A novel microcalcification shape metric to classify regions of interests

This paper proposes a microcalcification shape feature to aid in classifying regions of interest that are difficult to diagnosis. The proposed feature extraction method is based on a wavelet approximation of a microcalcification´s contour distance sequence, which is the Euclidean distance of each contour point to the centroid. A novel metric is proposed to quantify the roughness of a microcalcification. Experimental results from evaluations that include several published features are presented to establish the robustness of the proposed roughness feature.