A Comparitive Study of Various MicroCalcification Cluster Detection Methods in Digitized Mammograms

The presence of microcalcification clusters in mammograms contributes evidence for the diagnosis of early stages of breast cancer. Computer aided diagnosis (CAD) can be used as a useful tool for improving the accuracy of the diagnosis process, and for helping the radiologists with film interpretation. In this paper, digitized mammograms are decomposed using filter banks at several levels in the transform space. Global nonlinear operator was applied on decomposed detail subband images using multiscale adaptive gain method to enhance the images. Skewness & kurtosis were applied as detection method of the previous modification image with a specific size of region of interest (ROI). The DCT co-efficient taken as spectral features for classification of positive and negative region of interest. A three layered BPN employed as a classifier to evaluate classification efficiency. Distinction between microcalcification clusters (nodular components) and normal tissues such as blood vessels and mammary ducts (linear components) made using the eigenvalue of the Hessian matrix. Bayes discriminant function was employed for distinguishing among abnormal ROIs with a microcalcification cluster and two different types of normal ROIs without a microcalcification cluster. An integrated approach of using a filterbank, DCT and Bayesian classifier has shown to have the potential to detect microcalcification clusters with a clinically acceptable sensitivity and low false positives. The detection performance was evaluated by using 40 mammograms and showed 99% accuracy.