ESTRO 2022 - Abstract Book

S1379

Abstract book

ESTRO 2022

Here, the results are simplified to present only the best filter (or non-filter) that led to the maximum number of stable and robust RFs (different filters for different RFs families). Table 2 summarizes the percentage of best-filtered stable & robust RFs derived from T2w, ADC, and DCE-SUB maps.

% of stable features (% of robust features)

Augmentation Scenario

T2w

ADC

SUB wash-in

SUB wash-out

in-plane-random

99.06 (81.31)

98.13 (98.13)

97.20 (95.33)

98.13 (96.26)

in-plane-systematic

98.13 (48.60)

98.13 (90.65)

97.20 (94.39)

98.13 (97.20)

out-plane

100 (97.20)

100 (86.91)

100 (98.13)

100 (98.13)

in&out-plane-random

98.13 (79.44)

98.13 (82.24)

97.20 (92.52)

98.13 (95.33)

in&out-plane-systematic

98.13 (50.47)

97.20 (73.83)

96.26 (90.65)

97.20 (90.65)

Conclusion Variations in segmentation impact the stability of RFs. We propose a novel method that can be easily integrated into any pipeline, which can be used to identify RFs stable to such variability. Since this is a fully automated method, it avoids the need for multiple radiologists to annotate the dataset. The robust subset of features identified can be considered for further modeling or analysis.

PO-1596 Non-invasive measurement of the AIF for dynamic PET: Simulation of clinical workflow

Withdrawn

PO-1597 Evaluation of synthetic CTs generated from T2-weighted MRIs of prostate cancer patients

S. Andersson 1 , I. Steinseifer 2

1 RaySearch Laboratories, Research, Stockholm, Sweden; 2 Isala, Department of Radiation Oncology, Zwolle, The Netherlands Purpose or Objective In radiotherapy, MR imaging is used for delineation in an increasing degree due to its superior soft tissue contrast. However, CT images are still needed for treatment planning, as MR images lack tissue density information. Accurate MR-to-CT synthesis is a crucial step towards an MR-only workflow in radiotherapy. By removing the need of CT imaging, a clinic can both save time and get rid of potential MR-CT registration uncertainties. This work evaluates an algorithm for synthetic CT (sCT) generation, available in the research version of RayStation 10A (RaySearch Laboratories, Stockholm, Sweden). Materials and Methods This study included T2-weighted TSE MRI (VISTA) (Philips Healthcare, Best, The Netherlands) and CT Big Bore (Philips) pelvic images of 55 patients in treatment position. The MR and CT images were registered deformably to reduce the anatomical differences. The synthetic CTs were generated in a semi-supervised fashion, using a CNN architecture similar to CycleGAN but including an extra paired term to account for the existence of paired data in this context. 35 patients were used for training of the sCT model and 20 patients were used for evaluation of the resulting sCT images. Mean absolute errors (MAEs) were calculated between the Hounsfield units (HU) of the deformed CT and the sCT for each patient. The original VMAT treatment plans were recalculated on both the deformed CT and the sCT. The dose differences between the CT based and the sCT based dose distributions were calculated for D1, D2, D50, D95, D98, D99 and average dose, for the following ROIs: CTV, PTV, rectum, anal canal, bladder, left femoral head and right femoral head. A research version of the commercial treatment planning system RayStation 10A was used for both image conversion and dose calculation. Results The mean MAE between the deformed CT and the sCT was 41.5 ± 9.6 HU. Figure 1a shows the fusion between the MRI and the corresponding sCT, with a very good conformity of the structures. Figure 1b shows a fusion of the deformed CT and the sCT with minor differences. Furthermore, the dose distribution on the deformed CT and the sCT of one patient (c+e), as well as the DVH’s (d) and a dose difference map (f). The DVH’s overlay each other, no difference between the deformed CT and the sCT can be seen. In Table 1 the mean dose difference of the prescribed dose is shown in percent for all dose statistics. The dose comparison shows a very good agreement with the deformed CT, the mean dose differences were close to 0% and no standard deviation above 0.5%.