ESTRO 2020 Abstract book

S1054 ESTRO 2020

Segmentation of OARs in radiation therapy (RT) is tedious and time‐consuming. However, it is a necessary step in estimating the OARs toxicity and a high segmentation accuracy is required. Recently many commercial deep learning‐based auto segmentation solutions have appeared with acceptable performance. Among them, AccuContour™ from Manteia is a promising one, but not many independent validations of this software have been reported. The AccuContour software comes with models trained on the Vendors proprietary datasets for various body sites. In this work, 20 abdominal CT and 20 head and neck CT datasets of patients who underwent RT were run on AccuContour™ to auto segment normal organs which were then compared to manually segmented organs. The Dice Similarity Coefficient (DSC) was applied to evaluate the degree of concordance between the two sets. The performance of AccuContour™ was evaluated on three OARs from the abdominal datasets (liver, kidney (left and right) and esophagus) and four OARs from the Head & Neck datasets (brain, brainstem, parotid (left and right) and submandibular gland (left and right)). Results The auto segmentation process was completed in less than two minutes for each dataset which represents a significant reduction over the manual process which can take between 30‐60 minutes. The mean and standard deviation of DSC for each OAR structure is listed in table 1 and the boxplot for each OAR is plotted in Figure 1. From these tables, it is seen the accuracy of auto segmenting for various organs is: brain>liver>kidney>brainstem>parotid>esophagus>subman dibular gland. The outliers seen on the boxplots are due to incomplete patient datasets. Overall, the software produced smooth, regular segmentation of organs with correct edge delineation.

Conclusion The AccuContour™ software appears to be a useful tool to auto segment the structures tested in this work, although manual inspection and revision is still required. A blinded randomized test comparing auto segmented to manually segmented normal organ structures is planned to determine the clinical significance of differences between the two structure sets. PO‐1796 Framework for Competency and Performance Assessment in Radiation Oncology: MRIdian Linac Case Study C. Kota 1 1 Reading Hospital. Tower Health, Radiation Oncology, West Reading, USA Purpose or Objective To create a framework for assessing the knowledge, competency, performance and action of staff in Radiation Oncology while introducing a significantly different technology and workflow into the clinic with the ViewRay MRIdian Linac. Material and Methods The pyramid structure in Framework for clinical assessment introduced by Miller (The Assessment of Clinical Skills, Miller, 1990) is proposed to create a comprehensive framework for staff training while introducing the MRIdian Linac into the clinic. In the current context, the first base level, “Knowledge” is defined as the collection of basic facts and understanding of a). MR imaging and treatment on the MRIdian and b). adaptive treatments. The next higher level “Competency” is defined as the ability to draw on this knowledge to formulate a solution for individual patient treatments under various “live” situations. The third level “Performance” is defined as the ability to demonstrate the Competency on phantoms and volunteers in real time. The final level at the apex of the pyramid “Action” is the execution of Performance in the real world setting on actual patients. The roles and responsibilities of each member of the treatment team (Radiation Oncologist, Medical Physicist, Dosimetrist, Therapist) for adaptive treatments on the MRIdian were analyzed. The specific knowledge,