Multiresolution Shape Descriptions and their Applications in Medical Imaging

Medical image analysis: progress over two decades and the challenges ahead

“…Another interesting notion related to segmentation brought up during this time frame was the concept of applying forms of scale space theory to the problem, addressing the idea that relevant medical image features show up at a variety of scales, even within the same image dataset. Intending to apply the basic theory developed by Koenderinck [37] and others, a variety of researchers, many from the Universities of North Carolina in the United States and Utrecht in The Netherlands [38], [39] vigorously pursued this direction during this time period with the idea being that alternative spaces derived from scale-space hierarchies of intensity extrema would be useful in negotiating the segmentation (automatic or semiautomatic) of complex medical image data. These approaches ultimately lead to the notions of combining medial primitives with scale space concepts to develop a unified approach to shape description, useful for guiding object segmentation.…”

Multi-scale gradient magnitude watershed segmentation

“…Also split and merge techniques [9] have been introduced. However, this strategy is captured more elegantly in multiscale linking approaches [4,10,11,12,13].…”

“…Multi-scale watershed segmentation has been carried out based on the intensity and ridges: Gauch [12] used the image intensity function directly as local measure of homogeneity. Eberly [13] defined a homogeneity measure based upon local "ridgeness'.…”