SkinGuardian: Skin Health Management System

Abstract

The rate of skin cancer and non-cancerous skin disorders is rising globally, and this necessitates the development of more efficient and easily available diagnostic techniques. Traditional diagnostic techniques frequently fall short in terms of precision, usability, and quick response, which causes treatment to be delayed and healthcare costs to rise. Convolutional Neural Networks (CNNs) which is an advanced machine learning approach, is employed in the SkinGuardian: Skin Health Management System project to create a web-based application that improves the precision and effectiveness of skin-related issue diagnosis and treatment. This novel strategy provides a complete solution for early detection and continuous patient care by combining the creation of custom datasets, model training, and application development. The goal of the research is to use machine learning and convolutional neural networks (CNNs) to improve the diagnostic accuracy for skin disorders. To ensure specificity and reliability, a three model architecture was created: one to distinguish between cancerous and non-cancerous skin condition, one for skin cancer, and one for non-cancerous skin diseases which includes vitiligo too, where a custom dataset was created for vitiligo Since it was unrepresented in Asian countries like Sri Lanka. The accuracies of the three models were gradually increased when tested with several layer structures, activation functions, and optimization techniques. During the testing stage, the models demonstrated great performance. The accuracy of the model used to differentiate between non-cancerous and cancerous conditions was 94%, and the model used to identify skin cancer had a accuracy of 84% and an accuracy of 96% for model which classifies non-cancerous skin conditions, including vitiligo, highlighting the importance of the customized dataset for model performance.