ESTRO 2023 - Abstract Book

S2113

Digital Posters

ESTRO 2023

Table 1: Overall results of ART-NET DSC vs inter-expert DSC per organ

Figure 1: Example of non-edited automatic contours (blue) in comparison to contours from 2 experts (orange in red) for a pelvis male patient. Conclusion This study demonstrated the feasibility of using a commercially available auto-contouring tool to generate clinically acceptable contours on synthetic CTs generated from CBCT images. This is an important step towards automated and AI- based offline adaptive workflow. Future work will include an end-to-end evaluation of the tool for offline adaptive radiotherapy.

PO-2349 implementation of daily adaptive workflow with Elekta Unity for head&neck and glioblastoma patients

L. Spiazzi 1 , M. Buglione 2 , S. Nici 3 , A. Guerini 4 , S. Riga 3 , C. Cozzaglio 3 , D. Farina 5 , F. Vaccher 6 , S.M. Magrini 2 , R.G. Pellegrini 7

1 ASST Spedali Civili di Brescia, Medical Physics, Brescia, Italy; 2 University and ASST Spedali Civili, Radiation Oncology, Brescia, Italy; 3 ASST Spedali Civili, Medical Physics, Brescia, Italy; 4 ASST Spedali Civili, Radiation Oncology, Brescia, Italy; 5 University and ASST Spedali Civili, Radiology, Brescia, Italy; 6 University of Brescia, Radiology, Brescia, Italy; 7 Elekta AB, Medical affairs, Stockholm, Sweden Purpose or Objective The implementation of an MR-based Adaptive Workflow is described for Head and Neck (H&N) and Glioblastoma Patients (GBM), where daily Adapt to Shape (ATS) and Adapt to Position (ATP) are performed with Elekta Unity. Materials and Methods For Automated Contouring 9 H&N CT atlases have been created with Deep Learning Models and manual recontouring and the addition of 2 Air Cavities levels, 2 Bone levels and low density tissue created by histogram conversion. The nomenclature of AAPM TG263 was utilized over 38 structures with a numerical prefix specifying the Layer Priority of each contour for propagation onto MR studysets for creation of synthetic CTs (sCTs) Each reference CT is automatically contoured for OARs by Atlas Based Autosegmentation with Elekta ABAS. Contours are then propagated to the reference MR-set with the Adapt Anatomy Option of Elekta Monaco. This DIR function automatically assigns bulk electronic densities and layer priority to each organ, thus creating planning sCTs. For the reference Offline Plan, during the first phase of the optimization Monaco Multi Criteria (MCO) is active for all the tissue near PTV, whereas in the second phase of Segment Shapes Optimization it is disabled and the cost function weights are locked for direct manual correction of the scorecards out-layers. This plan is always compared to a Helical Tomotherapy (HT) as reference and back-up plan. During the On-line session a daily MR is acquired to assess shifts and adaptive strategy. ATS strategy is the mostly utilized one for its better control of organ daily deformation and planning scorecards, despite a slightly longer optimization time. For small corrections ATP can be used. The plan is finally approved and sent to MOSAIQ for the delivery. More sequences can be acquired online for evaluation of disease progression and organs at risk response. Each fractional dose is warped and accumulated onto the original Reference set to evaluate possible degradation of dosimetry during the adaptive treatment. Results ABAS for Automated contouring and layering has decreased the offline contouring times and lowered possible errors in the creation of sCTs. The creation of MCO-based plans is competitive with HT Plans and the use of the cost function weights blocking reduced a conventional H&N plan to less than 1 minutes. GBM cases achieve planning faster and more automated convergence for their lower clinical complexity and plan modulation. The creation of such a workflow has automated most contouring and planning steps, leading to online sessions in the range of 50-60 min and 30-40 min respectively.