Active segmentation of micro-CT trabecular bone images

The segmentation of biomedical images is a required stage for the quantification of biomedical features. In micro-CT images the segmentation of the scanned datasets is required in order to compute the morphometrical characteristics of the tissue structure. In literature the use of a fixed threshold, selected by means of different evaluation techniques, is usually suggested for this kind of images, but limits and critical scenarios were already highlighted. Different active segmentation methods were already suggested in literature for different biomedical purpose. Aim of the present study is the application of the Chan-Vese active segmentation technique to micro-CT images, and its comparison to the fixed threshold technique. Three micro-CT trabecular datasets were selected. The two segmentation techniques (Chan-Vese and fixed threshold) were applied and morphometrical parameters were computed for both of them. For every slice of every dataset, two dimentional (2D) bone density and trabecular thickness were computed. The resulted were compared between the two segmentation techniques by means of a paired t-test. Moreover a visual comparison was performed. The two segmentation techniques showed statistically significant differences in the measured parameters. According to this preliminary study, the Chan-Vese segmentation technique produced better results than the global thresholding in term of visual inspection, in scenarios with critical low bone density. Negligible differences were observed in the high density dataset.