Skip to Main Content
Mammography is a widespread imaging technique for the early detection of breast cancer. Microcalcification clusters, visible in X-ray images, are important indicators for the diagnosis. In the past, many image processing methods were developed to detect and to classify lesions as being malignant or benign using only cluster data extracted from the 2D-images. However, a microcalcification cluster is a 3D-entity whose shape is also an important information for radiologists. This paper presents a method for the 3D-reconstruction of microcalcification positions defining the cluster shapes. The key idea of the reconstruction algorithm lies in the modelling of mammographic units using a camera with virtual optics. This model can be used to calibrate digital systems with different geometries and with various physical acquisition principles. The different steps of the computer vision problem related to the cluster reconstruction (namely the acquisition system calibration, the microcalcification segmentation, the microcalcification matching and the 3D-reconstruction) are described. First results are then given for two phantoms. Tests with one phantom show that the inherent mean accuracy of the 3D-microcalcification localization algorithm is 16.25 μm. The other phantom is made of materials simulating the behaviour of both mammary tissue and microcalcifications towards X-rays. Tests using this phantom prove that the algorithm is effectively able to restitute true cluster shapes. Finally, patient data are used to reconstruct real clusters and to check the algorithm validity. These results prove that the proposed cluster reconstruction algorithm is the first one which is usable in clinical situations.
Date of Conference: 7-9 Sept. 2005