ESTRO 36 Abstract Book

S480 ESTRO 36 2017 _______________________________________________________________________________________________

the treatment planning system, along with a three- dimensional motion model accuracy (defined as the 75th percentile motion error in each voxel) map. The patients still undergo a commercial 4DCT protocol to provide a comparison between the current standard of care and the model-based process. Comparisons between the commercial and model-based approaches have been conducted on 19 patients to evaluate the magnitude of sorting artifacts in each process on a scale of 1-4, 1 having no artifacts and 4 having severe artifacts. The average CT noise for both protocols was described by examining a region of interest in the liver. Results Mean tumor displacement was 11.5 +/- 6.9 mm and the mean motion model error was 1.77 +/- 0.79 mm. The mean artifact severity ratings for the 4DCT and model-based CT approaches were 2.2 and 1.2, respectively. There were three instances of grade 4 artifacts and no instances of grade 3 or worse artifacts for the 4D and model-based approaches, respectively. The average CT noise was reduced from 57.7 HU to 11.6 HU. Conclusion The model-based approach provides the clinic with motion artifact free images that have lower noise and whose geometry accurately reflects the tumor and other lung tissues during the CT scanning session. We are still limited by the treatment planning system's input requirements for a series of breathing-phase defined images. Work is ongoing to develop treatment planning protocols that better match the data resulting from the model-based approach. PO-0884 Availability of MRI improves interobserver variation in CT-based pancreatic tumor delineation O.J. Gurney-Champion 1 , E. Versteijne 1 , A. Van der Horst 1 , E. Lens 1 , H. Rütten 2 , H.D. Heerkens 3 , G.M.R.M. Paardekooper 4 , M. Berbee 5 , C.R.N. Rasch 1 , J. Stoker 6 , M.R.W. Engelbrecht 6 , M. Van Herk 7 , A.J. Nederveen 6 , R. Klaassen 8 , H.W.M. Van Laarhoven 8 , G. Van Tienhoven 1 , A. Bel 1 1 Academic Medical Center, Department of Radiation Oncology, Amsterdam, The Netherlands 2 Radboud University Medical Center, Department of Radiation Oncology, Nijmegen, The Netherlands 3 University Medical Center Utrecht, Department of Radiotherapy, Utrecht, The Netherlands 4 Isala Clinics Zwolle, Department of Radiotherapy, Zwolle, The Netherlands 5 MAASTRO Clinic, Department of Radiation Oncology, Maastricht, The Netherlands 6 Academic Medical Center, Department of Radiology, Amsterdam, The Netherlands 7 University of Manchester and Christie NHS trust, Faculty of Biology- Medicine & Health- Division of Molecular & Clinical Cancer Sciences, Manchester, United Kingdom 8 Academic Medical Center, Department of Medical Oncology, Amsterdam, The Netherlands Purpose or Objective To assess whether the availability of magnetic resonance images (MRIs) alongside the planning CT scan for target volume delineation in pancreatic cancer patients decreases interobserver variation. Material and Methods Eight observers (radiation oncologists) from six institutions delineated gross tumor volume (GTV) on contrast- enhanced (CE) 3DCT and internal GTV (iGTV) on 4DCT for four pancreatic cancer patients. At least six weeks after submitting these delineations, the observers were asked to repeat the delineations, now with MRIs available in a separate window (3DCT+MRI and 4DCT+MRI). The MRI included plain and CE T1-weighted gradient echo, T2- weighted turbo spin echo, and diffusion-weighted imaging. Interobserver variation in volumes of (i)GTVs was analyzed. Also, the generalized conformity index (CI gen ), a

measure of overlap of the delineated volumes (1=full overlap, 0=no overlap), was calculated. Furthermore, the local observer variation was calculated for approx. 32,000 points on the median delineated surface (i.e. the surface of the volume that ≥50% of the observers included in their delineation). Local observer variation was defined as the standard deviation (SD) over the perpendicular distances between delineated surfaces at that point and is also known as local SD. The overall observer variation was defined as the root-mean-square of all local SDs. These parameters were compared between CT-only and CT+MRI delineations, for 3DCT and 4DCT (Wilcoxon signed-rank test; significance level α=0.05). Results Delineations differed substantially between observers in both CT and CT+MRI, as illustrated for the GTV in the figure. For both GTV and iGTV, the mean volume on CT+MRI was 32% smaller than on CT only (p<0.0005) (Table). Although smaller volumes were delineated on CT+MRI, the CI gen was similar in both studies (CT+MRI: 0.33, CT: 0.32). Furthermore, CT+MRI showed smaller overall observer variations (average SD=5.9 mm) in six out of eight delineated structures compared to CT only (average SD=7.2 mm). The median volumes from the (i)GTV on CT+MRI were included for 97% and 92% in the median volumes from GTV and iGTV on CT, respectively. Finally, iGTV delineation on 4DCT increased uncertainty with and without MRI, compared to GTV delineation on 3DCT. Both CT and CT+MRI delineations had regions of large local observer variation (SD>0.8) close to biliary stents and enlarged lymph nodes. This was largely due to ambiguous instructions (near stents) and poor protocol compliance (near lymph nodes).

Figure: GTV delineations by the eight observers (each a different color) for 3DCT+MRI and 3DCT.