Research On Predicting Clinical Outcomes Of High-grade Serous Ovarian Cancer Based On Radiomics Features

Objective: The purpose of this study is to explore whether the models based of quantitative parameters can accurately predict the progression-free survival(PFS),overall survival(OS)and platinum resistance of High-grade serous ovarian cancer(HGSOC),and whether the predictive models are related to the expression of p53 and Ki-67.Furthermore,to assess the power of a cross-modal integrative framework based on quantitative radiological and histopathological image features for predicting overall survival in patients with HGSOC and to identify potential molecular mechanisms underlying survival-associated traits preliminarily.Methods: 1.The patients with HGSOC confirmed by pathology from January 1,2010 to December 30,2020 in our hospital were retrospectively collected,and all the patients were examined by PET/CT before surgery.The conventional PET/CT measurements of both primary and metastatic lesions were recorded,and used to obtain the spatial heterogeneity indicators.and the spatial heterogeneity indicators based on the texture features were obtained using the Haralick texture features of VOIs.The conventional,spatial and integrated PFS predictive model,OS predictive model and platinum resistance predictive model were constructed.The Least absolute shrinkage and selection operator-Cox(LASSO-Cox)method was applied to feature selection and modeling of PFS and OS models.The feature selection and modeling method of platinum resistance analysis used automatic machine learning,and the model with largest area under the curve(AUC)of the receiver operating characteristic(ROC)curve was selected as the best model.p53 and Ki-67 immunohistochemical staining were performed on the pathological samples of patients after surgery.Then Spearman correlation coefficient(ρ)was used to evaluate the correlation between p53,Ki-67 immunohistochemical scores and the riskscores obtained from each model.2.The information of ovarian cancer patient in the Cancer Genome Atlas(TCGA)and the Cancer Imaging Archive(TCIA)databases was collected retrospectively.The LASSO-Cox regression was performed to establish the models,including radiological predictive model,histopathological predictive model and integrative model based on both computed tomography image features and histopathological image features.Finally,an available nomogram was developed to predict the overall survival of patients with HGSOC.Further,weighted gene co-expression network analysis and functional enrichment analysis(gene ontology and Kyoto Encyclopedia of Genes and Genomes)were performed for survivalassociated traits.Results: 1.A total of 292 patients with high-grade serous ovarian cancer were included.The prediction efficiencies of the integrated model was higher than that of conventional model and spatial model.The concordance indices(C-indices)of training set and verification set in the integrated PFS predictive model were 0.898(95% Confidence Interval [CI]: 0.881-0.914)and 0.891(95%CI: 0.860-0.921)respectively,and the C-indices of training set and verification set in integrated OS predictive model were 0.894(95%CI:0.871-0.917)and 0.905(95%CI: 0.873-0.936)respectively.The AUCs of training set and verification set in the best integrated platinum resistance model were 0.979 and 0.967 respectively.The area under the PRC is 0.952 and 0.953 respectively.p53 and Ki-67 had varying degrees of correlation with riskscores of predictive models,And the integrated PFS predictive model had the strongest correlation with p53(ρ =0.859,p<0.001)and Ki-67(ρ=0.829,p<0.001).2.A total of 62 patients with HGSOC were included in this study.For radiological predictive model,the concordance indices of training set and the verification set were0.916(95%CI: 0.858-0.975)and 0.865(95%CI: 0.669-0.989),respectively.For histopathological predictive model,the concordance indices of training set and the verification set were 0.943(95% CI: 0.908-0.978)and 0.866(95% CI: 0.691-0.981),respectively.For integrative model,the concordance indices of training set and the verification set were 0.925(95% CI: 0.869-0.980)and 0.874(95% CI: 0.684-0.986),respectively.A practicable nomogram was built based on the integrative model data.Male sex differentiation,neuroactive ligand-receptor interaction,negative gene expression regulation,epigenetic mechanisms,and systemic lupus erythematosus were identified as significant potential molecular mechanisms related to survival-associated traits.Conclusions: 1.The models based on PET/CT spatial heterogeneity indicators had good predictive performance on predicting progression-free survival,overall survival and platinum resistance of high-grade serous ovarian cancer.The riskscores of the predictive models were correlated with the expression of p53 and Ki-67 in varying degrees.2.The cross-modal integrative framework based on radiological and histopathological image features could reliably predict the prognosis of HGSOC.The potential underlying molecular mechanisms may provide new insights for future HGSOC research.