ESTRO 2021 Abstract Book

S1386

ESTRO 2021

Conclusion With our interactive DL segmentation tool, we have shown that we can substantially improve the segmentation accuracy with limited adaptations. Further improvement of the update speed is needed for practical usability.

PO-1664 Direct Density algorithm evaluation M. Garcia Cutillas 1 , B. Fornes García 1 1 Hospital Clinic de Barcelona, Radiation Oncology, Barcelona, Spain

Purpose or Objective Computed tomography imaging is usually used in radiation oncology to calculate radiation dose distributions. In order to do so, HU must be converted into electron densities by an energy dependent conversion curve, therefore dependent on the X-ray tube voltage. As far as variation on tube voltage among several acquisitions is needed in order to maximize image quality and minimize patient dose, it would be desirable, for the sake of simplicity and workflow robustness, to have a single calibration curves for all tube voltages. In the present study, we analyze HU variation as function of X-ray tube voltage among several materials on a phantom and different acquisition protocols, using reconstruction algorithms with and without direct Density (DD) algorithm. The aim of this study is to test HU independence when varying kV tube. It potentially enables the use of a single calibration curve when calculating dose distributions. It is then, the first step to implement the DD algorithm in clinical routine. Materials and Methods • CIRS CBCT Electron Density Phantom Model 062MA is made up of two structures to simulate head and body. Employed inserts were lungInhale, lungExhale, adipose, breast, plastic water, muscle, liver, bone200, bone800 y bone1250 with relative electron densities (RED) showed in table 1. • Serial images were acquired in a TC Siemens Somatom Go.Open Pro (Siemens Healthineers) which incorporates DD. HU analysis were performed using Eclipse TPS version 13.7 (Varian Medical Systems) Variations of tube voltage from 70 kV up to 140 kV in 10 kV increments were made. Tomographic reconstruction was implemented using standard and DD kernels. For determining HU of each material, a 2 cm ROI centred in each insert was created, making sure the ROI did not cover the edge of the insert. Calibrations curves were introduced into the TPS using the average of all kV voltages in the case of implementation of DD workflow. Results Adjusting all the points representing HU values for all tissues, kV and protocols to a line regression, we obtain a coefficient of determination R 2 =0,995, showing a good correlation of the data. Figure 1 shows the different calibration curves obtained using standard and DD reconstructions kernels when varying tube kV. It can be seen how DD curves tend to converge in a single one. Table 1 shows the different RED inserts and the HU value for each one of the inserts when averaging for all the tube voltages used. HU standard deviation is displayed showing how, when using DD reconstruction,It does not rise as much as in the case of high electronic density materials, such as bone inserts.