NIMG-26. RADIOMIC FEATURES OF GLIOBLASTOMA ON PRE-TREATMENT GD-T1W MRI ARE PREDICTIVE OF RESPONSE TO CHEMO-RADIATION THERAPY AND ASSOCIATED WITH AKT AND APOPTOSIS PATHWAYS. (5th November 2018)

• Record Type: Journal Article
• Authors: Beig, Niha; Prasanna, Prateek; Verma, Ruchika; Hill, Virginia; Varadan, Vinay; Madabhushi, Anant; Tiwari, Pallavi
• Is Part Of: Neuro-oncology. Volume 20(2018)Supplement 6
• Page Start: vi181
• Page End: vi181
• Publication Date: 2018-11-05
• Subjects: Brain Neoplasms -- Periodicals; Brain -- Tumors -- Periodicals; Brain -- Cancer -- Periodicals; Nervous system -- Cancer -- Periodicals; 616.99481
• DOI: 10.1093/neuonc/noy148.752
• Languages: English
• ISSNs: 1522-8517

Abstract

Abstract: BACKGROUND: >40% of Glioblastoma patients do not respond to chemo-radiation (Ch-Rx) treatment and recur within 6–8 months, suggesting that they could be better candidates for experimental therapies. We seek to discover radiomic features on pre-treatment MRI that are predictive of response to Ch-Rx. Further, in an attempt to establish biological underpinning of these predictive radiomic features, we then identified radiogenomic correlations between these features with molecular pathways that are known to impact response to Ch-Rx. METHODS: 49 GBM studies were obtained from publicly available IVYGap (n=29) and TCIA (n=20) databases, with pre-treatment MRI scans (Gd-T1w, T2w, FLAIR) and corresponding RNA-Seq data. Responders were defined as patients with progression-free survival (PFS) of >= 4-months, while non-responders had PFS of <4-months. A total of 1305 3D-radiomic features (Gabor, Haralick, and Laws energy) were extracted from each MRI protocol from expert annotated regions (enhancing, edema/non-enhancing, necrosis) for every study. Top 5 predictive features were obtained from a training set and validated using a support vector machine classifier. 13 signaling pathways that are known to be implicated in Ch-Rx response were curated from the MSigDB HALLMARK cohort. Gene Set Enrichment Analysis (GSEA) scores of these pathways was correlated with the top radiomic features with Bonferroni correction. RESULTS: Laws energy features that characterize appearance of ripples and spots from enhancing region on Gd-T1 MRI were found to best predict Ch-Rx response (sensitivity of 73.3% on validation set). These features were statistically significantly correlated with AKT and apoptosis signaling pathways (p<0.02, FDR = 5%). CONCLUSION: Activation of AKT pathway facilitates angiogenesis and develops Ch-Rx resistance. Similarly, higher expression of apoptotic proteins in GBM is associated with favorable Ch-Rx response. Our radiogenomic approach may allow for non-invasive CRT response prediction by providing biological understanding of molecular pathways implicated in Ch-Rx response, as manifested on pre-treatment Gd-T1 MRI.