Click here to

Session: Machine Intelligence Efficacy and Quality I [Return to Session]

Development and Maintenance of a Knowledge-Based Planning Quality Assurance Tool for a Statewide Radiation Oncology Quality Consortium

C Matrosic1*, K Dess1, K Torolski1, R Marsh1, M Grubb1, D Dryden2, D Jarema1, C Fraser3, P Paximadis4, M Dominello5, J Hayman1, S Jolly1, M Matuszak1, (1) University of Michigan, Ann Arbor, MI, (2) Covenant HealthCare, Boyne City, MI, (3) Henry Ford Health System, Detroit, MI, (4) Spectrum Health Lakeland, St. Joseph, MI, (5) Wayne State University School Of Medicine, Detroit, MI

Presentations

MO-A-BRC-2 (Monday, 7/11/2022) 7:30 AM - 8:30 AM [Eastern Time (GMT-4)]

Ballroom C

Purpose: Knowledge-based planning (KBP) has shown to be valuable for treatment plan quality assurance (QA). One challenge of KBP is the creation and maintenance of models can be complex and time consuming. Leveraging large, well-maintained databases and streamlining the update process could potentially overcome this problem. This work discusses the experience of updating a lung KBP model originally developed in 2017 based on recent statewide quality consortium data.

Methods: 60 lung cancer patient plans treated during 2021 that met consortium planning goals were collected from 24 institutions with a variety of planning systems. Plan and contour quality were reviewed over 6 hours of video conferences by a group of physicists and dosimetrists, resulting in 47 high quality plans that were used to retrain the 2017 model with an updated cost function. The model was tested by creating VMAT plans for the remaining 13 patients and 7 2017 QA patients and DVH metrics were compared between the current model plans, clinical plans, and 2017 model plans.

Results: The consortium’s database filter for specific DVH goals resulted in high quality training plans. 60%, 65%, and 65% of the updated model plans improved the lung V20Gy, mean heart dose, and spinal canal D0.03cc from clinical plans, respectively. The new model improved hot spots and spinal canal doses at the cost of slightly higher lung doses compared to the 2017 model. Difficulties discovered during the update were inconsistencies in contour naming conventions, contouring methods, and hot spot planning constraints amongst institutions, causing potential model variance.

Conclusion: A multi-institutional KBP model was updated using plans from a statewide consortium and when applied, demonstrated improvements in independent plans. With the implementation of additional policies and further web-based plan review tools, the model update process could be further streamlined, resulting in a valuable statewide plan QA tool.

Funding Support, Disclosures, and Conflict of Interest: MROQC is financially supported by Blue Cross Blue Shield of Michigan and the Blue Care Network of Michigan as part of the BCBSM Value Partnerships Program. Drs. Matrosic and Matuszak have been consultants and Dr. Jolly participated in an advisory board for Varian Medical Systems.

Keywords

Treatment Planning, Quality Assurance, Radiation Therapy

Taxonomy

TH- External Beam- Photons: Treatment planning using machine learning/Knowledge Based Planning/automation

Contact Email

Share: