Early melanoma skin cancer detection by processing dermoscopic images using traditional deep-learning models

Abstract

Skin cancer is one of the most dangerous health issues in the world. A huge number of people are affected every year. There are three types of skin cancer, and the most common one is melanoma, which is also the most dangerous skin cancer with a higher chance of metastasis and death. Roughly 1-2% of cases of skin cancer are caused by melanoma and affect 3-5% of the population. Current diagnostic methods are time-consuming and prone to human error, which requires more precise and effective detection methods. The goal of this project is to build a deep learning model architecture that can identify melanoma with high accuracy and low time consumption, utilizing the ISIC 2019-2020 Melanoma dataset. Our model aims to improve the accuracy of advanced image processing with deep learning techniques, which could facilitate early intervention and potentially life-saving treatments. The ISIC 2019-2020 datasets provide a huge quantity of dermoscopic images, which can serve as an excellent foundation for training and validating our deep-learning models. We use some recent famous and well-performed convolutional neural networks (CNNs) models and train the models, such as ResNet50, ConvNeXt, EfficientNetV2, VGG16, MobileNetV3, and Swin Transformer, to extract features indicative of melanoma skin cancer images and predict results accurately based on all these model’s prediction accuracy. We also created a web-based application to do the cancer prediction by comparing the results of all these 6th models to get an accurate result. This study shows the essential role of technology: to keep coming up with new ideas for advancing medical diagnostics to fight against the deadliest melanoma skin cancer. We tried to build up a deep-learning model that can detect exact skin cancer cells with high accuracy and low time consumption based on the ISIC 2019-2020 melanoma dataset.