MENU

Click here to

×

Are you sure ?

Yes, do it No, cancel

A Comparison of Interpolation Methods in Head and Neck Radiotherapy Contouring

J Spencer1*, K Chen2 , D Boukerroui3 , M Gooding4 , J Fenwick5 , (1) ,,,(2) University of Liverpool, Liverpool, Merseyside, (3) ,,,(4) Mirada Medical Ltd., Oxford, Oxfordshire, (5) Clatterbridge Cancer Centre, Birkenhead, Merseyside

Presentations

(Sunday, 7/29/2018) 4:30 PM - 5:00 PM

Room: Exhibit Hall | Forum 9

Purpose: Manual contouring of organs at risk (OAR) in RT planning is often time consuming and laborious. Interpolation methods are a useful generic tool, and can supplement manual approaches in a clinical setting. We compare two interpolation
methods: data-driven, which was developed in-house, and linear. Neither use prior knowledge of the target structure, and we investigate their dependence on the initial user input.

Methods: We consider certain OAR in the head and neck as an example, with complete outlines for seven structures from five data sets. We assume knowledge of multiple axial contours, as well as two orthogonal contours (from the sagittal and coronal views). These are distributed evenly throughout the structure, although in practice they would be provided manually.The data-driven interpolation (DDI) algorithm automatically approximates the surface of the structure using information from the user input, and refines the contour with a fast regularisation step. We compare this to results obtained by linear interpolation (LI) of the known axial contours. The quantitative measure used to evaluate performance in terms of accuracy is the Dice Similarity Coefficient (DSC).

Results: There is significant improvement for the cord and the hyoid, where DDI consistently outperforms LI. For structures with minimal contrast (brainstem and submandibular glands) on the known contours, the performance is similar for each method. However, there is some minor improvement for DDI as slice spacing increases for the parotid glands.

Conclusion: By utilising user-provided data, interpolation methods can achieve improved results for some structures in the head and neck. DDI incorporates manually acquired contours from multiple views and potentially offers significant advantages in terms of interactivity, and flexibility in terms of the OAR considered compared to LI. In a clinical setting this could be useful with respect to the input requirements on the user.

Funding Support, Disclosures, and Conflict of Interest: This work was supported by EPSRC grant number EP/N014499/1. D. Boukerroui and M. Gooding are employees of Mirada Medical Ltd.

Keywords

Contour Extraction, Segmentation, CT

Taxonomy

IM/TH- image segmentation: CT

Contact Email