Performance analysis of melanoma early detection using skin lession classification system

Melanoma has been proved to be very tedious and statistical analysis which provides the majority of deaths occurs from skin cancer. The earlier detection and treatment is the best way of survival from melanoma. Clinical diagnosis of melanoma is extremely difficult due to its irregularity in edges and its shape. This paper proposes a novel scheme for early detection of melanoma using Multiclass support vector machine (MSVM). There are five different skin lesions which are grouped as Solar Keratosis or actinic keratosis, Basal Cell Cancer, Nevocytic nevus, Squamous Cell Cancer, Seborrhoeic Verruca. The proposed system uses an automatic procedure, where the queried images are grouped and matched with higher probability type to classify the type of melanoma. The multi class [6][7] support vector machine is a powerful tool for solving classification problem. The algorithm is based on learning of each stage with some training sample. Here, the color and texture features such as gradient, contrast, edges are extracted. The proposed system contains an image database which has the all five types of melanoma for testing and classification purposes. From the result of simulation, the accuracy of the proposed support vector machine scheme has comparatively high among all five types.