2135 - Optimal Apparent Diffusion Coefficient (ADC) Thresholds for Predicting Geographic Recurrence in Glioblastoma during Chemoradiation (CRT)

Purpose/Objective(s): ADC from diffusion-weighted imaging (DWI) detects early treatment response in glioblastoma. We hypothesized that ADC changes during radiation identify glioblastoma regions prone to recurrence. Our objective was to determine optimal thresholds for predicting geographic recurrence from ADC values or voxel-wise ADC changes.

Materials/Methods: In this prospective observational serial imaging study (Dec/2017–Apr/2021), patients underwent DWI at radiation planning (Fx0), fraction 10 (Fx10) and fraction 20 (Fx20) during a standard 6-week course of CRT. The gross tumor volume (GTV, surgical cavity + residual tumor) was contoured at each time point. Recurrence was contoured at the earliest magnetic resonance imaging timepoint showing progression. The intersection of GTV and recurrence was labeled as radiation-resistant GTV (R-GTV), while the non-overlapping portion of the GTV as radiation-sensitive GTV (S-GTV). ADC values and percentage changes from Fx0 were compared between these regions using t-tests and logistic regression with stepwise model selection. Temporal changes were assessed using a linear mixed-effects model (fixed effects: region, time; random effects: subject). Predictions of R-GTV were defined via a thresholding method at Fx10 as regions of low ADC value or change from baseline within the GTV with varying thresholds. Sensitivity and specificity were calculated for the prediction and the Youden index (sensitivity+specificity-1) was maximized to obtain the optimal threshold.

Results: 80 patients were analyzed. Median ADC changes from baseline for R-GTV and S-GTV were +2.5% vs +15.1% at Fx10 (P<0.001) and +8.1% vs +23.1% at Fx20 (P<0.001), respectively. The linear mixed-effects model revealed a significant difference in ADC changes between R-GTV and S-GTV (p<0.001). Multivariable analysis identified smaller ADC changes at Fx10 (OR 0.95, P=0.005) and Fx20 (OR 0.95, P=0.010), MGMT unmethylation, and biopsy/subtotal resection (STR) as independent predictors of increased risk of GTV failure. Table 1 details the R-GTV prediction performance for the optimal ADC thresholds at Fx10 for the entire population and stratified by predictors of GTV recurrence.

Conclusion: The optimal ADC-based predictions of geographic recurrence demonstrated moderate predictive performance in an unselected population, with improved accuracy in a risk-subgroup analysis. While our findings may help inform the design of MR-guided, biologically adapted clinical trials, the small sample size may have influenced the results and thus warrants internal and external validation.