Click here to


Are you sure ?

Yes, do it No, cancel

Architecture, Implementation, and Performance of An Automatic and High-Throughput Treatment Planning Service: The Radiation Planning Assistant

L Zhang*, C Cardenas, D Rhee, C Nguyen, A Olanrewaju, L Court, MD Anderson Cancer Center, Houston, TX


(Sunday, 7/12/2020)   [Eastern Time (GMT-4)]

Room: AAPM ePoster Library

Purpose: To build a web-based automated contouring and planning service for clinics in resource-limited countries, focusing on high-quality, high-throughput and high-availability.

Methods: CT images are uploaded to the Radiation Planning Assistant (RPA) website, and then automatically distributed to different job schedulers to generate contours, treatment plans, and automated QA checks.

The deep learning scheduler prepares contouring jobs, performs load balancing, and submits jobs to GPU-clusters. Completed jobs are passed to the post-processing scheduler, which merges all segmentation result and generates a PDF report and DICOM-RTSTRUCT file for the user to download.

For treatment planning, pre-processing and pre-plan schedulers compute marked isocenter position and beam geometry, generates basic plan and sends these to Eclipse. The optimization scheduler then distributes jobs to script engines for plan optimization and dose calculation. Finally, the report scheduler generates plan PDF report, DICOM-RTSTRUCT and DICOM-RTPLAN files.

For plan QA, the QA service imports DICOM-RTSTRUCT or DICOM-RTPLAN, generating the following QA jobs: plan quality reporting, dose verification, isocenter QA, contouring QA, and beam aperture QA.

System performance was tested using unloaded (single patient) and heavily loaded (many patients at once) scenarios.

Results: On average, normal tissue (n=21) and CTV contouring a single head-and-neck patient takes 9 minutes. Treatment planning takes 24, 15 and 9 minutes and automated plan QA takes 8, 5 and 4 minutes for head-and-neck (VMAT), chestwall (tangents/sclv, field-in-field) and cervix (4-field) plans, respectively. Under loaded conditions, the system hourly throughput for head-and-neck contouring is 18 patients/hour. For planning, the throughput is 6, 9, and 16 for head-and-neck, chestwall and cervix patients/hour. For QA, the throughput is 18, 22, and 28 patients/hour.

Conclusions: We have implemented a high-throughput automated planning service. The system’s performance is scalable to meet demands of a service designed to offer high-quality radiotherapy contouring and planning to clinics with limited resources.

Funding Support, Disclosures, and Conflict of Interest: This work was partially funded by NCI (UH3CA202665) and Varian Medical Systems.


Not Applicable / None Entered.


Not Applicable / None Entered.

Contact Email