weighting factors by minimizing the sum of squares error between CT- SIM and synCT. Clinical plans were recalculated using fixed monitor units, plan dosimetry was evaluated, and local dose differences were characterized using gamma analysis (1%/1 mm dose difference/distance to agreement). Results: Overall mean absolute error between synCT and CT-SIM was 71.7  14.6 HU across all patients. While synCT yielded the smallest target dose differences from CT-SIM for D95, D99, and mean dose (<0.7 Gy (1%)) compared to MR w and MR w+b , pairwise comparisons were not significant. Small but significant (P < .05) improvements were observed for synCT for bladder D35 and D15 compared to MR w and MR w+b . No differences in rectum metrics between datasets were detected. For penile bulb, MR w provided marginally better D90 agreement than synCT or MR w+b . Synthetic CT gamma analysis pass rates (97.1  3.3%) exceeded MR w (94.1  5.5%) or MR w+b (93.3  9.2%), although differences were not statistically significant (P Z 0.06 and P Z 0.09, respectively). One subject’s gamma analysis results were lower for synCT (89.7%) than MR w (98.5%) and MR w+b (96.7%) due to an increase in rectal gas during MR-SIM and the ability of synCT to account for air. For this case, synCT rectal doses were greater than MR w for all metrics (1.6e2.6 Gy or 4.3e5.6%). Conclusion: Synthetic CT values provided better agreement to CT-SIM for dose metrics and gamma analysis than bulk density assignment-based MR images, with most improvement observed for the bladder. Synthetic CTs offer the potential to generate digitally reconstructed radiographs and may provide additional clinical value in treatment sites with greater air-to-soft tissue ratio. Author Disclosure: K.S. Garbarino: None. J.P. Kim: None. N. Wen: Research Grant; Varian Medical System. L. Schultz: None. K.J. Levin: None. B. Movsas: Research Grant; Philips Medical System, Varian Medical System. Chair; ASTRO. M.U. Siddiqui: Research Grant; Varian Medical System. Honoraria; Varian Medical System. I.J. Chetty: Research Grant; Varian Medical System, Philips Medical System. C.K. Glide-Hurst: Research Grant; Philips Medical System. Board of Di- rectors; AAPM. 1004 MR Based Tumor Delineation for Lung Cancer Radiation Therapy Planning S. Saraiya, 1 G.D. Hugo, 1 J.M. Schuster, 2 M.E. Schutzer, 1 N. Jan, 1 L. Fahrner, 1 T.D. Catron, 1 J.C. Ford, 1 K. Karki, 1 R. Groves, 1 and E. Weiss 1 ; 1 Virginia Commonwealth University, Richmond, VA, 2 Virginia Commonwealth University Medical Center, Richmond, VA Purpose/Objective(s): Magnetic resonance imaging provides excellent soft tissue contrast and higher image resolution than PET, but has seen limited use in target definition for lung cancer radiation therapy. In this study, we test the feasibility of MR-based contouring by comparing interobserver contouring variability on MRI to the current standard of CT and PET-CT. Materials/Methods: Nine patients with locally advanced lung cancer underwent T1 weighted 1.5T MRI (Volumetric Interpolated Breath-hold Examination e VIBE), diffusion weighted MRI (DW-MRI), CT, and PET-CT scans prior to radiation therapy. Seven of 9 patients had findings of atelectasis or pneumonia associated with their tumors; MRIs were acquired using expiration breath-hold for VIBE and respiratory naviga- tion for DW-MRs. Following detailed contouring instructions, 2 radiol- ogists, and 5 radiation oncologists delineated the lung tumor on the following protocols: 1) CT images without PET (CT); 2) CT images rigidly registered to PET-CT (CT/PET); and 3) morphological VIBE images rigidly registered to functional DW-MRI (MRI). Overlap frac- tions (OFs) were calculated as the intersection over union of physician volumes for each scan and protocol. Additionally, the contour boundaries were measured in 6 cardinal directions and the local standard deviation (local SD) was computed as the SD across all observers for each boundary. Overall standard deviation was computed as the mean of all local SDs for each scenario and used as a spatial measure of interob- server variability. Results: Mean of  standard deviation of CT, CT/PET, and MRI were 0.38  0.14, 0.36  0.12, and 