ANN-based diagnosis method for skin cancers using dermoscopic images

Abstract

Melanoma is one of the most dangerous skin cancers in the world. It accounts for 55% of all deaths associated with skin cancer. Researchers believe that skin cancer increases the risk of other cancers if not diagnosed early. Therefore, prompt and timely diagnosis of this disease is very important for the successful treatment of the patient. This system can detect melanoma lethal carcinoma from other skin lesions without the need for surgery, with a low cost, accuracy of about 98.88% and specificity 99%. In this article, a new, intelligent and accurate software (Delphi) system has been used to diagnose melanoma skin cancer. To detect malignant melanoma, the ABCDT rule, asymmetry (A), boundary (B), color (C), diameter (D) and textural variation (T) of the lesion are calculated and finally, an artificial neural network (ANN) is used to obtain an accurate result. The ANN with Multi-Layer Perceptron (MLP) contains the five extraction Characteristics (ABCDT) of lesions is used as inputs, two hidden layers, and two outputs. Very good results were obtained using this method. It was observed that for a dataset of 180 dermoscopic lesion images including 80 malignant melanomas, 20 benign melanomas and 80 nevus lesions. Due to its automatic recognition and ability to be installed on a computer, this system can be very useful for dermatologists as well as the general public.

Keywords

• melanoma skin cancer
• abcdt rule
• feature extraction
• artificial neural network