Melanoma Malignancy Prognosis Using Deep Transfer Learning

Abstract

Melanoma is a type of Skin cancer that spreads rapidly and has a significant death risk if it is not detected early and treated. A prompt and accurate diagnosis can improve the patient’s chances of survival. The creation of a skin cancer diagnostic support system based on computer technologies is highly essential. This study suggests a unique deep transfer learning model for categorizing melanoma malignancy. The proposed system comprises of three main phases including image preprocessing, feature extraction and melanoma classification. The preprocessing phase employs image filters such as mean, median, gaussian and non-local means filter along with histogram equalization techniques to obtain the preprocessed images. Feature extraction and classification are performed using pre-trained Convolutional Neural Network architectures such as DenseNet121, Inception-Resnet-V2 and Xception. Using the ISIC 2020 dataset, the suggested deep learning model’s effectiveness is assessed. The experimental findings show that, in terms of precision and computational expense, the suggested deep transfer learning model performs better than cutting-edge deep learning algorithms.