CT Delta-Radiomics Algorithm Predicts Progression-Free Survival (PFS) in Metastatic Renal Cell Carcinoma (RCC) Treated with Anti-Angiogenic (AAG) Therapy
Amir Khadivi1, Edward Florez1, Niki Patel1, Khalid Manzoul1, Benjamin Rushing1, Sarah Miller1, Elliot Varney1, Charlene Claudio1, Juliana Sitta1, Rana Gordji1, Amy Krecker1, Gerri Wilson1, Stella Powell1, Seth Lirette2, Andrew Smith3, Candace Howard1
1Department of Radiology, University of Mississippi Medical Center, Jackson, MS
2Department of Data Science, University of Mississippi Medical Center, Jackson, MS
3Department of Radiology, University of Alabama at Birmingham, Birmingham, AL
A multi-institutional prospective phase III trial evaluating sunitinib as first-line agent in patients with metastatic RCC was conducted on 275 patients with digital CT images in an effort to develop a CT delta-radiomics algorithm to predict progression-free survival (PFS) in patients with metastatic RCC treated with AAG therapy. CT radiomic features (RCT=250) were measured on baseline and initial post-therapy CT images using a quantitative software. Tumor length and CT radiomic features with high inter-observer agreement (ICC>0.60; RIO=14 candidate parameters) among 11 readers who evaluated 20 random patients were incorporated into a statistical model (CT delta-radiomics algorithm) and associated with PFS using Cox-proportional hazards ratio and log-rank test. The final CT delta-radiomics algorithm included: change in both target lesion length and tumor area, gray level non-uniformity, and run length non-uniformity. CT delta-radiomics algorithm non-responders (NNR=135) on the initial post-therapy CT exam were 2.6 times more likely to progress than responders (NR=140; HR=2.6, p<0.001). The median PFS of 0.7 years for non-responders was significantly different than the median PFS of 1.6 years for responders (p<0.001). Delta-radiomics analysis in CT images has the ability to measure changes in tumor heterogeneity. Two radiomic features had both high inter-observer agreement and a statistically significant association with PFS.