Identification and Classification of Melanoma Using Deep Learning Algorithm

Abstract

Melanoma also referred to as the most dangerous skin cancer, grow in the melanocytes that generate the pigment melanin. These cancers spread to different parts of the human skin in different ways and gradually lead to death. So, this cancer needs to be detected soon and easily. People often ignore skin problems and do not want to take complex medical tests. The aim of this article is to classify melanoma and non-melanoma by applying a deep learning-based model through dermoscopic images from a lesion dataset of Kaggle. Transfer learning with inception v3 model is utilized for melanoma classification. Convolutional neural networking (CNN) helps to achieve an encouraging result. A dataset from Kaggle containing total 2,750 dermoscopic images was being used. It consisted of three classes. The test accuracy is 77.25% in Jupyter notebook and 96.19% in colab google drive was achieved. The model can distinguish melanoma and non-melanoma (nevus and seborrheic keratosis) with satisfactory level of prediction. People need to take any dermoscopic image of their lesion and the get result can be generated within 5 minutes. Thus, the system can be used in a simple way to detect and distinguish between melanoma and non-melanoma.