Pigmented skin lesion diagnosis using geometric and chromatic features

Skin cancer appears to be one of the most dangerous types among others by the presence of malignant melanoma as one of pigmented skin lesion forms. Automated system for the purpose of pigmented skin lesion diagnosis mentioned through that paper is recommended as a non-invasive diagnosis tool. To obviate the problem of qualitative interpretation, two different image sets are used to examine the proposed system, a set of images acquired by standard camera (clinical images) and another set of dermoscopic images captured from the magnified dermoscope. Images are enhanced and segmented to separate the lesion from the background. Different geometric and chromatic features are extracted from the region of interest resulting from segmentation process. Then, the most prominent features that can cause an effect are selected by different selection methods; which are the Fisher score ranking and the t-test method. Most prominent features were introduced to two different classifiers; artificial neural network and Support vector machine for the discrimination of the two groups of lesions. System performance was measured regarding Specificity, Sensitivity and Accuracy. The artificial neural network designed with the combined geometric and chromatic features selected by fisher score ranking enabled a diagnostic accuracy of 95% for dermoscopic and 93.75% for clinical images.