Abstract
Micro-calcifications are known to be early signs of breast cancer. Realizing that early detection is important in cancer prevention, our group developed an automatic micro-calcification detection algorithm. This algorithm can be subdivided into four processes: (1) First, using a threshold selection method based on discriminant and least squares criteria, two threshold are chosen to separate the breast region from the background. Afterwards, additional noise eliminating processes are applied to improve the extraction precision. (2) Next, Tophat processing, a mathematical morphology application, is used to smoothen non calcific regions in the breast region. Consequently, the micro calcifications are emphasized. (3) The Tophat image is then binarized to specify the region of calcification candidates, by a threshold value obtained directly using a semi-automatic (image histogram-dependent) thresholding method. (4) Finally, to improve extraction precision, further elimination of noise components in calcification candidates using a medically based area thresholding is performed. Evaluation of the algorithm was carried out on 14 samples. From these, 25 candidates (14 were calcific and 11 were normal) were obtained. Misclassification occurred in three cases; 1 calcific candidate was classified as normal and 2 normal candidates were classified as calcific.
