Automated system for skin cancer classification
Skin Melanoma is the most deadly, and statistically, one of the most widespread diseases in the world. It primarily represents the hysterical growth of abnormal cells in the skin and is an aggressive cancer which can be healed by surgical excision successfully only if it is recognized in the early stage. The usual clinical practice of melanoma diagnosis is a visual inspection by the dermatologists and calls for extreme caution. However, to account for the loss of accuracy in clinical diagnosis, dermatoscopy is employed. Computerized analysis of the digitized images is receiving prominence in the last two decades due to its significant ability to accurately classify melanoma. Several motivating systems have been developed. Each of them have their advantages and disadvantages. The prime goal of this thesis is to solve the problem of early diagnosis of skin cancer by utilizing a new automated imaging system that classifies the images of skin cancer into benign and malignant melanoma. A new class of feature vectors called Averaging Centre-Symmetric Local Binary Pattern provides critical asymmetric texture description. Furthermore, the Gaussian image pyramid stages are integrated and uniquely used for Gray-level Co-occurrence Matrix texture-based autocorrelation feature extraction. A modification to l1/l2/l3 space using HSV color model delivers color difference information. The developed system is tested using the clinical dataset from the PH2 database . Computer simulation provided accuracy as high as 95.8%, sensitivity of 91.6% and specificity of 98.6%, which illustrates that it was able to overhaul the performance levels of several existing systems. The proposed solution to the existing problem of early diagnosis of skin cancer will provide symbolic assistance to dermatologists.