Multi-Aspect Comparative Detection of Lesions in Medical Images
Juliusz Lech Kulikowski, Malgorzata Przytulsk · InTech eBooks · 2011
Symmetry is an easily observable property of a normal human body.It also occurs in the anatomy of some of its organs: motion or sensory organs, brain, dentition, breasts, kidneys, etc.This property is often used as a basis of visual diagnosis of anatomical defects or of pathological lesions in the organs, expressed by local disparities between the (generally symmetric) pairs of compared images (Rogowska J., Preston K., Hunter G.J. & al., 1995).Such approach, based on an assumption that in most cases the defects or lesions have been caused by asymmetrically acting factors, leads to a simple algorithm of lesions detection by pixel-from-pixel subtraction of matched pairs of images.However, for several reasons this approach does not lead to satisfactory results: 1 st a general symmetry of normal body organs does not mean that small anatomic differences in them cannot occur, 2 nd small local differences in compared pixel values can also be caused by image acquisition defects, 3 rd substantial differences may be hidden in specific subtle local morphological structure of analyzed organs.A comparative detection of lesions is thus a non-trivial problem needing advanced solution approach.This remark also concerns a comparison of acquired at distanced time-instants medical images of a given organ aimed at an assessment of the results of its medical treatment.A comparative lesions detection should consist not so much in a detection of any formal but rather of medically significant differences between the compared images.Medically significant image details may be manifested by occurrence of both simple differences between the local (monochromatic or multi-chromatic) pixel values as well as by occurrence of more subtle features characterizing local sub-areas in the examined images.This leads to a concept of comparative image analysis based on a multiaspect dissimilarity measure (Kulikowski J. L., Przytulska M., 2009a).The notions of similarity and dissimilarity are evidently related: the more similar two objects are, the less they are dissimilar.In certain cases, when the objects can be considered as elements of a metric (e.g.Euclidean) space their dissimilarity can strongly be connected with a distance between them.However, not all objects of medical interest, usually described by combinations of their quantitative and qualitative features, as the elements of a formally defined metric space can be considered.That is why it seems more reasonable to define dissimilarity (as well as similarity) measure as a normalized dimensionless parameter.Using the notion of multiaspect dissimilarity to comparative lesions detection seems not only to be intuitively justified but also more suitable to distinguish between the normal and pathological tissues than a distance notion.