Apparent diffusion coefficient (ADC), derived from diffusion imaging is considered a surrogate for tumor cellularity and is routinely used in combination with tumor volumetrics for assessing early treatment response in patients with glioblastoma multiforme (GBM). However, ADC is heavily influenced by treatment-related edema, especially in the context of radiotherapy (RT). The objective of this study was to test the predictive value of Restriction Spectrum Imaging (RSI) based tumor cellularity in relation to clinical outcomes and to assess if it provides information complementary to conventional imaging. Our hypothesis is that RSI (due to its multi-b-shell acquisition and ability to decouple tumor cellularity from edema) provides a reliable estimate of tumor cellularity independent of volume changes in patients with GBM.We analyzed the diffusion and structural images of twenty nine patients with newly diagnosed GBM treated with external beam RT (average dose of 60Gy in 30 fractions) and standard chemo and/or tumor vaccines and receptor inhibitors. The baseline scans were obtained post-surgery (16-STR, 13-GTR) and prior to chemo-radiation (median 10 days pre-RT) and post scans were acquired within 50 days post end of RT. Only imaging obtained prior to start of anti-angiogenic therapy was included. Progression free survival (PFS based on imaging and/or clinical deterioration) and overall survival (OS) were defined w.r.to baseline. All scans were acquired and corrected for geometric distortions and motion as described previously¹. Briefly, for RSI, a single-shot pulsed-field gradient spin-echo EPI sequence was used (TE/TR = 96 ms/17 seconds; FOV = 24 cm) with 4 b-values (b = 0, 500, 1500, and 4000 s/mm²), and 6, 6, and 15 unique diffusion directions for each nonzero b-value, respectively. We used an RSI model with: a spherically restricted component for tumor cells, a cylindrically restricted component for the neurite fraction, a cylindrically hindered component for extracellular space and a free water component for CSF-filled compartments. The RSI model was fit to the unnormalized diffusion signal using least-squares estimation with Tikhonov regularization. Cellularity² values were computed by combining the parameter estimates from the isotropic restricted component and the isotropic component of the neurite fraction. ADC values were calculated from a tensor fit to the full dataset. Contrast enhancing (CE_vol) and FLAIR hyper-intensity (FLAIR_vol) volume were defined semi-automatically in co-registered T1 and FLAIR images at various time-points excluding necrotic core, blood products and resection cavity. The ADC histogram in the CE and FLAIR volumes were characterized using the 10th and 50th percentiles³ and RSI-cell using the 90th and 80th percentiles as ADC is inversely³ and RSI-cell is directly² related to tumor cellularity. Cox Proportional Hazards (CoxPH) were employed to describe PFS and OS on basis of clinical metrics (age, gender and extent of resection) combined with diffusion metrics and structural volumes. Significant diffusion metrics were co-varied with the corresponding structural volumes to assess if they provided any additional value. Due to the exploratory nature, no formal adjustment of type I error was performed (R3.2.2).

Median OS was 617 days (8 censored) and median PFS was 313 days (3 censored). Of the baseline clinical factors, only extent of resection was significantly associated with OS. Serial RSI cellularity and ADC maps of two patients with comparable structural volumes at baseline are shown in Figures 1 and 2. One patient (Patient1) progressed and died early (PFS-216 and OS-275 days) compared to the other (Patient2) who is still alive (PFS-458, OS – 508 days). CE_vol and FLAIR_vol increased for both patients with Patient2 showing pseudo-progression. RSI cellularity was higher in Patient1 than Patient 2 at both baseline (RSI-FLAIR_80%) and post-RT (RSI-CE_90%). Despite the ADC-CE_10% being larger in Patient1 than Patient 2, it is harder to track changes in the images longitudinally.

The summary of multivariate CoxPH analysis are summarized in table 1. For OS, CE_vol and RSI-FLAIR_80% were the strongest predictors at baseline and FLAIR_vol and RSI-CE_90% were the strongest predictors post-RT, with an increase in the metrics being associated with worse survival. ADC-CE_10% and ADC-FLAIR_10% were predictive of PFS only post-RT. Correlation analysis revealed that RSI-FLAIR_80%, RSI-CE_90% and ADC-CE_10% were independent of the structural volumes (r<0.3 for RSI and r<0.45 for ADC). For ADC, this independence held true only when high b values were used for its estimation compared to b=1500s/mm² (r~0.7).

RSI provides a better estimate of tumor cellularity, which may translate to a more reliable measure of treatment response relative to ADC. Specifically, it may provide a robust estimate of response to RT compared to conventional imaging, notably in patients who underwent STRs.