| Abstract: |
Medical image segmentation is an essential technique in diagnosis and image-guided interventional procedures. Anatomical structures must be identified and extracted from the images for understanding their relationship to therapeutic agents and devices and their response to treatment. Current clinical segmentation practices are often manual and inherently inefficient, requiring skilled and well-trained users, thus motivating the need for computer aided methods. In this paper we present an accurate semiautomatic segmentation method that dramatically reduces the time and effort required of expert users. We formulate the segmentation problem in terms of a Conditional Random Conditional Random Fields (CRFs) model of the image. With this model, maximum a posteriori (MAP) inference computes the optimal segmentation by minimizing an energy function. The energy function we present has both statistical regional and boundary terms. We estimate these statistical terms by giving a medical expert user an intuitive graphical interface to indicate samples of the target and non-target tissue by loosely drawing a few brush strokes on the image. Thus the need of assumptions on boundary contrast previously used by many other methods is eliminated. We show that boundary statistics provided on one or a few 2D slices of a volumetric medical image can be propagated through the entire 3D stack of images along with local regional statistics to achieve high accuracy. In addition, in terms of contour delineation from interpolation, we show the advantage of our method over a traditional surface-based interpolation method. The combination of a fast segmentation and minimal user inputs that are reusable, make this a powerful technique for the segmentation of medical images. |
| Notes: |
---
- Proceedings of the 2009 International Conference on Image Processing, Computer Vision, and Pattern Recognition, IPCV 2009
- Proc. Int. Conf. Image Process., Comput. Vis., Pattern Recogn., IPCV
- "Conference code: 91960"
- "Export Date: 4 September 2012"
- "Sponsors: United States Military Academy, Network Science Center; HST Harvard Univ. MIT, Biomed. Cybern. Lab.; Argonne's Leadersh. Comput. Facil. Argonne Natl. Lab.; Univ. Illinois Urbana-Champaign, Funct. Genomics Lab.; University of Minnesota, Minnesota Supercomputing Institute"
- 13 July 2009 through 16 July 2009
- "Source: Scopus"
|
