ESTRO 38 Abstract book

S1129 ESTRO 38

patients’ sizes differ from the size of the calibration phantom used to generate the curves. EP-2053 Pelvic plan adaptation to manage systematic rotations without CT re-imaging A. Licup 1 , S. Van Kranen 1 , M. Buijs 1 , F. Koetsveld 1 , J. Sonke 1 , P. Remeijer 1 1 Netherlands Cancer Institute, Radiotherapy, Amsterdam, The Netherlands Purpose or Objective In IGRT workflows for pelvic treatments, rotations often remain uncorrected. Especially for large target volumes (TVs), e.g. cervical tumors that include pelvic and para- aortic lymph nodes, large CTV to PTV margins are required to account for the residual errors. Replanning on a repeat CT (rCT) scan may correct for systematic rotations, however this increases clinical workload and patient imaging dose and may still fail to image the systematic rotational error. Therefore we propose to simulate an rCT scan based on the clinical (rigid) registrations of TVs in CBCT scans. To overcome large displacements outside the registered region, often seen as considerable displacements of the patient skin contour, we developed and validated a technique that restricts skin deformations while correcting for the observed systematic rotations in the pelvic area. Material and Methods To simulate an rCT scan, we propose to use compactly- supported radial basis functions (CSRBF) [1]. CSRBF is a method in which rigid registrations are confined in a local region, outside of which non-rigid deformations reduce smoothly to zero within a set RBF radius. Figure 1 illustrates this method for a cervical cancer patient model.

Results The CT number variance was very small for the low- to medium-density inserts and could be assumed to originate from CT noise rather than BH. In contrast, large differences were seen for the CT number of the high- density bone inserts. The lowest BH effect was found at slightly different monoEs for the four high-density bone inserts. The average optimal monoE was found at 90 keV. At this monoE, the BH was at least reduced by half compared to SECT. The resulting calibration curves for the small and large phantom differed only slightly at very high monoenergetic CT numbers, above the clinically relevant region (Fig. 2)

We validated the simulated rCT scan based on a single fraction ( N =1) on the following aspects: (1) skin contour preservation, (2) interior pelvic anatomy rigid registration, and (3) delineation shape preservation. Results Offline simulation of the rCT took 6mins on a 64-bit 3.40GHz Intel® Xeon® PC. Figure 2A shows a translation- only registration between the planning CT (pCT) and a single CBCT with a large rotation, the effect of uncorrected rotations are clearly seen. Figure 2B shows an example where we simulate an rCT on the same CBCT. The patient outline is preserved while the internal anatomy rotations are corrected. Figure 2C is a grey level difference image between pCT and rCT registered on the skin. We evaluated the root mean squared differences (RMSD). An RMSD≈0 suggests high similarity. Figure 2C shows minimal RMSD outside the ROI, validating (1). To validate (2), the entire rCT was rigidly registered back to the pCT (figure 2D), with minimal RMSD inside the ROI. Figure 2E summarizes the RMSD statistics. Finally, to

Conclusion Using monoenergetic images at 90 keV derived from TB DECT scans, the BH effect can be markedly reduced compared to SECT. The difference between the calibration curves for the small and large phantom was minor, which would decrease the SPR deviation when