A fully automated segmentation of radius bone based on active contour in wrist MRI data set

Advanced medical imaging techniques require high performance segmentation algorithms. Extracting the structures of interest accurately is one of the main challenges in medical imaging segmentation. In this paper a methodological approach based on active contour is proposed for fully automated segmentation of radius bone. As soft tissues like tendons, muscles and fat around the bone have close intensities to internal parts of the bone, accurate segmentation is difficult. Here the designed framework takes an MR image including radius bone as input and produces the segmented radius bone in 3D voxel set. The multi-step approach for segmentation is as following. Since our data set was much noisier, at first we denoised and enhanced contrast of the data using wavelet transform. Then an initial segmentation was produced focusing on edge map. Next, according to anatomical knowledge about the radius bone shape and size in intermediate slices, radius bone was extracted in this slice and used as the mask slice for adjacent slices. This masking procedure was applied to all slices with a 3D approach. After that we derived a convex hull region around the radius bone. This step was done for whole slices as regions of interest. Finally the estimated convex region is used as an initial mask for active contour. This framework was tested on more than 600 coronal MR slices of 23 subjects. In comparison to manual segmentation our method showed an average Dice similarity coefficient DSC and kappa statistics of 94.82% and 92.46% respectively. In the future works we utilize the proposed approach as part of a computer-aided diagnosis system for bone age estimation.