Segmentation of Micro-CT images: an understudied problem

Aims Segmentation of biomedical images is a very wide subject and it is a mandatory step for the analysis and quantification of biomedical features. Any expert of imaging knows that does not exist a single segmentation technique able to segment all kinds of images, therefore the choice of the technique, and its settings, are related to the problem at hand. In high resolution micro-CT images of trabecular bone a fixed threshold was found to be a reasonable solution for segmentation of trabecular structure [1-4]. In order to choose the right threshold, however, a comparison with the golden standard is required, usually synchrotron facilities, histological studies or Archimedes’ principle [3, 5, 6]. By comparing micro-CT images to higher resolution images it is possible to find the right threshold, but only for the specific set-up. Moreover, limits of the fixed threshold for the segmentation of micro-CT images were already pointed out for specimens with bone volume fraction (BV/TV) below 15% [7]. Generally, when the dimensions of the analyzed objects become comparable to the image resolution, the partial volume effect starts to play an important role in image segmentation, and the error in quantification of biomedical features rises. In order to reduce this kind of errors, many techniques able to adapt to the analyzed image were introduced. In the Otsu method a fixed threshold is automatically identified for every new image depending on its histogram [8], but more sophisticated active segmentation techniques were also developed along the years [9-11]. The purpose of the active segmentation techniques is the definition of contours around the object(s) to be segmented, in which the curve deforms to minimize the contour energy that includes the components of internal energy from the contour and external energy from the object. The aim of the present study is to compare some of the most used or promising techniques on a dataset of micro-CT images of trabecular bone in order to highlight the advantages and disadvantages of each.