A systematic review of radiogenomic applications in prostate cancer

Introduction: A priority in prostate cancer (PCa) research is development of precise risk stratification tools to enable early identification of men with aggressive tumours while minimizing overtreatment of others with indolent disease. With imaging playing a central role in the diagnosis and management of PCa, radiogenomics has been explored as a personalised medicine approach to improve risk stratification. This study aims to review and describe radiogenomic applications reported in the literature. Methods: Medline, Embase and Cochrane libraries were systematically searched using variations of search terms for articles reporting radiogenomic applications in PCa after 2010. Articles were included if the following were reported (1) genomic platform used; (2) method of determining region of interest (ROI) for radiomic feature extraction; (3) correlation analyses between radiomics and genomics. Results: A total of 267 articles were screened and 13 met the inclusion criteria following independent review by two authors. Majority (n=10/13) reported MRI-based applications involving 715 patients. Remaining modalities included ultrasound (n=1) and PET scan (n=2). Most (7/10) studies evaluating an MRI imaging modality, correlated bulk RNA-sequencing with radiomic features. Textural radiomics (n=6) features were most commonly reported to correlate with gene expression followed by histogram (n=2) and volumetric features (n=1). MRI radiomics significantly correlated with hypoxia related genes in 4 studies. The textural feature (Gray Level Co-occurrence Matrix) was seen to correlate with ANGPTL4 expression in 3 studies. Median AUC for a radiogenomic model to predict presence of clinically significant PCa was 0.746. Only 4 studies (MRI-based n=3, ultrasound-based n=1) externally validated the developed model. Most (9/13) studies used a manual qualitative approach to register imaging loci with site of tissue acquisition for genomic analysis. Conclusion: There is significant heterogeneity in the reporting and design of prostate cancer radiogenomic studies. A signal suggesting and association of MRI textural radiomic features have been consistently observed in several studies but lack validation