Online Poster Portal

Background

Radiation therapy generates three-dimensional dose distributions that provide critical spatial context for interpreting post-treatment imaging. This information can help distinguish tumor recurrence from treatment-related toxicity, yet radiation dose data often remain siloed within radiation oncology treatment planning systems rather than being available in routine diagnostic imaging workflows. Prior institutional work demonstrated the value of sharing radiation dose distribution maps in the electronic medical record, but the existing workflow required manual preparation, delivery confirmation, and selective export, limiting scalability.

Objective

To develop and evaluate Radiation Dose Routing for Oncology into PACS (Rad-DROP), an open-source, vendor-agnostic, automated framework that identifies completed radiation therapy plans, generates clinically interpretable dose visualizations, and routes them into PACS for use by radiologists and multidisciplinary oncology teams. The broader goal is to make prior radiation history available at the point of imaging interpretation and tumor board review without requiring manual retrieval from separate oncology systems.

Methods

Rad-DROP was designed as an end-to-end pipeline spanning clinical data ingestion, treatment validation, automated anatomic classification, dose rendering, quality control, and PACS integration. Radiation oncology data were retrieved from the ARIA oncology information system using standard DICOM Query/Retrieve operations, including C-FIND and C-MOVE, and organized into structured patient-, study-, and series-level directories. For each case, the pipeline ingested the planning CT, RTPLAN, RTDOSE, and RTSTRUCT objects. Treatment completion was determined by integrating planned fractionation, treatment delivery records, beam-level meterset information, delivered dose calculations, treatment dates, and temporal criteria for clinically concluded courses.

Automated anatomic segmentation was performed on the planning CT using TotalSegmentator. The treated site was inferred through overlap analysis between segmented anatomy, clinical target contours when available, and multiple relative isodose masks. These site assignments were used to select clinically appropriate window and level settings and to standardize visualization across disease sites. Dose visualizations were generated using 3D Slicer and SlicerRT with custom Python automation. Because most diagnostic PACS systems do not fully support native DICOM-RT objects, Rad-DROP converted rendered dose overlays into PACS-compatible Secondary Capture DICOM series. Each output included standardized isodose lines at 100%, 90%, 80%, 75%, and 50% of delivered dose, along with treatment metadata including treatment dates, delivered and prescribed dose, fractionation, and display settings.

A quality-control framework was incorporated before PACS transfer. Initial manual review was performed to identify systematic rendering errors, followed by automated detection of missing CT data, absent isodose information, and anomalous outputs. Validated Secondary Capture DICOM objects were then transferred into PACS using an authenticated DICOM node and pydicom-based automation. The system was deployed on institutional compute infrastructure with shared storage, containerized processing, asynchronous task orchestration, CPU/GPU resource allocation, and automated retry mechanisms to support retrospective batch processing and prospective nightly operation.

Results

Rad-DROP has enabled institutional-scale integration of radiation dose information into PACS. More than 15,000 radiation dose maps were routed into the institutional PACS over a period of weeks, and a prospective nightly service was established to identify completed treatment courses and automatically route corresponding dose visualizations for ongoing clinical access. Representative outputs demonstrated consistent visualization across multiple treatment sites, including brain stereotactic radiosurgery, head and neck radiotherapy, spine stereotactic body radiation therapy, breast radiotherapy, lung stereotactic body radiation therapy, and prostate stereotactic body radiation therapy.

The broader validation plan leverages an institutional archive of approximately 28,000 patients and more than 140,000 radiation therapy plans. Planned technical endpoints include treated-site classification performance, PACS transfer success rate, end-to-end pipeline success rate, runtime distribution, quality-assurance flag frequency, and categorized failure modes. A prospective pilot will evaluate workflow impact during multidisciplinary review, including time required to retrieve dose context and clinician confidence in post-treatment imaging interpretation and decision-making using Likert-scale measures.

Conclusions

Rad-DROP transforms radiation dose from a treatment-planning artifact confined to radiation oncology software into an accessible imaging resource within routine PACS workflows. By automating radiation dose visualization and PACS delivery, this framework reduces manual barriers to treatment-history review, supports more informed interpretation of post-treatment imaging, and improves access to radiation context for multidisciplinary oncology care. As an open-source and vendor-agnostic system, Rad-DROP provides scalable infrastructure for future applications including cumulative dose review, re-irradiation assessment, automated registration of prior dose onto follow-up imaging, and data-driven clinical decision support.