Introduction: Mammography has recently become the main diagnostic method for detecting breast abnormalities. Achieving an optimal image quality in mammography is very technically difficult for radiologic technologist. With regular assessment of the quality of mammographic images technical errors can be discovered and improved.
However, since image estimation can be time consuming, and at the same time estimation is quite subjective, the use of automatic segmentation would significantly shorten the whole process. Segmentation is the process of determining a homogeneous region of interest (ROI) and demarcating it from the background. Basically, we divide the segmentation methods into methods, that are based on combining the pixels into homogeneous areas and the methods based on determining the edges between areas. Purpose: The purpose of the master's thesis is a theoretical overview of existing segmentation methods by individual types of diagnostic procedure. In the next step we will try to perform segmentation procedures for the detection of certain elements used to evaluate radiological engineers in the DORA program. Based on image testing, we will determine how successful the use of automatic and semi-automatic segmentation procedures in detection of pectoral muscle in images is. Methods: in 250 mammographic MLO images, we will first manually mark the breast and the pectoral muscle. Then we will perform the automatic segmentation of breast and pectoral muscle. We will use thresholding to mark the breast and the region growing and clustering for segmentation of the pectoral muscle. We will be mainly interested in the difference in the areas of the curves between the obtained and the reference image. Results: We used thresholding to separate the background from the object. There were minimal deviations, namely the average value of the relative error rate was 1.13 %. Bigger errors occurred in pectoral muscle segmentation. In the method of region growing, the average value of the relative error rate is 22,83 %, and in the method of clustering 32,7 %. Discussion and conclusion: Due to a big difference in contrast between the background and the breast, the thresholding is suitable for segmentation of the entire breast. However, if we compare the methods for segmentation of the pectoral muscle, the region growing method is more appropriate. Major deviations occur due to a technically inadequate picture and a larger amount of glandular tissue.
|