Room: ePoster Forums
Purpose: We analyzed a cohort of patients treated for nasopharengeal carcinoma with proton therapy, who subsequently developed treatment related brain imaging changes on MRI. We used a voxel-level analysis to fit a generalized linear model incorporating dose and linear energy transfer (LET).
Methods: A group of 7 nasopharynx patients were identified as having post-treatment MR imaging changes observable as T1 and T2 sequences in temporal lobes. MR images were registered with treatment planning CT images and regions of change contoured and inspected by a practicing radiation oncologist. The contoured regions were identified as response with voxels represented as 1 while voxels within the ipsilateral temporal lobe outside of the response region were represented as 0. An in-house fast Monte Carlo system was used to recalculate treatment plans to obtain dose and LET information. Dose and LET were taken to be the predictors in the generalized linear model. The probit model assumed in the fitting was the normalized cumulative distribution function. Response parameters were calculated for T1, T2, and T1 + T2 imaging changes. Dose with 50% probability (D50) of local imaging changes was interpolated from the model.
Results: The response model shows clear dependence on both dose and LET with T2 response being the most sensitive and T1 + T2 the least sensitive to dose. D50 decreased with increasing LET (avg. slope of -5.25 GyÎ¼m/keV), indicating an increase in biological dose effectiveness. These results indicate that for higher LET values, the dose is more effective, which is consistent with variable RBE models for proton therapy.
Conclusion: A generalized linear model with dose and LET used as predictors was fit using voxel-level analysis of post treatment MR imaging changes in proton therapy patients. Interpolated D50 values show a decreasing trend with LET supporting the typical models of proton RBE.
Funding Support, Disclosures, and Conflict of Interest: Research supported by Cancer Prevention and Research Institute of Texas grant RP160232.