Objective To develop the models based on radiomics and genomics respectively for predicting the his-topathologic nuclear grade in patients with localized clear cell renal cell carcinoma(ccRCC)and to explore the correlation between radiomics features and RNA sequencing data for demonstrating the driving mechanism underlying that macro-ra-diomics models can predict the microscopic pathological changes.Methods In this multi-institutional retrospective stud-y,a CT radiomics model was developed for the nuclear grade prediction.Based on genomics analysis cohort,nuclear grade-associated gene modules were identified,and gene model was constructed based on top 30 Hub mRNA to predict nuclear grade.Using radiogenomics development cohort,biological pathways were enriched from Hub genes and radiogenomics map was created.Results The 4-features-based SVM model predicted nuclear grade with an AUC of 0.938 in validation set.A 5-gene-based model predicted nuclear grade with an AUC of 0.736 in genomics analysis cohort.A total of five gene mod-ules was identified to be associated with nuclear grade.Radiomics features were only associated significantly with 271 genes of 603 genes in 5 gene modules and 8 genes of top 30 Hub genes.Difference existed in the enrichment pathway between as-sociated and unassociated with radiomics features,which were associated with 2 genes of five-gene signatures in mRNA mod-el.Conclusion The CT radiomics models had higher predictive performance than mRNA models.The association be-tween radiomics features and mRNA which related to nuclear grade is not universal.
RadiogenomicsClear cell renal cell carcinomaNuclear gradeThe cancer genome atlasWGCNA a-nalysis
NETL
NSTL
万方数据
