Click here to

Session: Treatment Planning: Planning Automation and Assessment [Return to Session]

Landmark-Based Field Aperture Generation for Automatic Whole-Brain Radiation Planning

Y Xiao1*, D Rhee1, T Netherton1, L Zhang1, A Aggarwal2, C Trauernicht3, H Simonds3, L Court1, C Cardenas1, (1) MD Anderson Cancer Center, Houston, TX, (2) King's College London, London, UK,(3) Stellenbosch University, Cape Town, South Africa


SU-F-TRACK 5-3 (Sunday, 7/25/2021) 4:30 PM - 5:30 PM [Eastern Time (GMT-4)]

Purpose: To develop and compare two landmark-based approaches for automatic whole-brain treatment planning that can be configured to suit local clinical practices.

Methods: Field apertures for both approaches used landmarks based on organ contours seen in beams-eye-view digitally reconstructed radiographs (DRRs), with the relationship between the landmarks and field borders configurable to suit local clinical practice. For Approach 1, a previously trained deep learning model was used to create contours on CT scans, and these contours were projected onto DRRs. For Approach 2, organ contours were directly segmented on DRRs using a deep learning model trained/validated/tested using 724/182/224 patients. Field apertures were then created using the DRR contours from both approaches and compared with manually outlined field apertures using Hausdorff Distance (HD) and Mean Surface Distance (MSD). Additionally, quality assurance for the contours on DRRs from Approach 1 and 2 were compared using Dice-Similarity-Coefficient (DSC). In total, eight landmark-based rules consisting of anterior, carinal, posterior, caudal, and anterior-caudal bounds were applied, and five setting options included selecting vertebral bodies, anterior-caudal bound shapes, and extent of brain expansion and skin flash were provided for creating field aperture variations.

Results: The average DSC of organ contours segmented on DRRs regarding brain, eyes, lens, C1, and C2 were 0.97±0.01, 0.88±0.05, 0.55±0.18, 0.89±0.04, and 0.87±0.07, respectively. The performance of the generated field apertures from a single configuration in terms of the average HD and MSD for Approach 1 resulted in 16.43±6.95 and 11.63±7.26 mm, respectively, and 16.93±6.87 and 12.18±6.92 mm for Approach 2.

Conclusion: We have developed and tested two landmark-based methods to define whole-brain field apertures. Both gave similar results compared to the manually outlined field apertures. The flexible configuration for creating field aperture variations also provided flexibility in adapting this traditional treatment approach to local clinical practices.

Funding Support, Disclosures, and Conflict of Interest: This work is fully supported by Wellcome Grant UNS82174.



    Not Applicable / None Entered.


    Not Applicable / None Entered.

    Contact Email