2565 - MRI Relaxometry for Glioblastoma Response Assessment during Radiation Therapy on a 0.35 T MRI Linear Accelerator

N. Lutsik¹, S. Nejad-Davarani¹, K. Cullison¹, D. Maziero², A. Valderrama¹, M. De La Fuente³, G. J. Kubicek¹, J. J. Meshman¹, G. Azzam¹, J. Ford¹, and E. A. Mellon¹; ¹Department of Radiation Oncology, Sylvester Comprehensive Cancer Center, Miller School of Medicine, University of Miami, Miami, FL, ²UCSD Health Radiation Medicine and Health Sciences, La Jolla, CA, ³Department of Neurology, Sylvester Comprehensive Cancer Center, Miller School of Medicine, University of Miami, Miami, FL

Purpose/Objective(s): Radiation therapy (RT) by an MRI-linear accelerator (Linac) enables daily MRI for glioblastoma (GBM) RT adaptation. Multiparametric MRI (mpMRI) may further enhance adaptive RT decisions using physiologic tumor changes reflected in quantitative imaging. This abstract analyzes the ability of T1, T2, and T2* relaxometry to serve as predictors for early GBM progression. The purpose of our study was to discover how multiparametric MRI values change during RT and compare them to patient outcomes. Materials/

Methods: 29 patients (11 Female; median age of 64) with newly diagnosed GBM underwent chemoradiation on a 0.35 T MRI-Linac with 60 Gy in 30 fractions and temozolomide. IRB-approved research imaging was acquired on alternate days of treatment with a scanning protocol including Strategically Acquired Gradient Echo (STAGE: TR =42 ms; TE= 5/20.63/34.14 ms; Voxel size = 1.5x1.5x4mm; FA=10/50°) and multi-echo T2 (TR = 4000 ms; TE=29/73/131 ms; Voxel size = 1.6x1.6x6 mm) sequences. T1, PD (Proton Density), T2*, and T2 maps were generated using developed code. Map values within region of interest (ROI) (Lesion/Cavity on balanced Steady State Free Precession image +1 cm and excluding the resection cavity) were calculated. Parametric response maps (PRM) were defined as voxel-by-voxel percent change between every fraction and pre-RT by classifying each ROI voxel as either decreased >5 %, increased >5 % or unchanged (+/-5%). Patients were divided into 2 groups based on early clinical response to treatment: NP (no progression) and TP (true progression). Non-progressors included patients with pseudoprogression as well. Statistical analysis included analysis of variance of PRMs for every response group and stepwise multinominal logistical regression with two outcomes: NP and TP. Correlation with overall survival (OS) was evaluated by Cox proportional hazards models. The results were considered statistically significant at the two-sided 5% comparison-wise significance level.
Results: 18 patients (62 %) had NP and 11 (38 %) had TP. We observed a significant difference between non-progressors and true-progressors in voxels with increasing of T2 (NP: 40 ± 2.3 %; TP: 47 ± 3.7%, p = 0.02) and T2* values (NP: 8 ± 0.3 %; TP: 10 ± 0.5 %; p = 0.04) more than 30% on 1^st and 2^nd week of RT. The increasing of T2 and T2* values more than 30% at week 2 was correlated with OS (HR = 4 and 3.08, respectively, p < 0.05). Based on Akaike’s information criteria, the significant predictors of response in the models were increasing of T2 and T2* values more than 30% (p < 0.001).
Conclusion: mpMRI PRMs demonstrated significant differences mostly in T2* and T2 values in the 1^st and 2^nd weeks of RT in poorly responding GBMs. As such patients have the worst prognosis, they might benefit from early therapy intensification or adaptation with MRI-Linac and/or other therapies.