CT Segmentation

A second application for segmentation in 3d is the classification of CT images into the classes background, soft tissue, enclosed air and bones. Especially the automatic classification of bones from CT images is an important task in medical diagnostic, surgery planning and visualization. Often the classification of bones in CT images is done just by global or local thresholding methods. Such methods work well as the gray values of a CT image are calibrated (Hounsfield scale) and every tissue has a defined gray value range. Unfortunately, the gray level ranges of tissues may overlap. Thus some spatial and topological information should be used to distinguish different types of tissue.xOne essential step of our CT bone classifier is the CSC segmentation. As CT images are calibrated we choose a fixed small similarity threshold. A preprocessing step like the Kuwahara filter may be used to reduce noise, however this might destroy very thin bone structures especially in low resolution images. The segmentation results in many homogeneous segments which have to be classified using topological information.The histogram of a CT bone image has three significant peaks corresponding to the classes (from dark to bright) air (this means background and enclosed air), soft tissue and bones. Those peaks are detected and together with topological information used to classify the CSC segments.

CT Segmentation Video (5.6Mb)