Medical image segmentation using deep learning

In this thesis, first of all, detailed information about the deep learning approach which is attracted much attention in the field of machine learning in recent years is given. Then, an automatic application about skin lesion segmentation is performed to assist in the detection of melanoma, the most dangerous type of skin cancer. This medical image segmentation application is implemented by using Fully Convolutional Network (FCN) architectures obtained by modifying deep learning-based Convolutional Neural Network (CNN) architectures. ISIC 2017 is used as the dataset and four different FCN architectures named FCN-AlexNet, FCN-8s, FCN-16s and FCN-32s are used in the experimental studies performed for the application. Considering these architectures and dataset, this study is carried out for the first time in the literature. For the experimental studies, FCNs are first trained separately and accuracies on the validation dataset and Dice coefficients of these trained network models are compared. Besides, lesion segmentation inferences are visualized to take account of how precisely FCN architectures can segment lesions. The experimental results obtained show that FCNs are suitable for skin lesion segmentation. In addition, it is thought that the experimental results will contribute to the scientific literature and the researchers who are working on medical image segmentation. Keywords: Deep learning, Convolutional Neural Network, Fully Convolutional Network, Medical image segmentation.