Evaluation performance of local adaptive binarization algorithms for trabecular bone on simulated µCT

In this work, five local adaptive thresholding methods (named Bernsen, Niblack, Sauvola, Wellner and White) were evaluated to segment three-dimensional X-ray microtomographies (μCT). The basic objective was to use a computer algorithm to automatically quantify bone morphology. The performance of these methods was quantitatively evaluated by using 128 images (pixel size of 0.015 mm) obtained from a three-dimensional numerical phantom, which mimics the porosity of trabecular bone. Thereafter, standard histomorphometric parameters (bone-volume to total-volume ratio [BV/TV (%)], bone-surface to bone-volume ratio [BS/BV (mm^-1)], trabecular thickness [Tb.Th (μm)], trabecular number [Tb.N (mm^-1)], and trabecular separation [Tb.Sp (μm)]) were estimated and usad as a means of comparison between the true segmentation values (obtained from the numerical phantom) and the outcome of each binarization algorithm. The results pointed out that Wellner method presented in general smaller errors than the other ones. The next step is to apply this method on real μCT for quantifying bone morphology.