Click here to

Session: Automated Software and Clinical Tools [Return to Session]

Radiation Planning Assistant: International Physician Review of Automatically Generated Contours and Radiotherapy Plans

L Court1*, L Zhang1, A Olanrewaju1, R Mumme1, R Douglas1, D Rhee1, Y Xiao1, C Cardenas2, T Netherton1, C Nguyen1, A Aggarwal3, A Jhingran1, C Trauernicht4, H Simonds4, B Beadle5, (1) UT MD Anderson Cancer Center, Houston, TX, (2) The University of Alabama at Birmingham, Birmingham, AL, (3) Guy's and St. Thomas' NHS Foundation Trust, London, UK, (4) Stellenbosch University, Cape Town,ZA, (5) Stanford University, Stanford, CA


MO-C930-IePD-F4-1 (Monday, 7/11/2022) 9:30 AM - 10:00 AM [Eastern Time (GMT-4)]

Exhibit Hall | Forum 4

Purpose: Globally, there is insufficient access to high-quality radiotherapy. Here we describe efforts to address this inequity by developing automated, web-based tools for cancers of the cervix, breast, head/neck, and brain.

Methods: The Radiation Planning Assistant (RPA) is being developed to offer automated contouring/planning to clinics across the world. The user uploads a patient’s CT image and completes a web-form detailing disease extent and radiation prescription. Deep learning and other automation algorithms then complete the contouring and planning tasks for simple (1-step) and complex (2-step) processes. For 1-step processes (3DCRT), contouring and planning is completed without user intervention following the initial input of the CT and prescription. For 2-step processes (VMAT), users can edit auto-contours prior to automated plan optimization. Clinical acceptability of the RPA was determined for 75 patients for each task by at least 3 radiation oncologists, each from a different institution. Overall, 31 radiation oncologists (from 15 institutions and 5 countries [India South Africa, UK, USA, Zambia]) provided feedback.

Results: Use-As-Is or Use-After-Minor-Edits scores for 1-step planning were 81%/12% (bony landmark 4-field box for cervical cancer), 79%/17% (soft-tissue 4-field box for cervical cancer), 44%/47% (tangents/SCLV for post-mastectomy breast) and 76%/24% (whole brain). Automated contouring, the first step in the 2-step planning process, was scored as Use-As-Is or Use-After-Minor-Edits for 89%/8% and 92%/7% for normal tissues in the head/neck and cervix regions, respectively. CTVs for head/neck and cervical cancers were scored 40%/53% and 83%/9%, respectively, with larger inter-observer variabilities.Automated VMAT planning, the second step in the 2-step planning process, was scored 87%/9% and 99%/1% for head/neck and cervical cancer plans, respectively. For both HNC CTVs and VMAT plans, one physician did not consider the contouring/planning style acceptable, indicating that some customization may be necessary.

Conclusion: Physician reviews indicate a high level of clinical acceptance of the RPA

Funding Support, Disclosures, and Conflict of Interest: This project was partially supported by the NCI, CPRIT, and the Wellcome Trust.


Not Applicable / None Entered.


TH- External Beam- Photons: General (most aspects)

Contact Email