ESTRO 36 Abstract Book

S435 ESTRO 36 2017 _______________________________________________________________________________________________

3 University of Santo Tomas, Department of Radiation Oncology, Manila, Philippines 4 MD Anderson Cancer Center, Department of Radiation Oncology, Houston, USA Purpose or Objective To fully automate radiotherapy planning for cervical cancer (4-field box treatments) and head/neck cancer (VMAT/IMRT). Material and Methods We are using a combination of in-house software, Eclipse Treatment Planning System, and Mobius 3D to create and validate radiotherapy plans. Most planning tasks have been automated using a primary algorithm for the treatment plan, and a secondary independent algorithm to verify the primary algorithm. The first step is to automatically determine the external body surface and isocenter (based on radiopaque markers in a 3-point setup) using two independent techniques. For H/N cases, the radiation oncologist manually delineates the GTV. Normal tissues (parotids, cord, brainstem, lung, eyes, mandible, cochlea, brain), cervical neck nodes (levels II-IV, IB-V or IA-V) and retropharyngeal nodes are automatically delineated using an in-house multi-atlas segmentation tool. The RapidPlan tool (Eclipse) is used to create a VMAT plan. For 4-field box cervical cancer treatments, the field apertures (jaw and MLC positions) are automatically calculated based on bony anatomy using two techniques: The primary technique uses atlas-based segmentation of bony anatomy, and then calculates apertures based on the projection of these bones to each beam’s-eye-view. The secondary technique deformably registers atlas DRRs to the patient’s DRR for each beam, then uses the deformation matrix to deform atlas blocks (MLC positions) to the patient’s DRR. Relative beam weighting is determined based on a least-squares fit, minimizing heterogeneity in the treatment volume. Final dose distributions are automatically sent to Mobius for secondary dose calculation. Results Primary and secondary techniques for identifying the body surface agreed within 1.0mm/0.99 (mean distance to agreement/average DICE coefficient). Primary and secondary techniques for determining isocenter agreed within 3mm. H/N normal tissue and lymph node segmentation was evaluated by a radiation oncologist (128 patients), and found to be acceptable for all structures, except for esophagus and cochlea and in situations where the head position was non-standard. The figure below shows a fully automated plan including contours and optimized doses.

In terms of robustness evaluation, PTV-based MFO showed reduced robustness against both anatomical changes and uncertainties, i.e. wider DVH bands and a disagreement between planned and summed dose, whereas the robust MFO is less influenced. Both SFO approaches resulted in robust plans on the CTVs (Figure 1).

Conclusion The PTV-based MFO approach showed less robustness against uncertainties in setup and range, as well as for anatomical changes during the treatment course. Both SFO plans are robust in terms of CTV coverage; however, they present higher doses to the ipsilateral parotid gland. Robust MFO approach presents the lowest doses to the ipsilateral parotid and more robustness against uncertainties. The dose to more organs at risk and the difference in normal tissue complication probabilities for the 4 planning approaches will be presented as well. PO-0820 Full automation of radiation therapy treatment planning L. Court 1 , R. McCarroll 1 , K. Kisling 1 , L. Zhang 1 , J. Yang 1 , H. Simonds 2 , M. Du Toit 2 , M. Mejia 3 , A. Jhingran 4 , P. Balter 1 , B. Beadle 4 1 MD Anderson Cancer Center, Department of Radiation Physics, Houston, USA 2 Stellenbosch University, Radiation Oncology, Stellenbosch, South Africa