Melanoma Detection Using Convolutional Neural Networks and Group Normalization

Melanoma is one of the most dangerous types of skin cancers. It is very difficult to detect even for experienced dermatologists. Thus, there is a need to develop artificial intelligence-based detection systems to aid in the early diagnosis and treatment of melanoma. The paper proposes a Convolutional Neural Network (CNN) based approach to classify benign and malignant melanoma. The proposed system includes a web application that can be used by the end user. The CNN is trained using a large number of benign and malignant melanoma images. The proposed CNN architecture uses data augmentation and Group Normalization techniques for improved results. The proposed model produces an accuracy of 89.82% with a sensitivity of 92.20% and a specificity of 87.27%. The proposed model is compared to two other experimental architectures and is found to produce the best results. The proposed architecture is also compared to pre-built CNN architectures like ResNet50, MobileNet and InceptionV3 which have been trained on the same dataset. The proposed model outperforms the pre-built architectures and produces appreciably better results.