Skin cancer image classification using convolution neural network with transfer learning based on VGG16

Abstract

Skin cancer is a very dangerous type of cancer and is still one of the most common causes of death around the world. It is important to find and correctly diagnose things early to raise survival rates, but old ways of doing things aren't always reliable. Recent advancements in deep learning have significantly improved the automation of skin lesion evaluations, bringing positive outcomes in the work of dermatologist. A skin lesion is an abnormal area of skin where malignant tumors grow out of control, which is different from benign tumors. This research developed a Convolutional Neural Network (CNN) model for skin cancer detection, utilizing VGG16 as the foundational architecture via transfer learning. The research utilized the ISIC2018 dataset to categorize two primary tumor types malignant and benign while also incorporating a variety of seven different skin cancer types from the HAM10000 dataset for a more comprehensive examination.The CNN model performed exceptionally well, achieving a training accuracy of 96.46%, a training loss of just 0.092%, a precision of 0.90%, a recall of 0.91%, and an F1-score of 0.90%. These findings go beyond previous work in the area, showcasing the increasing power of deep learning methods in skin cancer detection. They offer more efficient and simplified models that can handle a variety of data types. Although deep learning continues to make rapid strides, there is still a need for further research to address existing challenges and discover new approaches to automating the detection of skin cancer.