P11.24.A The anisotropic component of the decomposed diffusion tensor predicts overall survival in patients with glioblastoma

• Published in: Neuro-oncology (Charlottesville, Va.) Vol. 24; no. Supplement_2; pp. ii61 - ii62
• Main Authors: Simon, N L, Sinha, R, Sravanam, S, Mayrand, R, Li, C, Wan, Y, Wei, Y, Price, S
• Format: Journal Article
• Language: English
• Published: 05.09.2022
• ISSN: 1522-8517; 1523-5866
• DOI: 10.1093/neuonc/noac174.213

Abstract Background The diffusion tensor can be decomposed into isotropic (DTI-p) and anisotropic (DTI-q) components (Peña et al., 2006). Regions of DTI-q abnormality have a high tumour burden and increased surgical resection of abnormal DTI-q positively correlates with overall survival (OS) (Yan et al., 2017). We aimed to establish if median voxel DTI-q (MVQ) or a distribution measure of DTI-q (DMQ) could act as a neuro-imaging biomarker, to predict OS in patients with glioblastoma. Material and Methods Diffusion tensor decomposition was used to create DTI-p and DTI-q maps, using FSL software (FMRIB Software Library, Oxford). MVQ and DMQ (the 95th centile minus the 5th centile of the DTI-q distribution, divided by the MVQ) were calculated from the preoperative whole brain (WB), contrast-enhancing (CE), and non-contrast-enhancing (NCE) hemisphere DTI-q maps, using fslstats, for 33 patients with glioblastoma. Using R Studio, multiple linear regression (MLR) models were computed to establish if MVQ or DMQ of the WB, CE and NCE hemispheres or age, significantly predicts OS. The Breusch-Pagan Test, on package “lmtest” in R, was calculated for all MLR models, to determine if heteroscedasticity was present and, if so, bootstrapped multiple regression (BMR) models were computed using package “boot” in R. Results Evidence for heteroscedasticity was present in MLRs that modelled the relationship between DMQ of WB, age, and OS (BP = 6.032, p = 0.014) and DMQ of CE hemisphere, age, and OS (BP = 7.163, p = 0.028). In the BMR of WB DMQ, age, and OS, the 95% bias-corrected accelerated confidence intervals (BCa-CI) for the WB DMQ regression coefficient was 133.5 - 1851.4 and included the WB DMQ estimated coefficient of 803.9. The BMR of CE hemisphere DMQ, age, and OS, computed a 95% BCa-CI for the CE hemisphere DMQ coefficient of 101.8 - 1579.6, containing the CE hemisphere DMQ coefficient estimate of 612.414. For both BMRs, the 95% BCa-CI for age coefficients crossed a value of 0. From computed MLR models, WB MVQ (t = -2.569, p = 0.015), CE hemisphere MVQ (t = -2.143, p = 0.040), NCE hemisphere MVQ (t = -2.567, p = 0.015) and NCE hemisphere DMQ (t = 2.39, p = 0.024) were significant predictors of OS. Age did not significantly predict OS in any models and was not significantly related to WB, CE and NCE hemisphere MVQ or DMQ. Conclusion WB, CE and NCE hemisphere MVQ and DMQ predict OS in our tested subgroup of patients with glioblastoma. Age is not a significant predictor of OS and does not significantly correlate with MVQ or DMQ. The median value and distribution spread of DTI-q may act as a prognostic biomarker in glioblastoma, facilitating patient stratification.