Click here to

Session: Multi-Disciplinary General ePoster Viewing [Return to Session]

Automated Delineation of CTVs for Nasopharyngeal Cancer

C Sjogreen1*, T Netherton1, C Cardenas2, A Lee1, B Beadle3, D Rhee1, S Gay1, C Nguyen1, R Mumme1, J Duryea1, L Court1, (1) UT MD Anderson Cancer Center, Houston, TX, (2) The University of Alabama at Birmingham, Birmingham, AL,(3) Stanford University, Stanford, CA,


PO-GePV-M-335 (Sunday, 7/10/2022)   [Eastern Time (GMT-4)]

ePoster Forums

Purpose: Contouring guidelines for the delineation of CTVs for nasopharyngeal cancer radiotherapy can be complex and depend on the location and disease extent. This is particularly true for low-risk CTVs, which are defined based on coverage of various anatomic structures. Furthermore, different protocols have different guidelines. We are developing an auto-contouring solution, which follows the NRG HN001 protocol but can be adjusted to meet other protocol requirements.

Methods: We first used 2-channel 3D U-Net and nnU-Net architectures to auto-contour 22 normal structures (Clivus, Nasopharynx, Pterygoid Fossae, Maxillary Sinus, Sphenoid Sinus, … etc.) that are used in the NRG HN001 protocol to define the CTVs. We then extracted morphological and geometric landmarks from these structures and created a polygon set describing the high-risk CTV (CTV1), intermediate-risk CTV (CTV2), and low-risk CTV (CTV3, CTV4). While CTV1 and CTV2 are mostly expansions from the GTVs, CTV3 and CTV4 are a combination of variable expansion from the GTV, and partial or full inclusion of various anatomical structures, depending on the location and extent of the disease. The auto-contouring styles were then processed to correctly mimic physician interpretation of the protocol rules. Finally, contours were assessed quantitatively with Dice-Similarity-Coefficient (DSC), Mean-Surface-Distance (MSD), and subjectively (physician review).

Results: DSC and MSD comparing auto-contours and clinical contours were 0.82±0.02 and 2.5±0.3mm between CTV3, CTV4 and the clinical contours (ground-truth) for 9 nasopharyngeal cancer patients with varying stages; 100% of the cases had DSC > 0.78 and MSD ⩽ 3.3 mm. The clinical review was scored on a 5-point scale system; 55% were scored with≥ 4 use-as-is, and the other 45% were scored 3 use-as-is after some minor edits.

Conclusion: We have successfully developed an auto-contouring tool that can follow specific guidelines for the delineation of nasopharyngeal clinical target volumes, with minimal physician interaction.

Funding Support, Disclosures, and Conflict of Interest: Varian Medical Systems


Not Applicable / None Entered.


IM/TH- image Segmentation: CT

Contact Email