Melanoma Detection using Convolutional Neural Networks

Abstract

In humans, most severe and common type cancer is skin cancer. Skin cancers are basically 3 types: basal cell carcinoma (BCC), squamous cell carcinoma (SCC) and Melanoma. Among these Melanoma is dangerous skin cancer. Melanoma is classified as two types: Benign Melanoma and Malignant Melanoma. If Melanoma can be identified in early stages it can be cured easily. The conventional method for detecting Melanoma is very painful. In this study deep learning techniques like CNN is used to detect Melanoma. CNN consists of convolutional layers, pooling layers and fully connected layers. Both training and testing of images can be done using CNN. ISIC Archive 2017 dataset is given to the network. By comparing different number of epochs and batch size, accuracy is noted. Highest accuracy 88.89% is achieved at 45 epoch count and batch size 2.