Author: Shatha Al Afifi, Hana Baroudi, Leonard Che Fru, Laurence Edward Court, Suzanne B. Evans, Kent A. Gifford, Adam D. Melancon, Melissa P. Mitchell, Issa Mohamad, Patricia Murina, Manickam Muruganandham, Tucker J. Netherton, Callistus M. Nguyen, Joshua S. Niedzielski, Deborah L. Schofield, Simona Shaitelman, Willie Shaw, Sanjay S. Shete, Adam Shulman, Brendon Smith, Sheeba Thengumpallil, Carlos Daniel Venencia, Conny Vrieling 👨🔬
Affiliation: University of Cape Town, MD Anderson Cancer Center, The University of Texas MD Anderson Cancer Center, University of the Free State, UT MD Anderson Cancer Center, King Hussein Cancer Center, Instituto Zunino - Fundacion Marie Curie, Department of Radiation Physics, The University of Texas MD Anderson Cancer Center, Dra., Hirslanden Clinique des Grangettes, Department of Radiation Oncology, The University of Texas MD Anderson Cancer Center, Department of Therapeutic Radiology, Yale University School of Medicine, Yale University, The University of Texas, MD Anderson, Houston TX 🌍
Purpose:
Automated contouring and planning tools are usually trained on single-institution datasets, seldom tested across diverse patient populations. This introduces a risk of population bias, restricting their generalizability for global application. This study aims to develop an end-to-end automated planning tool for breast radiotherapy and assess its clinical performance and limitations in multiple patient populations.
Methods:
An automated contouring and planning tool for breast radiotherapy was developed, supporting nine treatment approaches, including tangents, supraclavicular nodal fields with/without axillary nodes, and internal mammary nodes treatments using partially wide tangents or matched electron fields. The tool automatically contours the targets and normal tissues and optimizes gantry angles and field shapes to minimize dose to critical structures.
Validation was conducted using a dataset of 475 patients from 8 institutions across 6 countries (Argentina, Iran, Jordan, USA, South Africa and Switzerland), representing diverse patient populations, imaging and positioning protocols. Evaluation involved physics review of dose distributions, dosimetric assessment of 4,275 plans following established guidelines, and fault tree analysis to identify and mitigate risks for safe implementation.
Results:
Dosimetric objectives and constraints were met for breast (99%), supraclavicular (98%), axillary (98%), internal mammary nodes (91%), heart(97%), ipsilateral lung(88%), and contralateral breast(96%). Contour and plan review confirmed 84% of auto-plans were acceptable as is, and 16% requiring edits. Edit rates ranged from 12–18% for four institutions (p>0.05) to 26% and 31% for two institutions (p<0.05), with one involving intentionally challenging cases. Failure modes were identified, primarily linked to previously unseen patient anatomy (e.g., large skin folds) and positioning (e.g., rotation). Mitigation strategies, including automated detection of high-risk features like implants, reduced these risks.
Conclusion:
The automated planning tool demonstrated high clinical performance and acceptability across diverse populations. By addressing population-specific challenges, the tool offers significant potential for global adoption, enhancing access to high-quality breast radiotherapy.