Fuzzy skeletonization improves the performance of characterizing trabecular bone micro-architecture

Abstract

Skeletonization provides a compact, yet effective representation of an object. Despite limited resolution, most medical imaging applications till date use binary skeletonization which is always associated with thresholding related data loss. A recently-developed fuzzy skeletonization algorithm directly operates on fuzzy objects in the presence of partially volumed voxels and alleviates this data loss. In this paper, the performance of fuzzy skeletonization is examined in a popular biomedical application of characterizing human trabecular bone (TB) plate/rod micro-architecture under limited resolution and compared with a binary method. Experimental results have shown that, using the volumetric topological analysis, fuzzy skeletonization leads to more accurate and reproducible measure of TB plate-width than the binary method. Also, fuzzy skeletonization-based plate-width measure showed a stronger linear association (R2 = 0.92) with the actual bone strength than the binary skeletonization-based measure.