MELANOMA DETECTION BASED ON DEEP LEARNING NETWORKS

Our main objective is to develop a method for identifying melanoma enabling accurate assessments of patient’s health. Skin cancer, such as melanoma can be extremely dangerous if not detected and treated early. Detecting skin cancer accurately and promptly can greatly increase the chances of survival. To achieve this, it is important to develop a computer-aided diagnostic support system. In this study a research team introduces a sophisticated transfer learning model that utilizes Resnet50 to classify melanoma. Transfer learning is a machine learning technique that takes advantage of trained models, for similar tasks resulting in time saving and enhanced accuracy by avoiding the need to train from scratch. The Resnet50 is a type of network that can distinguish between cancerous skin lesions in each sample. To evaluate its performance, we used data from the melanoma cancer dataset. However, the dataset has a percentage of samples which creates an imbalance between the classes. We addressed this issue by making the dataset more diverse through data augmentation techniques. In our project we implemented the Resnet50 pretrained model with learning rates and weight decay. This model consists of 50 layers organized into blocks that include batch normalization and skip connections (known as connections). We adjusted the depth of the model to improve its accuracy. Our experimental results demonstrate that our proposed deep learning technique performs better in terms of accuracy compared to state of the art algorithms in this field. iiiThe model achieves an accuracy of 91.70%, with a learning rate of 0.0001 and a model depth of 34. By tuning hyperparameters using RESNET 50 we can further enhance the accuracy of our trained models.