Classification of skin cancers from skin lesion images

• Main Author: Wee, Hui Ning
• Other Authors: Rajapakse Jagath Chandana
• Format: Final Year Project
• Language: English
• Published: 2017
• Subjects: DRNTU::Engineering::Computer science and engineering::Computing methodologies::Image processing and computer vision
• Online Access: http://hdl.handle.net/10356/72867
• Institution: Nanyang Technological University

The objective of this Final Year Project (FYP) is to develop a deep neural network that can identify a benign or malignant skin lesion, and to be able to identify the type of skin cancer a malignant lesion falls into, from the skin lesion images. Convolutional neural networks are a subset of feedforward neural networks, and can model many different types of datasets. It is also possible to finetune the parameters of the network for optimization such that they are able to better generalize the model to suit the required dataset. The deep learning framework Caffe is a popular choice in developing such neural networks, and allows for the utilization of a large number of related libraries such as OpenCV and CUDA, while also supporting C++ and Python. The Caffe framework is used in the development and optimization of the network in this project. In this project, we use skin lesion images from the ASCI database and convolutional neural networks (CNNs) to classify the skin images into benign or malignant cancers. 65% of accuracy was achieved in classifying these images into cancerous and non- cancerous types.