|
|
Intelligent Synchronized Magnifying Glasses for Assisting Reading of Temporal Mammograms Host Publication: Finds and Results from the Swedish Cyprus Expedition: A Gender Perspective at the Medelhavsmuseet Authors: F. Temmermans, R. Deklerck, M. Suliga, C. Breucq, G. Behiels and P. Dewaele Publication Date: Nov. 2008
Abstract: This paper describes a framework which aims at assisting the clinical analysis of temporal mammograms. The framework exists of a processing part and a visualization part. The core components of the processing part are (1) an algorithm for spatial registration of two successive mammograms, (2) a method for automatic detection of Micro-Calcifications (MCs) and Micro-Calcification Clusters (MCCs) and, (3) an algorithm that allows an automatic identification of detected MCs and MCCs between two successive images. For the presentation of registration, detection and identification results, we present a visualization tool, called "Intelligent Synchronized Magnifying Glasses" (ISMAG), which allows a visualization of the results in an intuitive and non distracting way. ISMAG provides two operational modes, the "random mode" and the "discrete mode". In the random mode, clicking on any position will cause the highlighting of the neighborhood of the selected location and the neighborhood of its corresponding location within the temporal counterpart as well. The corresponding position in the second image is determined by applying the affine transformation, which we computed during the processing phase, to the coordinates of the selected position in the clicked image. The registration algorithm is based on a mutual information metric. The user can browse through one of the images entirely by moving his mouse pointer and follow the corresponding content in the other image simultaneously. In this way, he can systematically inspect the complete images and immediately evaluate possible changes at any position. Although the affine registration may bring the overall breast shapes into correspondence quite well, still some local regions may be slightly displaced due the difference in compression level between the two mammography image acquisitions. In this case the discrete mode, in which the clickable positions are limited to Regions Of Interest (ROIs) containing MCs or MCCs, may be a better option. For the detection of MCs and MCCs a support vector machine based method is used. In the discrete mode the correspondence between ROIs is achieved using the results of the temporal analysis, which is based on a markov random field based graph matching algorithm.
|
|
