Radiomics Machine Learning Analysis of Clear Cell Renal Cell Carcinoma for Tumour Grade Prediction based on Intra-Tumoural Subregion Heterogeneity

Abstract

Background: Renal cancers are among the top ten causes of cancer specific mortality; of which the ccRCC subtype is responsible for most of the cases. Grading of ccRCC is important in determining the tumour aggressiveness and clinical management. Objectives: To predict the WHO/ISUP grade of ccRCC pre-operatively and characterise the heterogeneity of tumour subregions using radiomics and ML models including comparison with pre-operative biopsy determined grading in a subgroup. Methods: Data was obtained from multiple institutions across two countries from 391 patients with pathologically proven ccRCC. For analysis, the data were separated into 4 cohorts. Cohort 1 and 2 were data from the respective institutions from the two countries, cohort 3 was the combined data and cohort 4 data was a subset of cohort 1 where both biopsy and subsequent histology from resection (partial or total nephrectomy) was available. 3D image segmentation was done to derive a voxel of interest (VOI) mask. Radiomic features were then extracted from the contrast enhanced images and the data normalised. Correlation coefficients and XGBoost model were used to reduce the dimensionality of the features. Thereafter, 11 ML algorithms were implemented for the purpose of predicting the grade of ccRCC and characterising heterogeneity of subregions in the tumours; Results: For cohort 1, 50% tumor core and 25% tumor periphery exhibited best performance with an average AUC of 77.91% and 78.64% respectively. 50% tumor core had the highest performance in cohort 2 and cohort 3 with an average AUC of 87.64% and 76.91% respectively. Cohort 4 with 25% periphery showed an AUC of 95% and 80% for grade prediction using internal and external validation respectively while biopsy histology had an AUC of 31% for the prediction with final grade of resection histology as a reference standard; Conclusion: Radiomic signatures combined with ML have the potential to predict the WHO/ISUP grade of ccRCC with superior performance compared to pre-operative biopsy. Moreover, tumour subregions contain useful information that should be analysed independently when determining tumour grade. It is therefore possible to distinguish the grade of ccRCC pre-operatively to improve patient care and management.