An Approach to Semiautomatic Segmentation of Lung Region from 3D-color Images of Frozen Human Body

Segmentation of internal organs from 3D medical images is increasingly focused on by many medical engineering researchers. This paper proposes a new approach to extract the lung region from the 3D-Color Image of Frozen Human Body (Visible Human Male). One of the effective segmentation methods for the medical images is based on the watershed-algorithm. Generally this method has the following steps. First, we detect the edge of the original image. Next, we apply the watershed-algorithm to the image. Watershed-algorithm segments image into too many small regions, so usually, we merge these small regions into some groups. To merge regions, some measures of similarity is used. Finally, we select the target group. But with only these steps, the group which is selected is not good. In our case (we use Euclidean distance on RGB color space for the measure), the group includes some regions outside of the target region. It is because that the voxels of small regions have so similar color that the merging step combines the small regions unnecessarily. For example, the over-merged regions are liver, muscle, heart, bronchi, kidney and some other regions. Our goal is how we cut off these unnecessary regions from the selected group and obtain the target lung region. To avoid over-extraction, we introduce three other steps. They are “Edge-Marking”, “Reconstruction (shrinking, selecting, and expanding) based on small regions”, and “graph-partitioning algorithm”. “Edge-Marking” is the process that separates two small regions by giving strong edge feature to boundary voxels between them. “Reconstruction based on small regions” and “Graphpartitioning” are the processes that separates the regions which are connected with narrow bridge. Adding these three processes for the region, other than liver and kidney regions are separated from the lung region. This paper describes these processes and experimental results.