MENU

Click here to

×

Are you sure ?

Yes, do it No, cancel

Towards Quality Assurance for First AI-Driven Online Adaptive Radiotherapy Based On Failure Mode and Effect Analysis

J Booth1,2*, P Sibolt3, E Laugeman4, B Cai4, D Sjostrom3, S Mutic4, M Perez1, (1) Royal North Shore Hospital, Sydney, NSW, Australia (2)Institute of Medical Physics, School of Physics, University of Sydney, Sydney, Australia,(3)Radiotherapy Research Unit, Department Of Oncology, Herlev And Gentofte Hospital,Denmark, (4) Washington University School of Medicine, St. Louis, MO,

Presentations

(Sunday, 7/12/2020)   [Eastern Time (GMT-4)]

Room: AAPM ePoster Library

Purpose: To conduct failure mode and effect analysis (FMEA) of first AI-driven system with assessment of severity, detectability and likelihood to inform quality assurance measures ensuring safe delivery of AI-driven online adaptive radiotherapy.

Methods: The first AI-driven online radiotherapy systems are in use and contain several novel components, potentially requiring specific quality assurance measures, such as; automated treatment planning, AI-driven contouring on iterative CBCT, semi-automated deformable image registration, automatic online plan of the day, software-based independent monitor unit calculation of adaptive plan, and offline accumulated dose monitoring. Potential failures and their effects were estimated within an international consortium. The three clinics represented different staffing and workflows and three levels of system experience. A benchmarking table of descriptors from AAPM TG100 was supplied to assist with consistency of risk scoring for the scenario of current QA measures.

Results: A total of 170 potential failure modes were discovered, approximately 50 of these were given risk scores above 100. Each institution determined a different ranking order to highest scoring failure modes, however all agreed on the top 50 failure modes. Only 20% of these 50 failures modes were scored with detectability of 6 or higher, suggesting QA measures specific to adaptive radiotherapy may be required. The 3 main areas identified were; initial registration of CBCT, the associated contouring of targets and other structures using AI segmentation and the auto plan creation. The areas have been used to inform the type QA required. Detection of issues in the generation of the synthetic CT is challenging and the consortium is developing quality control steps mitigating these errors and the associated failure modes.

Conclusion: The AI-driven adaptive system was assessed for risk towards development of a quality assurance program. This gives the international community a platform from which to safely implement the system in their clinic.

Funding Support, Disclosures, and Conflict of Interest: This work has been supported by a collaborative research agreement from Varian Medical Systems to Royal North Shore Hospital

Keywords

Quality Assurance, Risk

Taxonomy

TH- External Beam- Photons: adaptive therapy

Contact Email