Bone marrow cavity segmentation using graph-cuts with wavelet-based texture feature

Hironori Shigeta, Tomohiro Mashita, Junichi Kikuta, Shigeto Seno, Haruo Takemura, Masaru Ishii, Hideo Matsuda

Abstract: Emerging bioimaging technologies enable us to capture various dynamic cellular activities in vivo. As large amounts of data are obtained these days and it is becoming unrealistic to manually process massive number of images, automatic analysis methods are required. One of the issues for automatic image segmentation is that image-taking conditions are variable. Thus, commonly, many manual inputs are required according to each image. In this paper, we propose a bone marrow cavity (BMC) segmentation method for bone images as BMC is considered to be related to the mechanism of bone remodeling, osteoporosis, and so on. To reduce manual inputs to segment BMC, we classi¯ed the texture pattern using wavelet transformation and support vector machine. We also integrated the result of texture pattern classi¯cation into the graphcuts- based image segmentation method because texture analysis does not consider spatial continuity. Our method is applicable to a particular frame in an image sequence in which the condition of °uorescent material is variable. In the experiment, we evaluated our method with nine types of mother wavelets and several sets of scale parameters. The proposed method with graph-cuts and texture pattern classi¯cation performs well without manual inputs by a user.