[Matching of mammary microcalcifications on images from different views]

In order to increase sensitivity and specificity in the diagnosis of breast cancer, more than one - generally two or three - mammograms of the same breast are obtained. In order to be able to carry out 3D reconstruction of microcalcification clusters on one hand, and efficiently fuse the information they carry on the other hand, one needs to match corresponding regions of microcalcifications from the different views. Unfortunately, this may be difficult at times. To help the situation, we have developed in this paper, a technique to automatically match microcalcifications found on pairs of mammograms from the same breast.

We used the computed morphological characteristics of individual microcalcifications to build a likeness function to match the microcalcifications. The geometrical constraint suitable for the system used was then applied to reduce the possibilities. From the remaining possibilities, the one with the highest likeness was selected as pair of corresponding microcalcifications.

This technique was tested on a number of real cases and yielded 77.14% of good matches.

This technique provides good results and could therefore be used either directly or for 3D-reconstruction of clustered microcalcifications.