ESTRO meets Asia 2024 - Abstract Book

S394

RTT – Treatment planning, OAR and target definitions

ESTRO meets Asia 2024

[1] “A Dosimetric Comparison Using Avoidance Sectors vs Avoidance Structures for Hip Prostheses in Cancer Patients Receiving Pelvic Irradiation”. Grace E.

[2] “Volumetric modulated arc therapy for synchronous bilateral whole breast irradiation – A case study”. Jan Seppälä

[3] “The choice of statistical methods for comparisons of dosimetric data in radiotherapy”. Abdulhamid Chaikh et al.

[4] “Evaluation of VMAT Planning Strategies for Prostate Patients with Bilateral Hip Prosthesis”. To D. et al.

232

Proffered Paper

Network-wide implementation and uptake of an AI contouring platform: 33 departments in 20 days

Aidan Leong 1,2 , Huong Nguyen 3 , Ben Archibald-Heeren 3 , Yunfei Hu 4

1 Radiation Therapy, Bowen Icon Cancer Centre, Wellington, New Zealand. 2 Department of Radiation Therapy, University of Otago, Wellington, New Zealand. 3 Clinical Care, Icon Group, Brisbane, Australia. 4 Medical Physics, Icon Cancer Centre Gosford, Gosford, Australia

Purpose/Objective:

AI contouring platforms using deep learning for autosegmention are increasingly prevalent in the clinical environment. While there is an established body of literature regarding the performance of such platforms, there is comparatively little research outlining the implementation and uptake of these applications in a real-world setting. This study outlines our international network’s early experience regarding the implementation, uptake and staff feedback of an AI contouring platform into standard practice.

Material/Methods:

A commercially available AI contouring platform was identified for implementation across the Australasian sector of our network, comprising 33 individual radiation oncology centres. The application was first installed at a single site to facilitate pre-clinical validation and preliminary workflow development. A radiation therapist-focused education program was developed including elearning, practical training, and peer-review components which could be flexibly delivered both remotely and in-person. A framework of partnered centres was established to focus initial training and consolidation of experience at early-adopter sites in each geographic region, who would then cascade training to subsequent centres alongside the accumulation of further learnings and contribution to workflow refinement. Clinical release of the platform at individual centres was sequenced across a defined implementation timeframe in coordination with the completion of staff training requirements. AI contouring use was tracked via tasks within the oncology information system, with each task prompting user feedback on perceived accuracy and time-saving benefit of the platform. Further data on utilisation was extracted from the platform’s API. Training completion was reported from the network’s learning management system.

Results: