Click here to

Session: AI/ML Autoplanning, Autosegmentation, and Image Processing II [Return to Session]

Automated Pipeline for Prostate Auto-Contouring and VMAT Planning

M El Basha1,2*, Q Nguyen2, D Fuentes1,2, J Pollard-Larkin1,2, F Poenisch2, Z Yu1,2, S Frank2, C Cardenas3, C Nguyen2, A Olanrewaju2, C Tang2, S Shah2, A Aggarwal4, L Court1,2, (1) University of Texas MD Anderson Cancer Center UTHealth Graduate School of Biomedical Sciences, Houston, TX, (2) MD Anderson Cancer Center, Houston, TX, (3) The University of Alabama at Birmingham, Birmingham, AL, (4) Guy's and St. Thomas' NHS Foundation Trust, London, UK


SU-F-BRB-2 (Sunday, 7/10/2022) 2:00 PM - 3:00 PM [Eastern Time (GMT-4)]

Ballroom B

Purpose: To develop a sole automated contouring and radiotherapy planning pipeline for intact and postoperative prostate cancer patients.

Methods: First, a fully-comprehensive auto-contouring model was developed. An ensemble of nnU-Net models was developed and validated for individual segmentation on 832 retrospective clinical CT volumes (8:1:1 train:validation:test split) of 10 target and OAR structures: prostate, seminal vesicles (SV), prostate fossa, pelvic lymph nodes, rectum (with/without rectal balloon), bladder, femoral-heads, penile-bulb, sigmoid, and SpaceOAR (with/without). The model automatically determines if a patient is post-operative and generates the corresponding structures that are corrected for overlap between targets and OARs. A RapidPlan (Eclipse) model was developed/evaluated on a separate 95/20 clinical intact and prostatectomy cases with/without SpaceOAR and rectal balloons. Isodose planning structures (105% and 104%) are progressively added in up to two additional optimization iterations to identify and remove hotspots in the PTV. The 20 model evaluation cases were reviewed by two radiation oncologists. Finally, the end-to-end pipeline of automatic contouring followed by VMAT planning was tested with 15 patients.

Results: Segmentations of planning structures were generated with the following average Dice/mean surface distances: prostate 0.85±0.06/1.68±0.83mm, SV 0.68±0.18/2.2±3.1mm, prostate fossa 0.80±0.10/3.1±0.6mm, rectum 0.88±0.05/1.53±0.9mm, bladder 0.96±0.04/0.7±0.8mm, penile-bulb 0.75±0.11/1.62±0.9mm, femoral-heads 0.94±0.04/1.6±1.0mm, sigmoid 0.52±0.21/9.3±8.0mm and lymph nodes 0.80±0.05/2.6±0.6mm. All VMAT plans on clinical patient contour sets were found to be clinically acceptable (90% use-as-is, 10% use after minor edits). End-to-end testing (auto-contouring, auto-planning) passed all OAR constraints for 12/15 patients. The remaining 3 did not pass constraints due to large deformation of anatomy from insertion of rectal balloons. Two-thirds of the plans passed PTV target V78Gy≥95% constraint with the remaining one-third missing the constraint by 0.67±0.33Gy.

Conclusion: We have automated the entire prostate contouring and planning process and demonstrated the ability to create high-quality contours and plans from a single, minimal interaction pipeline.

Funding Support, Disclosures, and Conflict of Interest: This material is based upon work supported by the National Science Foundation Graduate Research Fellowship Program under Grant No. 2043424. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the National Science Foundation.


Prostate Therapy, CT


TH- External Beam- Photons: Treatment planning using machine learning/Knowledge Based Planning/automation

Contact Email