Explainable AI (XAI) driven skin cancer detection using transformer and CNN based architecture

Abstract

Skin Cancer is a cancer form that has become very prevalent in recent times and, if left untreated, has the potential to cause premature death. That is why early diagnosis and treatment are important to cure this disease. For this, we can use Machine Learning based methods to effectively impact the identification and categorization of skin cancer. Previously it was seen that the CNN models had a notable impact on the performance of the classification tasks. However, Vision transformers (VIT) are also the solution chosen by the researchers which have displayed significant performance in classification works. To make the outcomes of diverse data as distinct as feasible, contrastive learning is utilized to make similar skin cancer data for encoding similarly. The categorization of skin cancer depending upon multimodal data is made possible by the transformer network’s exceptional performance in natural language processing and field of vision. In this paper, we have offered a detailed analysis of VGG-16, a CNN architecture, and ViT, a transformer-based method to classify skin lesion images for aiding the early diagnosis of skin cancer. The findings indicate that the VGG-16 model attained an accuracy of 82.14%, whereas the Vision Transformer achieved a slightly lower accuracy of 76.15%. A modified version of the original vision transformer, the shifted patch tokenization, and locality self-attention modified Vision transformer showed an accuracy of 74.55% with expectations for further improvement in the future. Moreover, nowadays people have to choose a model from several other models to solve an issue, and as the model keeps on improving, it becomes very difficult to understand how the model works internally. So, for this reason, Explainable Artificial Intelligence (XAI) is introduced to give an idea of a human-readable explanation for the decision-making process of a model. This will certainly benefit cosmetologists, health researchers, research scientists, and researchers working in various areas and offer patients more convenience.