A knowledge-based fuzzy clustering method with adaptation penalty for bone segmentation of CT images

Accurate segmentation is critical in many advanced imaging applications such as volume determination, radiation therapy, 3D rendering, and surgery planning. However, due to the complex anatomical structure of tissue and organs, as well as artifacts caused by patient motion, beam hardening, and partial volume effect in CT image, the boundaries between different regions are smeared. In addition, the intensities of bone voxels vary widely that some of them are so close to that of the muscle. They all make the extraction of bone out of surrounding tissue quite difficult in CT images. In this study, a knowledge-based fuzzy clustering method was proposed, which was formulated by modifying the objective function of the standard fuzzy c-means (FCM) algorithm with additive adaptation penalty. Since the membership of voxels in boundary regions is intrinsically fuzzy, unsupervised fuzzy clustering methods turns out to be particularly suitable for handling the bone segmentation problem. The knowledge-based fuzzy clustering method was tested by patient CT images. Experimental results demonstrated that while the conventional FCM methods might loss a significant amount of bone volume during segmentation, the proposed method could improve the performance of bone extraction significantly.