An Adaptive Similarity Measure for Automated Identification of Breast Lesions in Temporal Pairs of Mammograms for Interval Change Analysis

Abstract

An adaptive similarity measure (ASM) is designed to improve automated identification of corresponding lesions on prior mammograms. It combines adaptive filtering to enhance the lesion and a similarity measure (SM) as a figure-of-merit (FOM) measure. The filters are designed with a training set to maximize and minimize the FOM for the similar and dissimilar image pairs, respectively, by using a gradient optimization technique. The ASM was applied to the final stage of our multistage regional registration technique for identification of the mass on the prior mammogram. A search for the best match between the lesion template from the current mammogram and a structure on the prior mammogram was carried out within a search region, guided by the ASM. This new approach was evaluated by using 179 temporal pairs of mammograms containing biopsy-proven masses. 86% of the estimated lesion locations resulted in an area overlap of at least 50% with the true lesion locations. The average distance between the estimated and the true lesion centroids on the prior mammogram was 4.5 ± 6.7 mm. In comparison, the correct localization and the average distance using a conventional correlation SM were 84% and 4.9 ±7.0 mm, respectively. The ASM improved the identification of the corresponding lesions on temporal pairs of mammograms.