Investigating automatic techniques in segmentation accuracy of masses in digital mammography images

Segmentation techniques evaluation based on a single compact breast mass classification scheme

“…The results are in agreement with those presented in a previous work [2], where Level-Set was also considered the best segmentation when the comparison was made with the mass delineation by a specialist.…”

“…This usually results in the segmented region being very close to the actual mass, as shown in [2], where this technique proved to be the best among those tested compared to an expert's segmentation.…”

“…Following a previous research [2] , six segmentation techniques -Otsu [3] , SOM [4] , EICAMM [5] , Fuzzy C-Means [6] , K-Means [7] and Level-Set [8] -were tested. The resulting segmented nodule was evaluated using a simple centroid-based classification method, in order to decide which technique would be more recommended for using in a more complex classification method.…”

“…Usually such techniques are tested against a gold standard based on a radiologist manual segmentation [2] , but the purpose here is not finding the segmentation technique closest to an expert's segmentation, but determining which the technique provides the best classification results.…”

“…This current work focuses on six of them, each one using a distinct methodology in order to reach the results -all these techniques are briefly described in the next section. The selection of these techniques was based on a previous work [2] , in an attempt to analyze several different paths for segmentation, including neural network, adaptive contour, statistical analysis, fuzzy logic, etc.…”