Investigation of Color Difference in Diagnosis of Melanoma

Skin cancer has become more common in recent years. Melanoma is the deadliest type of skin cancers. Early diagnosis increases the success of treatment in all types of cancer. The success of early diagnosis depends on the attributes used to identify the disease. Weak attributes require more complex classifiers and increase diagnostic error. Color features were an important indicator in the diagnosis of melanoma. The most important of these was the number of color, but no specific method has been proposed in the literature for its calculation. In this study, a method that provides the determination of the required threshold values to accurately calculate the number of colors of the lesions was proposed and a new attribute named color difference has been defined. For the threshold values determined in the proposed method and the threshold values in the literature, the color numbers and color difference attribute of all samples were calculated. Calculated attributes were analyzed statistically. The results of the analysis showed that the number of colors and color difference attribute calculated with the proposed method were significant. However, it was observed that the number of colors calculated with the threshold values used in the literature did not have statistical significance. Finally, univariate classification was made with the proposed color difference attribute and the number of colors calculated according to the methods in the literature. The classification results were compared in terms of f-measure and it was found that the color difference attribute was 25.8% to 31.9% more successful than the other number of colors used in the literature. The results obtained showed that the importance of correctly determining the number of colors of the lesions and the proposed color difference attribute were quite effective.