Automatic Detection on Clustered Microcalcifications on Mammograms Based on 2D Particles

Clustered microcalcifications (MCs) in mammograms can be an important early sign of breast cancer for women and are still very difficult to reliably detect by either radiologists or computer-aided diagnosis (CAD) system. Traditional filtering based detection methods are vulnerable to complex textures on mammograms. And morphology based methods have some problems in determining the structure elements. To overcome these shortcomings, a novel detection method based on 2-dimensional particles is proposed in this paper. First, a new segmentation algorithm is used to obtain 2-dimensional particles on mammograms. Then suspicious particles are picked up according to their contrasts and background intensities. An inpainting based method is applied to extract features from these suspicious particles, followed by using a classifier to separate the microcalcifications from normal particles. Experiments on the DDSM database show that the new detection method can obtain 93.3% true positive rate with average 3.1 false positives per image. Compared with the original fast marching algorithm, the new segmentation method proposed greatly reduces the time consumption in 2-dimensional particles segmentation on mammograms. Experiments on the DDSM database show that the average time consumption of the new segmentation method decrease by nearly nineteen-twentieth. The new segmentation method presented can also be used in other image processing applications.