Automated detection of clustered microcalcifications on mammograms: CAD system application to MIAS database

To investigate the detection performance of our automated detection scheme for clustered microcalcifications on mammograms, we applied our computer-aided diagnosis (CAD) system to the database of the Mammographic Image Analysis Society (MIAS) in the UK. Forty-three mammograms from this database were used in this study. In our scheme, the breast regions were firstly extracted by determining the skinline. Histograms of the original images were used to extract the high-density area within the breast region as the segmentation from the bay area around the skinline. Then the contrast correction technique was employed. Gradient vectors of the image density were calculated on the contrast corrected images. To extract the specific features of the pattern of the microcalcifications, triple-ring filter analysis was employed. A variable-ring filter was used for more accurate detection after the triple-ring filter. The features of the detected candidate areas were then characterized by feature analysis. The areas which satisfied the characteristics and specific terms were classified and displayed as clusters. As a result, the sensitivity was 95.8% with the false-positive rate at 1.8 clusters per image. This demonstrates that the automated detection of clustered microcalcifications in our CAD system is reliable as an aid to radiologists.