ESTRO 36 Abstract Book

S825 ESTRO 36 _______________________________________________________________________________________________

Material and Methods Atlas Based Segmentation (ABS) was utilized for automatic volume delineation. Volume metrics and DICE coefficient scores were compared between multiple manual delineation, ABS and ABS with post processing. Automatic planning was achieved by python code in the Raystation treatment planning environment. Initial optimization objectives were determined by a min- difference optimization of database entries from previous clinical plans. Plan evaluation checks against both standard guidelines and previous plan quality scores are produced through python code and inbuilt look up tables. Non-linear scoring systems are incorporated for total plan scores that provide score weighting to crucial structures. Adaptive planning and dose tracking is achieved in a Varian-Mosaiq-Raystation environment. In all case time measurements were use to provide comparisons between manual and automated processing of typical radiotherapy planning tasks. Results Contouring of DICE scores showed strong agreement (over 0.90) for the vast majority of regions of interest, with an average DICE coefficient of 77.7 for breast patients and 81.8 for prostate. Results were improved with post processing. Breast and prostate plans show comparable plan quality with manual planning for both simple single phase and advanced 3-4 target volume techniques with dose volume histogram differences consistently within 5% TD Point to point comparisons between automatic deformation matches and manual user deformations showed varying results highlighting the current need for visual QA of deformable registrations. Time improvements over manual processes are recoreded for both breast and prostate patients in all areas of testing Conclusion The possibility of automation to provide efficiency and consistency on a departmental and larger scale is demonstrated. The work represents a step in the correct direction rather than a finished produce to radiotherapy automation and the current limitations and problems are opened to the audience for responses and questions. EP-1535 knowledge based planning for lung cancer patients with stereotactic ablative radiotherapy S. Smith 1 , P. Houston 1 , G. Currie 1 1 NHS Greater Glasgow & Clyde, Radiotherapy Physics, Glasgow, United Kingdom Purpose or Objective To determine whether a knowledge based treatment planning system can efficiently produce VMAT plans for lung cancer patients treated with SABR and to assess plan quality in relation to different training plan data and model parameters. Material and Methods Three Lung SABR models were developed using the Varian RapidPlan™ DVH estimation algorithm. 1. Model A was trained using 60 plans calculated with both standard and HD MLCs, 6MV flattened /10 MV flattening filter free (FFF) utilising one algorithm (AAA version 10.0.28) with no plans excluded.

generate plans for ten patients who had not been included in the model training process. Comparisons of plans generated by each RapidPlan model with corresponding clinical plans were performed against clinical objectives. Clinical plans were generated by an experienced physicist using 10MV FFF, HD MLC and AAA (version 13.6.23).

Results All of the plans generated by each model met the clinical objectives. The PTV V99 (Volume of the PTV receiving at least 99% of the prescription) was comparable between all three model plans and the plan generated by the experience physicist (p>>0.05). The R50 (Ratio of the 50% prescription isodose volume to the PTV) and D2cm (the maximum dose at 2cm from the PTV) values were significantly reduced when using the RapidPlan (p<0.05) for all three models. Model B gave the best results and was statistically better than model A (p<0.05). Model B also gave better results for the R100 (Ratio of the 100% prescription isodose volume to the PTV) than the experienced planner (P<0.05). This may be due to the differences in the out of field dose calculation between versions of AAA. OAR doses were comparable between all models and the experienced planner (p>0.05) Conclusion The RapidPlan™ system was able to generate clinically acceptable VMAT treatment plans for lung SABR patients, in a single optimisation, with comparable OAR sparing and equal or better plan conformity than the original clinically acceptable plans. allowing for improved consistency and efficiency in the treatment planning process. EP-1536 The advantages of Collimator Optimization for Intensity Modulated Radiation Therapy S. Pella 1 , B. Brian Doozan 2 1 21st Century Radiation Oncology- Florida Atlantic- Univeristy and Advanced Radiation Physics Inc., Radiation Oncology/Physics, Boca Raton, USA 2 Florida Atlantic University, Physics/Medical Physics, Boca Raton, USA Purpose or Objective The goal of this study was to improve dosimetry for pelvic, head and neck and other cancer sites with aspherical planning target volumes (PTV) using a new algorithm for collimator angles optimization for intensity modulated radiation therapy (IMRT). Material and Methods A retroactive study on the effects of the collimator optimization for 20 patients was performed by comparing the dosimetric effects, number of monitor units (MU), and the treatment delivery time from optimized plans in Eclipse version 11.0. Keeping all parameters equal, multiple treatment plans were created using four collimator angle optimization techniques: CA 0 , all fields have collimators set to 0°, CA E , using the Eclipse collimator angle optimization, CA A, minimizing the area of the jaws around the PTV, and CA X , minimizing the x-jaw gap. The minimum area and the minimum x-jaw angles were found by evaluating each field beam’s eye view of the PTV with ImageJ and finding the desired parameters with a custom script. The evaluation of the plans included the monitor units (MU), the maximum dose of the plan, the maximum dose to organs at risk (OAR), the conformity index (CI) and the number of split fields. Results Compared to the CA 0 plans, the monitor units decreased on average by 6% for the CA X with a p-value of 0.01 from an ANOVA test. The average maximum dose stayed within 1.1% between all four methods with the lowest being CA X . The maximum dose to the most at risk organ was best

2.

Model B included data from model A and a further 40 plans; 22 plans with statistical outliers were excluded. Model C included the data from model B. A further 78 plans calculated with a new algorithm version (AAA version 13.6.23) were added to the model. Statistical outliers were excluded.

3.

The resulting models were then used to