An Intelligent Deep Learning Framework for Accurate Skin Cancer Detection Using CNN, ResNet, and GoogLeNet Models

Abstract

Skin cancer is one of the most common diseases as well as one of the most threatening for patients' lives; thus, early and accurate detection of skin cancer becomes vital for successful treatment. The current paper studies the use of deep learning models (Convolutional Neural Networks, ResNet, and GoogleNet) in skin cancer classification using dermoscopic images. These deep learning models were chosen because of the possibility of learning hierarchy and distinguishing small differences in images of benign and malignant lesions. Transfer learning was used to speed up training process and improve generalization using pretrained weights for images. The experiment showed that the most accurate model reached the following metrics: 96.8% of accuracy, 95.5% of precision, 96.2% of recall, and 95.8% of F1-score working with a publicly available skin cancer dataset. The findings demonstrate the high efficiency of deep learning when combined with dermoscopic image analysis and transfer learning.