A confocal laser scanning microscope segmentation method applied to magnetic resonance images.

Segmentation is the process of defining distinct objects in an image. A semi-automatic segmentation method has been developed for biological objects that have been recorded with a confocal laser scanning microscope (CLSM). The CLSM produces a sequence of thinly "sliced" images that represent cross-sectional views of the sample containing the object of interest. The cross-sectional representation, or "seed" is created of the object of interest within a single slice of the image stack. The segmentation method uses this "seed" to segment the same object in the adjacent image slice. The new "seed" is used for the next image slice and so on, until the object of interest is segmented in all images of the data set. The segmentation method is based on the idea that the object of interest does not change significantly from one image slice to the next. The segmented information is then used to create 3D renderings of the object. These renderings can be studied and analyzed on the computer screen. Previous work has demonstrated the usefulness of the algorithm as applied to the CLSM images. This paper explores the application of the segmentation method to a standard sequence of magnet resonance imaging (MRI) images. Typical MRI machines can produce impressive images of the human body. The resulting data set is often a sequence, or "stack" of cross-sectional slice images of a particular region of the body. The goal then, is to use the previously described segmentation method on a standard sequence of MRI images. This process will expose limitations with the segmentation method and areas where further work can be directed. This paper illustrates and discusses some of the differences between the data sets that make the current segmentation method inadequate for segmentation of MRI data set. Some of the differences can be corrected with modification of the segmentation algorithm, but other differences are beyond the capabilities of the segmentation method, and can possibly be addressed in other ways. The lessons learned from this research process will lead to a more capable and robust segmentation algorithm.