Skin Cancer Detection Based on Deep Learning Methods

Abstract

One of the most common and dangerous types of tumors/cancer is skin cancer. It is an abnormal growth of the skin. Skin cancer that is not detected early spreads to other organs. In the era of deep learning technology and computer vision, determining whether or not a patient has cancer has become more accessible because the earlier the diagnosis is, the more it reduces the risk or prevents the patient’s death. Artificial intelligence has a lot of applications in healthcare, especially in skin cancer detection. In this paper, A convolutional neural network is used to make a diagnosis system more accurate in diagnosing whether the patient has cancer or not. HAM10000 is a Dataset which used to train the model, and it is used to classify between cancer and no cancer. MATLAB is a software tool we used when we train deep learning algorithms were used to make our diagnosis systems. To achieve the highest level of accuracy, two algorithms were used to optimize the parameters and apply them in GoogleNet, MobileNet-v2, NasNet mobile, SqueezNet, Darknet19, VGG16, Xception, ShuffleNet, Inception-v3, resnet18, resnet50 and nasnetlarge. Then the result of each of them is recorded, and the best algorithm is selected. Xception is used to make a diagnosis system with 96.66% accuracy to classify between cancer and no cancer.