Efficient selection of the most similar image in a database for critical structures segmentation

Abstract

Radiotherapy planning needs accurate delineations of the critical structures. Atlas-based segmentation has been shown to be very efficient to delineate brain structures. However, the construction of an atlas from a dataset of images, particularly for the head and neck region, is very difficult due to the high variability of the images and can generate over-segmented structures in the atlas. To overcome this drawback, we present in this paper an alternative method to select as a template the image in a database that is the most similar to the patient to be segmented. This similarity is based on a distance between transformations. A major contribution is that we do not compute every patient-to-sample registration to find the most similar template, but only the registration of the patient towards an average image. This method has therefore the advantage of being computationally very efficient. We present a qualitative and quantitative comparison between the proposed method and a classical atlas-based segmentation method. This evaluation is performed on a subset of 45 patients using a Leave-One-Out method and shows a great improvement of the specificity of the results.