| Malignant melanoma (MM) is a cancer that originates from the melanocytes, the cells that produce the skin color known as melanin. The incidence of melanoma has increased in a faster rate than any other cancer cases in the United States, which makes it the eight most common cancers in the country. This number is expected to grow in the future.;Superficial spreading melanoma is the most commonly diagnosed type of melanoma. Its manifestation has close similarity to benign Dysplastic nevi (DN), which gives rise to difficulties in its diagnosis. Clinical detection of MM has been done visually by experts. An ABCD guideline was established to assist experts and general practitioners with their diagnoses. The same method was also used to educate the public for self-examination. However the method is not viable due to great variability in the manifestation of the MM lesion. Therefore, success in diagnosis was highly subjective that it depends on the extent of experience of the experts performing the diagnosis.;The advancement in technical computing opens new windows of opportunities for their medical applications. In the area of skin cancer, people investigated different algorithms to characterize MM from DN---most of them were highly successful. Many of the methods were focus on color and morphological information of the lesion, and not too many on textures. Visual texture is a property of an image that provides its characteristics/definition. Visual textures had been studied in the past and technical definitions were established. Nonetheless, no unique term could adequately describe it. Malignant melanoma has visual texture and so is DN. Therefore, the objective of this study was to extract several relevant texture parameters from these skin lesions and use them to create a classification system. The system will be a viable tool not only for general physicians but also for the experts in providing more accurate and reproducible diagnosis.;In this research, a skin lesion classification system for MM and DN was constructed. Skin was categorized into MM, DN, and normal skin. First, appropriate segmentation method was determined by examining different method from frequency and spatial domain. Based on quantitative comparison, the wavelet method based on multi-scale edge detector (Mallat) was used and explored for its applicability to skin lesion segmentation. Afterwards, three visual texture descriptors were calculated that include contrast, homogeneity, and energy. The descriptors were then used to train a supervised, multi-layer artificial neural network system.;Based on the results, the decision system that was constructed produced 98.72% correct classification. The system has good performance in discriminating between MM and DN on both training and test dataset. . In addition, the sensitivity and specificity of the system to melanoma is 89.37% and 97.78% respectively. The sensitivity and specificity of the system to dysplastic nevi is 86.59% and 98.07% respectively. Nonetheless, the sample size(n) used for the construction was relatively small (MM=39, DN=39). Therefore as part of the future work, more samples will be included. Ideally, the samples desirably cover the whole manifestations of Superficial Spreading MM and DN. In addition, more relevant descriptors will be integrated to increase reliability of the system. |
