Classification of Skin Tumor Lesions Based on Transfer Learning Using Dermoscopic Images

Skin tumor is one of the brutal conditions of cancer types. An up-repaired deoxyribonucleic acid causes it in cells’ skin, which generates mutations on the skin. With the rapid growth of skin cancer cases with the high mortality rate, the detection of skin tumors is essential, especially at the earliest stage, because it starts tending gradually to spread over the body parts. In this regard, researchers have developed some detection techniques for skin lesion detection using the ISIC 2019 dataset. Assessing the gravity of skin tumors, this paper presents improved detection results by implementing state of the art deep learning approach known as transfer learning. Initially, three different pre-trained models are trained on the ISIC2019 dataset; eventually, ResNet50 has selected a final solution. The additional images are supplemented to the dataset because the dataset was imbalanced and which leads to biases of the model towards the class with more images. After addition, the further samples dataset contains five thousand samples each. Lastly, the results are compared with other existing studies on skin tumor detection using deep learning. Our results surpassed the existing models by performing well in training and testing time.