ESTRO 37 Abstract book

ESTRO 37

S481

plans. A 90% confidence level was used as action level to identify plans that could possibly be improved. Next parameters of the Pinnacle AP were adjusted to generate treatment plans that met or surpassed the predicted doses. Results Of the 50 validation patients, plans of 6 patients (9 DVH points) exceeded the action level. Of these 6 patients one had a double sided hip prostheses and an alternative, non-AP, planning strategy was used clinically. The other 5 cases (see table) could all be improved such that all dose metrics became within the predicted 90% CIs. Excluding these outliers, the mean difference (standard deviation) between the achieved value and the predicted expectation was: rectum Dmean: -0.14 (1.4) Gy; Rectal wall V75Gy: 0.4 (2) %; Rectal wall V65Gy: 0.7 (2) % ; anal sphincter Dmean -0.4 (1.2) Gy. Conclusion Independent KB treatment planning QA was used successfully to assess the quality of clinical Pinnacle AP plans. 6 Outlier patients were detected, including two large outliers (difference ≥5Gy / 5%), that could all be improved with re-planning. Apart from these outliers, predictions where accurate with the mean difference well within one standard deviation. Pinnacle AP created acceptable treatment plans for the far majority of patients, but suboptimal plans did occur. Therefore in clinical practice we recommend the use of independent treatment plan QA in combination with Pinnacle AP. [1] Y. Wang, B. Heijmen, S. Petit Prospective clinical validation of independent DVH prediction for plan QA in automatic treatment planning for prostate cancer patients. Accepted for publication by R&O. PO-0903 Evaluation of a four-pi approach to stereotactic treatment planning in cranial SRT J. Robar 1 1 Dalhousie University, Radiation Oncology, Halifax, Canada Purpose or Objective The objective of this study was to evaluate a new four-pi approach for treatment planning of skull-based indications for stereotactic radiotherapy, and to compare the dosimetric performance to that from standard VMAT Twenty patient treatment plans were performed, ten for patients with vestibular schwannoma, and ten with pituitary adenoma. A novel and recently commercialized four-pi algorithm (Cranial SRS Element, Brainlab AG) was used to automatically determine gantry start/stop angles as well as treatment couch positions, without user intervention by the planner. This algorithm begins with a user-defined template and adjusts couch angles as well as gantry arcs within limits specified by the user by minimizing i) overlap of PTV and OARs in the beam’s-eye- view and ii) radiological depth. The cases were re- planned using standard template VMAT arc geometries. Treatment plans were compared with regard to PTV dose homogeneity, organ-at-risk (OAR) sparing and monitor unit efficiency. planning geometries. Material and Methods

Eclipse 70.1±9.4% The table reports the results of PQM% for all the KPB models and TPS. While maintaining the same overall plan quality (p=0.657), as measured by PQM%, Eclipse tends to favor a better PTV coverage and homogeneity at the cost of a reduced OARs sparing if compared to RayStation and Pinnacle (see Figure). The differences among the outcomes of the three models were not statistically significant (p=0.492), but a bias toward better performances was observed when Rapid Plan predictions were based on plans optimized with Pinnacle. Model #1

Conclusion This work proved that RapidPlan predictions can be used to drive the optimization on different TPS installed in different Institutions with similar results. PO-0902 Independent knowledge-based treatment planning QA to audit Pinnacle Autoplanning T.M. Janssen 1 , Y. Wang 2 , G. Wortel 1 , E. Damen 1 , S.F. Petit 2 1 Netherlands Cancer Institute, Radiotherapy, Amsterdam, The Netherlands 2 Erasmus MC Cancer Institute, Radiotherapy, Rotterdam, The Netherlands Purpose or Objective With the advent of automatic treatment planning (AP) options like Pinnacle’s Autoplanning technique, the challenge arises how to assess the quality of a plan that you did not work on. In addition, since Autoplanning is basically a black box, independent quality assurance of treatment plans may be necessary. At our institute Pinnacle AP is used in routine practice since March 2016. In this work we study the need for independent treatment planning QA for Pinnacle AP. For this purpose a knowledge based (KB) model was trained and validated to detect outliers in plans quality. Subsequently we studied if these outliers indeed could be improved. Material and Methods 100 recent clinical prostate cancer patients were included, receiving 35x2.2 Gy either on the prostate, or on the prostate plus seminal vesicles. Patients were treated with VMAT plans that were generated using Pinnacle’s AP (v9.10) using a fixed AP technique. Patients were divided 50-50 in a training and validation group. The training patients were used to train a recently published KB treatment planning QA model, based on Overlap Volume Histograms and Principle Component Analysis [1]. The KB model was used to predict for the validation patients the achievable V65Gy and V75Gy of the rectal wall and Dmean in the anal sphincter and full rectum. The dose predictions were compared to the AP