KEY POINTS
- The authors developed an automated framework that starts from delivered-treatment records and reconstructs linked radiotherapy plans, doses, structure sets, planning images, registrations, and diagnostic imaging across heterogeneous clinical systems.
- The framework was deployed across four institutions using six combinations of treatment-planning and record-and-verify systems, including ARIA/Eclipse, MOSAIQ, RayStation, MIM, Pinnacle, and Brainlab-related data sources.
- In the completed two-clinic implementation covering 11 years, the system processed 6,164 patients and identified 13,904 referenced plans, successfully retrieving 13,871 plans for an overall success rate of 99.76%.
- Mean processing and transfer time was approximately 18 minutes per patient. Custom modules converted proprietary or non-queryable data into DICOM-compatible objects, while automated validation, logging, and retry mechanisms supported recovery from interruptions and missing links.
- Generalizability remains uncertain because complete collection was reported from only two clinics, both using ARIA as the record-and-verify system. The framework also does not yet integrate outcomes, toxicity, staging, or other non-DICOM clinical variables.
CLINICAL TAKEAWAY
This framework demonstrates that large retrospective radiotherapy datasets can be assembled with high technical completeness despite fragmented planning and delivery systems. Its immediate value is mainly for physicists, informaticians, and research infrastructure teams; clinical impact will depend on broader multi-institutional validation and integration with outcomes and toxicity data.