2021 - Response at the Voxel Level in HNSCC Using Daily Quantitative MRI from MR-guided Radiation Therapy
Presenter(s)
R. Bonate1, M. J. Awan2, M. E. Shukla2, S. Tarima3, H. A. Himburg2, J. O. Zenga4, and E. S. Paulson2; 1Department of Biophysics, Medical College of Wisconsin, Milwaukee, WI, 2Department of Radiation Oncology, Medical College of Wisconsin, Milwaukee, WI, 3Medical College of Wisconsin, Milwaukee, WI, 4Department of Surgery, Medical College of Wisconsin, Milwaukee, WI
Purpose/Objective(s): Recently, we demonstrated the utility of daily quantitative MRI (qMRI) obtained during MR-guided radiotherapy (MRgRT) in detecting both differential and regional responses to radiotherapy (RT) in head and neck squamous cell cancer (HNSCC) patients. We investigate here whether the behavior of the primary gross tumor volume (GTVp) can be reliably monitored at the single voxel level during RT in a cohort of HNSCC patients.
Materials/Methods: Seventeen HNSCC patients treated with hypo-fractionated MRgRT at 50-60 Gy in 15 fractions were included in the study (DEHART, NCT04477759). Five patients were treated with concurrent atezolizumab, and twelve patients were treated with RT alone. Daily intravoxel incoherent motion, variable flip angle SPGR, and CPMG sequences were acquired on a 1.5T MR-Linac in the idle time during adaptive plan generation. ADC, T1, T2, f, and D (derived from b-values 150 and 550 s/mm2) parameters were extracted from the physician-defined GTVp. T1 and T2 data, obtained from the GTVp contour at fractions 1, 6, and 11, was used to fit a linear regression model for each voxel and determine the first derivative of response on a voxelwise basis. Using the derivative, response at fraction 15 was predicted on a voxelwise basis and compared to the observed data. Three-dimensional plotting of the GTVp was performed using custom Python scripts to investigate spatial patterns in response and predictability. This analysis was performed using native 1mm2 voxels with a 2.5mm smoothing filter applied.
Results: High accuracy in direction of response prediction was found in both T1 and T2 (67.85% and 69.44% of voxels respectively). Voxels that were predicted incorrectly were slightly more likely to be trending downward rather than upward (54.67% negative change T2, 56.11% negative change T1). Three-dimensional plotting revealed spatial patterns in predictability, with the relatively immobile tumor core demonstrating higher predictability than the outer tumor rim. The quality of results was slightly higher in patients with endophytic tumors than in patients with exophytic tumors (74.10% vs. 68.93% voxels predicted correctly for T2, respectively). Smoothing did not noticeably change predictability at the voxel level (69.44% correct without smoothing vs. 70.59% with smoothing for T2). Smoothing did however increase 3D image clarity.
Conclusion: Voxelwise investigation of qMRI changes during radiation therapy for HNSCC suggests that the direction of response to RT can be reliably monitored using T1 or T2 maps. Three-dimensional plotting of the GTVp revealed that the tumor periphery has a lower level of predictability than the tumor core. These findings support qMRI-based monitoring of patient progress in a biologically guided adaptive RT approach for HNSCC.