ESTRO 2020 Abstract Book

S1095 ESTRO 2020

whereas the PPNN aiming at distinguishing real plans approved by the physician and plans generated by DPNN. We trained our system using 200 plans from 41 patients, and tested the performance on 20 testing plans from 4 patients not seen by the training process.

and diagnosis, are listed in blue to remind the checking physicist of the few checks not performed by BrachyVerification that must be performed manually. The second tab (Figure 2) compares the planning system calculated dose for five points placed during planning with a dose calculated by the software using TG-43 formalism. Discrepancies of <5% are highlighted in green, and greater discrepancies are highlighted red. The third tab creates a treatment record document using values from the treatment plan.

Results DPNN was able to predict EQD2 with an error of 8.1%±6.4%, 6.7%±5.2%, 6.9%±3.4%, and 3.8%±2.7% for bladder, rectum, sigmoid, and CTV, respectively. For PPNN, the sensitivity, specificity, and accuracy were 81%, 77%, 80% among the real and generated data. The error was partially ascribed to suboptimal quality of the testing plans that were generated in real clinic under time pressure. Conclusion We developed a novel model that predicts a physician’s intention when approving a treatment plan. Performances will be improved in future with the inclusion of more and high-quality plan data. DPNN and PPNN will be incorporated in our automated treatment planning workflow to provide physician-like guidance. OC-1043 Automated Plan Verification Software for Ultrasound-Planned High Dose Rate Prostate Brachytherapy T. HarriS 1 , C. Molodowitch 1 , E. Sugar 1 , D. O'Farrell 1 , M. King 1 , I. Buzurovic 1 1 Dana Farber/Brigham and Womens Cancer Center, Radiation Oncology, Boston, USA Purpose or Objective Ultrasound-based planning for high dose rate (HDR) prostate brachytherapy offers potential advantages over CT-planned HDR cases, such as completion of the brachytherapy treatment without moving the patient from the time of needle placement to dose delivery. Due to the need to minimize patient time under anesthesia, our clinic developed rapid and accurate software targeted specifically to ultrasound planning performed in the OR, in order to automate treatment plan checking by the physicist. Material and Methods BrachyVerification software is a stand-alone Java program (figure 1). The software accesses the Aria (Varian Medical System, Palo Alto, CA) SQL database, and the user then selects the appropriate course and intent. The physicist then uploads the written directive (WD) and a RTPlan DICOM file of the planned treatment, generated in Oncentra Prostate (Elekta, Stockholm, Sweden). The software parses the WD using the known format of our directives, then reports on the parsing success vs. failure of the WD and DICOM plan. The first tab (Figure 1) displays the results of 13 different verifications: WD dose/fraction matches Aria, WD dose/fraction matches intent, WD approved, WD patient information, Oncentra patient information, WD matches Oncentra, Oncentra dose/fraction and number of fractions matches Aria, Oncentra plan approved, catheter lengths, catheter channels, Aria dose/fraction matches intent, Aria activity x times value matches Oncentra to within 3%, and known Oncentra bugs. Passed checks are green; failed checks are red and moved to the top of the list to ensure visibility. Several additional tests, such as site, applicator,

Results BrachyVerification has been used for 35 ultrasound- planned HDR implants performed on 25 different patients since May of 2019. Defining a false positive as an erroneously red-flagged check and a false negative as an erroneously green-passed check, a retrospective review of the implants finds that the software reported 0 false positives and 0 false negatives compared to manual checking. Conclusion We developed and implemented software that automates most of the physics checks for ultrasound-planned HDR prostate cases. The expectation is that the software will reduce total physics time for cases, decreasing the time the patient spends under anesthesia, while ensuring accurate checks. A time study is planned to quantify any increased efficiency. With small modifications, the software can be adapted to CT-planned HDR prostate cases, as well as other HDR sites. OC-1044 Catheter Reconstruction Limits of an Afterloader With EMT Capabilities D. Tho 1,2 , L. Beaulieu 1,2 1 Université Laval, Département de physique de génie