The long term goal of this work is to ease the burden of melanoma detection by development of an automated screening device capable of distinguishing between malignant and benign pigmented lesions (melanomas and nevi). This device will be based on dermatoscopic imaging and computer vision technology. Support for this approach includes the increased diagnostic accuracy associated with clinical use of the dermatoscope and the documented suitability of dermatoscopic images for automated analysis. Clinical dermatoscopists base their analysis on a group of features that includes the p i g m e nt network, brown globules, black dots, diffuse pigmentation, depigmentation, and blue-gray veil. The histopathologic correlates of each of these features are known and the contribution that each of them makes towards a benign or malignant diagnosis is understood in quantitative terms. The immediate goal of this project is to develop software for automated detection and characterization of each of these features in digital images. Binary skeletonization procedures have already been developed for analysis of pigment network topology, but their performance varies with the width of the network lines. Pyramidal, multiscale processing and/or greyscale skeletonization procedures will be incorporated into algorithms capable of detecting network elements of any width. Algorithms for evaluation of network line width v a riability, hole size variability, radial streaming, pseudopods, and abruptness of the pigment network boundary will be developed. Algorithms will be developed for evaluation of the size and spatial distribution of black dots and brown globules, and for evaluation of the other """"""""classical"""""""" dermatoscopic features described above. Algorithms will be developed for rejection of artifacts caused by air bubbles and hairs. A classifier for discriminating nevi and melanomas will be constructed from the extracted features, using neural or Bayesian networks. The entire software package will closely model human perception and analysis of dermatoscopic images, and will be ready for incorporation with image acquisition hardware into an integrated screening device.
| Fleming, M G (2001) Digital dermoscopy. Dermatol Clin 19:359-67, ix |
| Fleming, M G; Steger, C; Zhang, J et al. (1998) Techniques for a structural analysis of dermatoscopic imagery. Comput Med Imaging Graph 22:375-89 |
