Skin Cancer Detection using VGG16, InceptionV3 and ResUNet

The rapid spread of skin diseases can be attributed to global warming and the harmful effects of ultraviolet rays from the sun. If left untreated, skin diseases can pose a significant threat to human life. Melanoma, a fatal form of skin cancer, is particularly challenging to diagnose due to its complex nature. Early and accurate detection of melanoma is critical for saving human lives. While dermatologists typically rely on examining color images of skin lesions to diagnose melanoma, this process can be error-prone. Therefore, computer vision-based methods have been proposed as an alternative, with the potential to provide more accurate diagnoses. In this study, a dataset consisting of 10,500 skin cancer images from the International Skin Imaging Collaboration (ISIC) was created by selecting images from 7 different skin cancer categories. Three convolutional neural network (CNN) models, namely VGG16, ResUNet and InceptionV3 were trained on this dataset to classify skin cancer. To improve classification outcomes, transfer learning is utilized on two specific pre-trained models, namely VGG16 and InceptionV3. In recent years, the popularity of utilizing machine learning models for diagnosing skin cancer has risen significantly, owing to its ability to decrease diagnostic errors and enhance patient outcomes. The ensemble of CNN architectures used in this study represents a promising approach to skin cancer diagnosis, as it allows for the combination of multiple models to improve overall accuracy. This particular combination of models has shown better performance after a lot of other combinations were trained. Based on the experimental outcomes, it is concluded that the suggested model exhibits commendable performance in precisely categorizing skin cancer, and the achieved results are satisfactory in comparison to other contemporary techniques.