Texture features in the classification of melanocytic lesions
The use of different texture features for the classification of melanocytic lesions is studied in an attempt to develop a computerized method for the early detection of melanoma. The computer laboratory at the University of Oulu has a strong tradition in applications of computer vision to visual inspection for industrial quality control, and some of the methods learnt in these applications are being transferred and experimented with in this medical context. This include the utilization of texture distributions for classification purposes.
To avoid the effect of different photographing systems, all the images are first converted to intensity images and then the lesion parts are divided into 3202 rectangles in order to obtain the maximal number of non-overlapping samples. The divided images are then normalized by z-score transformation and texture feature distributions are counted for the rectangular samples and classified into melanoma and benign nevus with a k-nearest neighbour classifier. 78–99% of the test samples were found to be classified correctly, depending on the texture feature used.
KeywordsTexture Feature Local Binary Pattern Melanocytic Lesion Dysplastic Nevus Benign Nevus
- Schmoeckel, C., Wagner-Grösser, G. and Braun-Falco, O., Klinische diagnostik initialer maligner melanome, Der Hautarzt, Vol. 36, 1985, pp. 558–562.Google Scholar
- Stoecker, W.V., Moss, R.H., Ercal, F. and Umbaugh, S.E., Nondermatoscopic digital imaging of pigmented lesions, Skin Research and Technology, Nr 1, 1995, pp. 7–16.Google Scholar
- Teppo, L., Ihosyöpä Suomessa, Suomen Lääkdrilehti, 28/91 vsk 46, pp. 2598–2601.Google Scholar
- Osterlind, A., Jensen, O. M., Hudpigmentering, naevi og sollys som aetiologiske faktorer til kutant malignt melanom, Nordisk Medicin, Vol. 195, 8-9/1990, pp. 223–229.Google Scholar
- Ries, L.A.G, Hankey B.F., Miller B.A. et al. Cancer statistics review 1973–1988. Bethesda, Md: National Cancer Institute. Surveillance Epidemiology and End Results (SEER) data base. NIH publication Nr. 91-2789, 1991.Google Scholar
- Ganster, H., Gelautz, M., Pinz, A., et al. (1995) Initial results of automated melanoma recognition, Proceedings of the 9th Scandinavian Conference on Image Analysis, Uppsala, Sweden, pp. 209–218.Google Scholar
- The Khoros Group (1991) Khoros User's Manual & Programmer's Manual. University of New Mexico, USA.Google Scholar
- Milligan, G.W. and Cooper, M.C., A study of standardization of variables in cluster analysis, Journal of Classification, Nr 5, 1988, pp. 181–204.Google Scholar
- Laws, K.I., Textured image segmentation, Image Processing Institute, University of Southern California, Report 940, 1980.Google Scholar
- Swain, M.J. and Ballard, D.H., Indexing Via Color Histograms, Proceedings of Third International Conference on Computer Vision, Osaka, Japan, December 4–7, 1990, pp. 390–393.Google Scholar
- Stolz, W., Hölzel, D., Riemann, A., et al. (1991) Multivariate analysis of criteria given by dermatoscopy for the recognition of melanocytic lesions. In: Book of Abstracts, Fiftieth Meeting of the American Academy of Dermatology, Dallas, Tex: Dec. 7–12.Google Scholar