Multispectral imaging methods for the diagnosis of skin cancer lesions

Abstract

Skin cancer is the most prevalent form of cancer, and melanoma is one of the most threat disease of it. But it can be cured if it is detected early enough. Multispectral imaging is a potential method to differenciate melanoma from nevi as it provides spectral images with information of absorbance and reflectance. With this aim, spectral images along the visible and near infrared range (from 415nm to 995nm) of 165 lesions including nevi, melanomas and basal cell carcinomas were processed in this master thesis. After obtaining all data in terms of reflectance and absorbance and other related parameters for each pixel of the segmented lesions, a statistical analysis was carried out to quantify their spatial distribution all over each lesion. Algorithms such as Support vector machine (SVM) and Discriminant Analysis (DA) were used as a means of classifying the lesions. The results show that DA linear classifier provides a better diagnosis than the SVM. BCCs are easier to discriminate from nevi than melanomas.