ليلى محمد جبر الكريطي

Cancer is one of the types of diseases that has caused more deaths in humans in the last century than in previous centuries. Skin cancers account for the largest share of all types of cancer, accounting for 40%. Environmental pollution, spending a lot of time under sunlight, etc. are among the main causes of the spread of this disease. If skin cancer cells are detected, classified, and identified in their early stages, it will play a fundamental and decisive role in the treatment of this disease. In recent years, with the emergence of deep learning and neural networks, the accuracy of diagnosis and classification has improved significantly. One of the most widely used deep learning methods in this field has been the convolutional neural network. The deep convolutional neural network, with its various types such as U-net, VGG-16, VGG-16, Alex-Net, etc. models, has played a tremendous transformation in the diagnosis and classification of various diseases, including the diagnosis of skin lesions. However, these models are very complex due to their multiple layers, and basically, these models are designed to respond to very large-scale challenges, and training these models using images from smaller databases will definitely not yield desirable results. Therefore, the best idea for newer challenges is to design a model inspired by these models, tailored to the existing conditions and data, and with less complexity. In this research, a model inspired by the VGG-16 neural network is presented, which uses only three of the five main depths of this model, but the number of convolution channels in the second and third depths has been doubled. Laboratory results on IBSB-2018 data show that the proposed method has significantly improved with an accuracy of 98.73%.