The objective of the thesis was developing methods to determinate the region of the lung in tomosynthesis made images. This task can be achieved by segmenting the intercept of the ribs with the slices. These can be used as markers to outline the position of the lung. Of course, this is applicable only if the detection of these markers is easier than identifying the lung itself. Mainly at the margin slices where the actual region of the lung is much smaller than of the chest. Therefore the main goal has become the segmentation of these intercepts.
In the thesis I applied two main phenomena, the clustering and the binary image processing combined with binary morphology. Nevertheless the most important part of the work is the used clustering method, namely the Markov Random Field (MRF). This a graph like structure which can solve optimization problems by minimizing an energy function. The form of this function guides the connections of the nodes which represent the voxels. As the graph optimizes the energy, the nodes change their state, and this state is their belongings of a cluster. The clustering methods use some kind of properties of the inputs, so does the MRF. I used the image gradient vector as the feature. The results of the MRF are promising, but in the current state of the development, it is unable to segment distinct rib interceptions by the clustering. So it was necessary to apply some other methods to extract rib candidates. Therefore I used binary morphology on the identical clustered voxels to filter out false regions and smooth their shape. After that these distinct regions are processed as blobs, and I made an attempt to utilize the 3D sequence information.