Radiation Planning Assistant (RPA)

The Radiation Planning Assistant (RPA) is an assistive software tool designed to streamline radiotherapy treatment planning for specific cancers by employing deep learning algorithms for automated contouring and optimized plan generation using CT images. Its methodology integrates with existing clinical workflows by requiring user review and finalization in their primary Treatment Planning System, with performance data indicating comparable dosimetric outcomes to conventional planning, often with a conservative approach to organ sparing.

The Radiation Planning Assistant (RPA) is a web-based software designed to aid in radiotherapy treatment planning for head and neck, cervix, breast, and brain metastasis cancers. It utilizes CT images for external beam photon therapy planning, offering automated contouring, automated planning with dose optimization, and quality control checks. The system's science relies on deep learning algorithms for contouring, trained on extensive patient CT datasets, and for VMAT planning, it employs pre-set optimization objectives leveraging some Eclipse system features. Critically, the RPA serves as an assistive tool; all its outputs require import, review, editing, and final dose calculation within a user's primary Treatment Planning System (TPS). Non-clinical software validation confirmed the RPA's adherence to specifications. Multicenter clinical studies evaluated its performance across various anatomical sites (cervix, chest wall, head and neck, whole brain) and CT scanner types. Results generally demonstrated high concordance with clinical plans regarding dosimetric metrics for safety and effectiveness, often with the RPA exhibiting a more conservative approach to organ-at-risk sparing (e.g., bowel bag in cervix, cochlea in head/neck, lenses in whole brain). Across all sites, the RPA showed proficient body contouring (Surface DSC > 0.8) and accurate isocenter identification (<3mm deviation), underscoring its utility in the initial stages of radiation treatment planning.

AI Regulatory Concerns

Powered by

Jev reads the full FDA letter and returns calibrated concern probabilities per regulatory category. These are AI estimates, not regulatory advice.

• Q1. How does the performance of the RPA's deep learning contouring algorithms vary across different CT image qualities or scanner protocols not included in the training/testing datasets?; Given that RPA-generated VMAT plans have pre-set optimization objectives and are not easily editable within the RPA, what is the typical extent of modification required when these plans are imported into a user's primary TPS, and how does this impact overall planning efficiency?; The document notes some differences in dosimetric outcomes (e.g., cochlea in head/neck, bowel bag/kidney/bladder in cervix VMAT) compared to clinical plans, often citing RPA's 'conservative' nature. What specific aspects of the RPA's algorithms lead to this conservatism, and could it potentially compromise optimal target coverage or lead to unnecessary organ-at-risk sparing in certain clinical scenarios?; For the 'simple plans' where the plan is generated before contour review, what mechanisms are in place to flag potentially suboptimal contours that could significantly impact the initial plan, especially if a user might overlook subtle contouring errors before proceeding to their TPS?; The RPA was trained on data from MD Anderson. How robust is the system's performance when applied to patient populations with significantly different anatomical variations, disease presentations, or imaging protocols encountered in diverse global oncology practices?