0.32  0.14, respectively. Differences be- tween the three contouring sessions were not significant. Union volumes on CT (262  226 cm 3 ) were larger than CT/PET (169  125 cm 3 , P Z .02), and larger than MRI (226  169.6 cm 3 , P Z .17). Differences be- tween CT/PET and MRI were borderline significant (P Z .05). Intersec- tion volumes on CT were larger than CT/PET (105.29  98.5 cm 3 vs 65.3  61 cm 3 , P Z .03) and MRI (66.87  57.74 cm 3 , P Z .09). Intersection volumes on CT/PET and MRI were comparable (P Z .43). In contour boundary analysis, paired t-test did not show significant difference in overall SD between CT/PET and MRI (P Z .19), and between CT/PET and CT (P Z .27). Conclusion: An MRI-based delineation protocol showed similar, although slightly higher, spatial, and volumetric interobserver variability as a standard CT/PET based protocol. Further training may be required to reduce the level of interobserver variability for the MR-based protocol to that of CT/PET. Author Disclosure: S. Saraiya: None. G.D. Hugo: Research Grant; Var- ian, Philips, NIH. Senior Editor; Practical Radiation Oncology. Vice Chair; AAPM Therapy Imaging Subcommittee. J.M. Schuster: None. M.E. Schutzer: None. N. Jan: None. L. Fahrner: None. T.D. Catron: None. J.C. Ford: None. K. Karki: None. R. Groves: None. E. Weiss: Research Grant; Varian, Philips, NIH. Patent/License Fees/Copyright; UptoDate. 1005 A Novel Tool to Assess Contouring Accuracy of Organs-at-Risk in Stereotactic Radiosurgery K. Stang, 1 E. Guenther, 2 Z. Kozel, 3 I.A. Rusu, 1 Z. Siddiqui, 1 E. Melian, 4 and A. Sethi 1 ; 1 Loyola University Medical Center, Maywood, IL, 2 Loyola University Chicago, Stritch School of Medicine, Maywood, IL, 3 Stritch School of Medicine, Loyola University Chicago, Maywood, IL, 4 Loyola University Medical Center, Maywood, IL Purpose/Objective(s): Autosegmentation of organs-at-risk (OARs) is desirable as it facilitates treatment plan automation and on-line adaptive radiation therapy; however, these OAR structures may be prone to con- touring variation caused by image/motion artifacts and limitations in the autosegmentation algorithm. This requires time consuming manual inter- vention and editing of contours. We sought to create an automated tool to assess and improve quality of computer generated OAR contours. Materials/Methods: Intracranial OAR contours (brainstem, optic chiasm, optic nerves, and eyes) were delineated by an expert user on MR axial T1 image sets for ten stereotactic radiosurgery patients using a predetermined window/level setting (200e800). These contours were analyzed for interpatient variation in absolute OAR volume, volume ratios, and distance between OARs. Each parameter was adjusted for patient anatomical var- iations by normalizing to AP and lateral head thickness. The AP separation was taken from the nasion to the back of the head, while the lateral sep- aration was measured between the top of the ears. Organs-at-risk separa- tions with respect to brainstem were obtained by measuring the distance between the center of mass of respective structures. Brainstem was chosen as the reference OAR as it was found to be least susceptible to contouring errors. Parameters with the smallest standard deviation (least interpatient variation) were used to create a tool to assess accuracy of OAR contours for future patients. This tool was applied to 9 SRS patients with varying degree of OAR contouring errors caused by the observer’s lack of con- touring experience and incorrect window level setting. Each new patient was subsequently added to the database to improve tool robustness. Results: Organs-at-risk separation was most effective in predicting con- touring errors, and thus, formed the core of the assessment tool. The spatial relationship data found small standard deviation (2.2e3.5%) for all OAR- brainstem distances among all patients; OAR volume based parameters showed significantly larger variations (10e42%) and poor predictive power, and therefore, had lower weighting in our assessment tool. Appli- cation of assessment tool to new patients was able to detect contouring International Journal of Radiation Oncology  Biology  Physics S158