Skin Cancer Detection from Macroscopic Images

Automatic diagnosis of skin cancer images is especially difficult in medical image processing. Moreover, proper segmentation is crucial for the partitioning of growths from the skin, which can aid in the differentiation between melanoma and benign skin lesions. To address these issues, this research work investigates the widely used ABCD rule (Asymmetry, Border Irregularity, Colour and Diameter) on macroscopic images and the Graph-Cut segmentation technique as it demonstrates capabilities for handling extremely textured, noisy and colour images which are present in macroscopic images. The accuracy rates achieved by the proposed model with the use of the TDS (Total Dermoscopy Score) classifier is 73,529%, SVM is 75,294% and KNN classifier is 74,706%.