UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Pre-clinical assessments of head and neck organs at risk atlas-based auto-segmentation with advanced methodology for parotid glands Khawandanh, Eman Hesham


Modern technology allows radiation therapy dose distributions to conform closely to targets, providing better sparing of adjacent anatomical structures (organs-at–risk, OARs) in the treatment of cancer. This has the potential to reduce radiation side effects. To take advantage of this technology, accurate delineation of both targets and organs at risk is essential. Routinely, organs at risk are delineated (segmented) on anatomical images using a manual process which can be both time consuming and error prone. Automated segmentation methodology is therefore an active area of research. The objectives of this thesis were: 1) to assess the variability in manual segmentation of head and neck OARs and provide benchmark data for comparisons with auto-segmentation; 2) to compile and categorize a set of image data sets with expertly validated segmentations of head and neck OARs, forming an in-house constructed atlas library for use with an automated segmentation software tool; 3) to evaluate the performance of an atlas based auto-segmentation tool (MIM Maestro™) using the in-house constructed atlas library; and 4) to improve the auto-segmentation performance by studying the impact of the number and quality of atlas library cases and different user-defined settings. Results of these studies indicate that the time required to segment a complete OAR set can be reduced to three minutes using the atlas-based auto-segmentation approach, versus 30 minutes for manual segmentation. With the exception of salivary glands, the auto-segmentation performance was clinically acceptable for all organs. Atlas-based auto-segmentation performance for salivary glands was improved by increasing the quantity and quality of atlas cases in the library. The results provide novel insight into the behaviour of auto-segmentation algorithms. This performance evaluation of OAR segmentation in head and neck radiotherapy provides the basis for clinical implementation of the MIM Maestro ™ auto-segmentation software at the British Columbia Cancer Agency, Vancouver Centre.

Item Citations and Data


Attribution-NonCommercial-NoDerivatives 4.0 International