Click here to

Session: Deep Learning for Treatment Planning [Return to Session]

Automated Whole-Brain Radiotherapy Treatment Planning Via Deep Learning Auto-Contouring and Customizable Landmark-Based Field Aperture Design

Y Xiao1*, D Rhee1, T Netherton1, L Zhang1, C Nguyen1, R Douglas1, R Mumme1, A Aggarwal2, S Skett2, T Patel2, C Trauernicht3, H Simonds4, C Cardenas5, L Court1, (1) The University Of Texas MD Anderson Cancer Center, Houston, TX, (2) Guy's and St. Thomas' NHS Foundation Trust, London, UK, (3) Stellenbosch University, Cape Town, ZA, (4) Stellenbosch University, Stellenbosch, ZA, (5) The University of Alabama at Birmingham, Birmingham, AL


TU-I345-IePD-F2-1 (Tuesday, 7/12/2022) 3:45 PM - 4:15 PM [Eastern Time (GMT-4)]

Exhibit Hall | Forum 2

Purpose: To develop and evaluate an automated radiotherapy treatment planning pipeline with design flexibility and quality assurance (QA) for whole-brain patients.

Methods: The pipeline consists of the following steps: (1) automatically contour normal structures on CT scans and digitally reconstructed radiographs (DRR) using deep learning techniques, (2) locate the landmark structures on beam’s-eye-view, (3) generate field apertures based on different landmark rules. Two treatment specifications based on MLC blocking and angled collimator (AC) were developed to suit different hospital requirements. Eight landmark rules based on MLC were designed to address different clinical purposes and physician preferences. Field apertures based on DRR contours were generated for QA purposes to check the consistency/accuracy. The performance of the generated field shapes and dose distributions were evaluated by the original clinical plans. The clinical acceptability of the plans was assessed by five radiation oncologists from four hospitals.

Results: Hausdorff Distance and Mean Surface Distance of the MLC apertures and the original clinical apertures were compared (182 patients), resulting in 16±7mm and 12±7mm for the MLC approach, and 17±7mm and 12±7mm for the QA approach, and the differences between the MLC and QA approaches are 1±2mm and 1±3mm, respectively. Clinical review of 75 patients for MLC, QA, and AC approaches achieved 100%, 100%, and 93% acceptance rate for use as is or after minor edits for field shape design, and 99%, 96%, 75%, respectively for dose distribution. All three approaches achieved a 100% acceptance rate for brain coverage (V95%). The acceptance rate for meeting lens dosimetric recommendations was 75%, 80%, and 91% for MLC, QA, and AC approaches, compared with 50% for the clinical plans.

Conclusion: This study provided a novel pipeline to automatically generate whole-brain radiotherapy treatment plans. Both quantitative and qualitative evaluations demonstrated that our plans are comparable with the original clinical plans.

Funding Support, Disclosures, and Conflict of Interest: Wellcome Trust Funding


Not Applicable / None Entered.


Not Applicable / None Entered.

Contact Email