Program Information
A Gaussian-Mixture Model Algorithm and Platform for HDR QA
N Thorne*, R He, C Yang, University of Mississippi Med. Center, Jackson, MS
Presentations
TU-L-GePD-T-1 (Tuesday, August 1, 2017) 1:15 PM - 1:45 PM Room: Therapy ePoster Lounge
Purpose: We purpose a method to perform HDR QA by tracking the source using a Gaussian-Mixture Model (GMM). We also provide an accompanying graphic user interface (GUI) for its implementation in the clinic that affords greater visualization and quantitative measurements of the delivery process. Using this system, we can monitor source position, dwell time, and activity; quantify uncertainty in applicator commissioning; perform HDR patient treatment plan QA; and construct structured data sets that could be used for machine learning.
Methods: Data were collected using an IBA MatriXX ion chamber array system and OmniPro software in movie mode and exported as a Dicom file to a custom MATLAB script. For each time frame, these data were normalized by the script and a GMM algorithm was implemented to ascertain the position, which was used to determine placement accuracy. Dwell times were assayed by counting the number of frames that the source was monitored in each position. Activity was measured from the amount of dose that was collected by the array system within a given time frame at a fixed position. Measurements were compared to EBT3 film. The patient HDR QA is performed using a Tandem and Ring plan. The algorithm was implemented with a custom GUI for daily QA that would store the data in structured data sets for data mining.
Results: The GUI noticeably increased the speed of daily QA, while producing structured data sets for measurements. A precision of 0.32 mm was observed for position accuracy measurements. Time was measured to within 50 ms. The results are in agreement with the film measurements.
Conclusion: The system may serve as an improvement over existing QA and commissioning practices as it allows for the establishment of structured data sets and direct visualization of the source motion.
Contact Email: