The final goal of my thesis is to create an algorithm, which is capable of segment the lung from other organs on chest tomosynthesis with a contour. This contour has several possible potential uses by Computer Aided Diagnosis systems. Two main options are to recognize Chronic Obstructive Pulmonary Disease by detecting shape disorders and determining the search area for a lung cancer detection method.
Digital tomosynthesis is an excellent screening technique for early detection of diseases, compared to CT (Computer Tomography), it is faster, cheaper and has lower radiation level. Considering these advantages, tomosynthesis should be used as screening method. However, after the reconstruction, the slices are not ideally thin and the are partly overlapping, causes that, images have less depth information, than CT images. This overlapping causes bigger problem on the noncentral images, the geometry of the machine causes this characteristic. This problem makes hard to segment the lung, and the evaluation as well.
During the thesis, a method is designed, which is capable of segmenting the lung on chest digital tomosynthesis images, and it is implemented in MATLAB environment. During the design the accuracy of the method is increased by using multiple methods together. For validating the results such reference images should be used where accurate lung contours are available. Because of the lack of such reference images, simulated tomosynthesis images are computed, where at least approximately correct lung contours can be determined. Based on these “reference images” the validation results can be considered as approximately correct too.