ESTRO 2023 - Abstract Book

S349

Sunday 14 May 2023

ESTRO 2023

To evaluate the accuracy of our sCT, mean absolute error (MAE), mean error (ME) and peak signal to noise ratio (PSNR) were computed. Results Table 1 shows MAE, ME and PSNR for each training cohort (monocentric and multicentric). No significant differences were found between monocentric trainings and multicentric trainings where some of the test and training data come from the same center.

Figure 1 shows some sCT results from the second center after three different types of training.

Conclusion To conclude, multicentric context doesn’t penalize trainings when both centers are in the training cohort. The learning cohort should include data from various acquisition devices to generalize our deep learning approach for use in all cancer centers. OC-0447 CBCT-based synthetic CT scans for tumor delineation of bone metastases in the pelvis N. Hoffmans-Holtzer 1,1 , A. Magallon-Baro 1 , I. de Pree 1 , C. Slagter 1 , M. Olofsen-van Acht 1 , M. Hoogeman 1 , S. Petit 1 1 Erasmus MC Cancer Institute, University Medical Center Rotterdam, Department of Radiotherapy, Rotterdam, The Netherlands Purpose or Objective Patients treated for bone metastasis with palliative RT are often frail and in pain. Hence, these patients would benefit from (1) minimal time delay between intake and start of treatment, and (2) minimal movements between hospital bed and treatment couch. This may be achieved with one-table treatment, i.e.: imaging for simulation, treatment planning and delivery, all occurring while the patient remains on the same treatment couch. Given recent developments in automated contouring and treatment planning, the prime bottleneck for one-table treatments is the image quality of regular CBCT scans. The aim of this study was to evaluate whether the Elekta deep learning (DL) algorithm could yield synthetic CT (sCT) scans of sufficient quality for accurate tumor delineation of bone metastasis, which often present soft tissue infiltrations beyond the bone boundaries. Materials and Methods A total of 22 CBCT and planning CT scans from 22 consecutive (and eligible) female patients with bone metastasis were included. For each CBCT, a corresponding sCT was created by using the female pelvis DL model v2 in ADMIRE 3.38.0. Three Radiation Oncologists (ROs) manually delineated 23 GTVs on the sCTs (GTVsCT) and gave a confidence score (0-10) for the sCT quality. Based on this score, scans were divided into an insufficient quality category (score<5.5) and a sufficient quality category (score ≥ 5.5). Next, the contours were propagated to the planning CT scan and manually adjusted if needed (GTVclin; the gold standard). Geometric accuracy of the GTVsCT was evaluated by comparing with the GTVclin using the Dice coefficient (DC) and Hausdorff distance (HD). The dosimetric effect of any GTVsCT inaccuracies was investigated by expanding the GTVsCT by different GTV to PTV margins, varying from 8 mm (current clinical standard) to 15 mm in 1 mm steps. For each PTVsCT, an automated VMAT treatment plan was generated; 184 plans in total (23 tumors x 8 plans). For each plan, the target coverage goal of V95%>98% was evaluated on the gold standard PTV (i.e.: PTVclin). Results For 13 tumors, the sCT quality was sufficient (Fig.1 - Tumors A and B) and for 10 insufficient (Fig.1 - Tumors C and D). For the sufficient category, the DC and HD were excellent (DC = 0.9 (range 0.9-0.9), HD = 11 (range 9-13) mm) while for the insufficient category they were considerably worse (DC = 0.7 (range 0.6-0.8), HD = 34 (range 27-51) mm). PTV coverage as function of PTV margin is shown in Fig.2. For the sufficient category, a median of only +1 mm additional margin was required to achieve good PTVclin coverage. For the insufficient category, however, even +7 mm margin (15 in total) was not sufficient to achieve good PTV coverage for 9/10 tumors.