On the Reliability of the Fractal Dimension as a Scalar Characteristic of the Medical Images’ Contours

Medical images typically have irregular and fragmented contours. This is a strong motivation to use fractal geometry, rather than Euclidian geometry, for their description and characterization. In this paper we analyse a set of 100 images of melanoma and non-melanoma moles. The moles have their contours extracted with several tools and then the contours have their fractal dimension computed with distinct estimators. We have used descriptive statistics to depict that the fractal dimension does not give clear classification or systematization of the moles. We have also applied the student’s t-test to show that in the considered cases the two sets of fractal dimensions of melanoma and non-melanoma moles are not statistically different. Key-Words: moles, contours, medical diagnosing, fractal dimension, reliability, student’s t-tests.