S80
International Journal of Radiation Oncology Biology Physics
Materials/Methods: A mathematical framework to predict achievable OAR DVHs was derived based on correlation of expected dose to the distance from a voxel to the PTV surface (r). OAR voxels sharing a range of r were computed as sub-volumes. A three-parameter, skewnormal probability distribution was used to fit sub-volume dose distributions, and DVH prediction models were developed by fitting the evolution of the skew-normal parameters as a function of r with generalized polynomials. A cohort of 20 prostate and 12 head-andneck IMRT plans with identical clinical objectives was used to train organ-specific models for rectum, bladder, and parotid glands. A sum of residuals analysis quantifying the integrated difference between the clinically-approved DVH and predicted DVH was utilized to identify outliers where the clinical DVH was sub-optimal to the predicted DVH. Small residual sums indicate strong agreement between the clinical and predicted DVHs, negative residual sums indicate a clinical DVH better than predicted, and positive residual sums indicate a clinical DVH worse than predicted. Outliers with large, positive residual sums were re-planned to evaluate feasibility of achieving DVHs predicted by the initial models, and refined models were obtained by excluding these outliers from the training cohort. The ability of the models to prospectively predict OAR DVHs was evaluated on an independent validation cohort of 20 prostate and 12 head-and-neck IMRT plans. Results: The models’ ability to detect sub-optimality and predict achievable DVHs was demonstrated by a significant reduction in the average sum of residuals (SRavg) in five re-planned prostate and headand-neck outliers (before re-planning: SRavg(rectum) Z 5.1 +/- 2.1; SRavg(bladder) Z 5.2 +/- 1.3; SRavg(parotid) Z 2.0 +/- 2.1 and after replanning: SRavg(rectum) Z -2.5 +/- 2.7; SRavg(bladder) Z 1.0 +/- 1.2; SRavg(parotid) Z -11.3 +/- 5.0). The predictive ability of the models on the validation cohort was the same or better than the training cohort: SRavg(rectum, training) Z 2.2 +/- 3.7 versus SRavg(rectum, val) Z 2.6 +/- 4.2; SRavg(bladder, training) Z 2.7 +/- 3.8 versus SRavg(bladder, val) Z 0.3 +/- 5.2; SRavg(parotid, training) Z 4.9 +/- 4.0 versus SRavg(parotid, val) Z 2.8 +/- 6.2. Conclusions: The results of this study demonstrate the ability to successfully predict achievable OAR DVHs based on individual patient anatomy. The models were capable of identifying sub-optimal plans which were brought within expected values via further optimization. Clinical implementation is in progress to evaluate the impact on real-time IMRT QC. Author Disclosure: L. Appenzoller: O. Patent/License Fee/Copyright; IMRT QC Patent. S. Mutic: K. Stock; Radiation Oncology Company. O. Patent/License Fee/Copyright; IMRT QC Patent. J.M. Michalski: None. W.L. Thorstad: None. K.L. Moore: O. Patent/License Fee/Copyright; IMRT QC Patent.
imaging 3-5 weeks before and 4-6 weeks after CRT, before surgery. Rigid image registration was performed between both images and a voxel-tovoxel correspondence was established. We study the mean SUV uptake within the tumor (defined using SUV 2.5) before CRT as a possible surrogate for initial tumor burden; the ratio of SUV after to SUV before CRT (/) as possible surrogates for cell survival. Since TCP in a heterogeneous tumor is a function of average cell kill, or equivalently cell survival, we use a maximum likelihood calculation to estimate parameters for TCP as a function of /. The TCP curve is simulated with a sigmoid function with two parameters: x50, the ratio of SUVs that give TCP Z 0.5 and s50, the slope of the curve at x50. We also study the voxel-to-voxel correlation between SUV before CRT and (correlation between initial tumor burden and cell kill). Results: Of the 20 patients, a total of 11 are non-responders and 9 are responders. We find that responders have significantly lower values of the possible surrogates of cell survival (e.g., / Z 0.41 for responders and 0.57 for non-responders, p value Z 0.03). The maximum likelihood estimate of x50 is 0.44 (95% CI Z 0.25-0.65) and for s50 is 0.71 (95% CI Z 0-1.7). The initial tumor burden (initial SUV) does not correlate with response; in fact there is a non-significant trend of higher SUVbef for responders than for non-responders (4.1 vs. 3.5, p value 0.08). However, a correlation exists between voxels with high initial tumor burden and larger metabolic response (larger SUVaft-SUVbef) and this correlation is stronger among responders. Conclusions: / is a possible surrogate for cell survival in esophageal cancer patients. Even though confidence intervals are large due to the small patient sample, parameters for a TCP curve can be derived and an individual patient TCP can be calculated for future patients. Small initial SUV did not predict for response and a correlation was found between surrogates for tumor burden and cell kill. Author Disclosure: M. Guerrero: None. S. Tan: None. W. Lu: None.
197 Radiobiological Modeling Based on FDG-PET Data for Esophageal Cancer M. Guerrero,1 S. Tan,2 and W. Lu1; 1University of Maryland, Baltimore, MD, 2Huazhong University of Science and Technology, Wuhan, China Purpose/Objective(s): FDG PET is routinely used to diagnose and evaluate response in many cancer sites. Due to the lack of a quantitative relationship between SUVs and tumor characteristics and response, FDG PET is often used in a qualitative way. Our goal is to investigate the relationship of SUVs with radiobiological parameters such as cell survival and tumor control probability (TCP) to allow for a quantitative prediction of tumor response based on SUVs from FDG PET before and after treatment. Materials/Methods: We analyze data for twenty esophageal cancer patients treated with chemo-radiation therapy (CRT) to a dose of 50.4 Gy followed by surgery. The tumor pathological response to CRT was assessed with the analysis of a surgical specimen. The patients underwent FDG PET
198 Radio-inducible Adaptive Response Effect on Hypofractionation of Lung Cancer Radiation Therapy D. Yan,1 B. Marples,1 I. Grills,1 S. McDermott,1 and L. Kestin2; 1 William Beaumont Hospital, Royal Oak, MI, 2MHP/21 Centry Oncology, Troy, MI Purpose/Objective(s): To evaluate the potential effects of radiation adaptive-responses of normal lung tissue on treatment evaluation and optimization for hypo-fractionated lung cancer radiation therapy. Materials/Methods: Dose distributions of 65 lung cancer patients treated using hypo-fractionated radiation therapy were included in this retrospect study. Dose-per-fraction effect on patient lung dose response was evaluated using (1) the LQ model without including the adaptive-response effect; and (2) the radiation induced-repair (IR) model that includes hypersensitivity and adaptive-response parameters. For each patient, the normalized biologically equivalent dose distribution in lung was determined with respect to the lung dose distribution treated by the conventional prescription dose 2 Gy per fraction using LQ model and IR model respectively. Using the
Oral Scientific Abstract 198; Table IR Model
MLD and V20 quantified using LQ- and
LQ-NTD IR-NTD LQ-NTD IR-NTD 30 2Gy (5 12 Gy) (5 12 Gy) (3 20 Gy) (3 20 Gy) V20 (%) 9.3 + 2.9 MLD (Gy) 6.7 + 1.7
14.4 + 4.8 16.1 + 4.5
8.6 + 2.7 10.5 + 3.1
17.4 + 6.0 23.7 + 6.8
9.8 + 3.1 15.5 + 4.8