Microcalcification detection using independent component analysis

In this paper, the Independent Component Analysis (ICA) is employed to estimate unknown, statistically independent sources that form the regions of mammograms according to the linear mixing model. The minimum description length criterion is used to estimate the number of sources. The transformation coefficients of ICA are used as classification features of regions of interest. The extracted features are then fed into a three-layer back propagation neural network classifier to decide whether the region containing microcalcifications (MC) or not. The free-response receiver operating characteristic (FROC) analysis is used to evaluate the performance of the scheme. The results demonstrate that the designed system is able to detect approximately 90% of the true clusters with an average of one false cluster detected per image. Thus, it can be effectively used in computer-aided diagnostic prompting.