-HCC liver controls,and its higher expression could act as a favorable indicator for the poor prognosis of HCC.Univariate Cox regression analysis was performed to screen out 21 radiomics features related to the prognosis of HCC(P<0.05),and two features were used to construct the radiomics-based prognostic index via multivariate Cox regression analysis.The nomogram established by integrating transcriptomic,radiomics and TNM stage showed great value for the survival assessment of HCC patients.Conclusion: The present study comprehensively analyzes the prognostic value of transcriptomics and radiomics.Furthermore,this study provides an effective and accurate prognostic evaluation system for HCC patients by combining transcriptomic,radiomics and clinicopathological data.