Bone Segmentation in CT-Liver images using K-Means Clustering for 3D Rib Cage Surface-Modeling

A 3D rib cage model helps to study anatomical structures in some medical applications such as biomechanical and surgical operations. Its quality directly depends on rib cage segmentation if it is reconstructed from image data. This paper presents an optional segmentation method based on K-means clustering. It uses a hierarchical concept to control the clustering, and it organizes clustered regions in subsequent indexes of background, soft-tissue, and hard-tissue regions. We applied the proposed method to 3D CT-liver images acquired by a 4D-CT imaging system. The proposed method was compared with 2D K-means (KM) and 2D fuzzy C-means (FCM) clustering. From our experiment, the proposed method gave more stable clustering results under a condition of randomization in initial cluster-centers, and it performed faster than 1.5 times of 2D-KM and 7.7 times of 2D-FCM on average. For 3D surface models, the results of the proposed method provided more information of bone regions in vertebra, ribs, and scapula areas than results of 2D-KM and 2D-FCM. Key–Words: 2D K-mean clustering, 2D fuzzy C-mean clustering, and rib cage segmentation