Resolution effects on the morphology of calcifications in digital mammograms

The development of computer assisted diagnosis (CAD) techniques and direct digital mammography systems have generated significant interest in the issue of the effect of image resolution on the detection and classification (benign vs malignant) of mammographic abnormalities. CAD in particular seems to heavily depend on image resolution, either due to the inherent algorithm design and optimization, which is almost always resolution dependent, or due to the differences in image content at the various resolutions. This twofold dependence makes it even more difficult to answer the question of what is the minimum resolution required for successful detection and/or classification of a specific mammographic abnormality, such as calcifications. One may begin by evaluating the losses in the mammograms as the films are digitized with different pixel sizes and depths. In this paper we attempted to measure these losses for the case of calcifications at four different spatial resolutions through a simulation model and a classification scheme that is based only on morphological features. The results showed that a 60 μm pixel size and 12 bits per pixel should at least be used if the morphology and distribution of the calcifications are essential components in the CAD algorithm design. These conclusions were tested with the use of a wavelet-based algorithm for the segmentation of simulated mammographic calcifications at various resolutions. The evaluation of the segmentation through shape analysis and classification supported the initial conclusion.