233 - Early Prediction of Radiotherapy Outcome Using ADC Metrics for Head & Neck Squamous Cell Carcinoma
Presenter(s)
B. McDonald1, S. Mulder2, T. Schakel3, F. Reinders3, K. Kuijer3, P. Doornaert3, L. McCullum4, N. West2, R. de Bree3, M. de Ridder3, C. Terhaard3, and M. Philippens3; 1MD Anderson Cancer Center, Houston, TX, 2UT MD Anderson Cancer Center, Houston, TX, 3University Medical Center Utrecht, Utrecht, Netherlands, 4UT MD Anderson Cancer Center UTHealth Houston Graduate School of Biomedical Sciences, Houston, TX
Purpose/Objective(s): The apparent diffusion coefficient (ADC) from diffusion-weighted MRI (DWI) shows promise as a tumor response biomarker in head & neck squamous cell carcinoma (HNSCC). Prior studies link ADC changes during RT to locoregional failure (LRF), potentially enabling biologically adaptive RT, but the optimal predictive time point (TP) is unknown. Functional diffusion mapping (FDM)—which quantifies the percentage of voxels with increased (VI), decreased (VD), or unchanged (V0) ADC—outperforms whole-tumor median ADC for brain cancer outcomes but has not been evaluated in HNSCC. Our aim is to determine the optimal TP for LRF prediction and compare the predictive potential of FDM vs. whole-tumor ADC metrics.
Materials/Methods: 60 patients with T2-T4 HNSCC undergoing standard-fractionation RT were enrolled in an IRB-approved prospective imaging trial and gave written informed consent. Patients received MRIs at pre-RT, weeks 2-5 of RT, and 3 months post-RT, including T2-weighted and SPLICE-DWI (b=0, 150, 800 s/mm2) scans. Clinical gross tumor volume contours were propagated to ADC maps at each TP. Median tumor ADC and percent changes from pre-RT were calculated. For weeks 2-5, ADC slope was computed using linear regression with that TP and all previous TPs. FDM metrics (VI, VD, V0) were calculated at each TP using a 0.25x10-3 mm2/s threshold. The study endpoint was locoregional control/failure (LRC/LRF) at 24 months post-RT. Wilcoxon signed rank tests were used to assess ADC changes from pre-RT in LRC & LRF groups (a=0.05). Univariate logistic regression with bootstrapping (1000 iterations) was used to identify both clinical and ADC predictors of LRF. Significant ADC metrics were adjusted for the two most predictive clinical factors using multivariate logistic regression with bootstrapping.
Results: The cohort consisted of 48/12 males/females; median (range) age: 63 (42-87); tumor site: 2 oral cavity, 19 HPV+ oropharynx, 21 HPV- oropharynx, 9 hypopharynx, 9 larynx; T stage: 26 T2, 21 T3, 13 T4; and N stage: 17 N0, 14 N1, 22 N2, 7 N3. At 24-month follow-up, 36 patients had LRC, 14 had LRF, and 10 were lost to follow-up or died of unrelated causes. All TPs showed significant ADC increases from baseline for both LRC & LRF groups. Tumor type (p<0.001), N stage (p=0.001), and smoking status (p=0.013) were significant clinical predictors of LRF. V0, VI, and slope at week 3 (p=0.017, 0.027, 0.043) and VD at 3-month follow-up (p=0.032) were significant in univariate analysis but lost significance in multivariate analysis when adjusted for tumor type & N stage.
Conclusion: Week 3 was the best TP for outcome prediction in univariate analysis. FDM metrics/slope may be better predictors than whole-tumor ADC metrics. However, in this heterogeneous cohort, tumor-specific factors play a dominant role in outcome prediction. Further validation is required before ADC can serve as an imaging biomarker for treatment response and adaptive RT.