A Transfer Learning-based Pre-trained VGG16 Model for Skin Disease Classification

Abstract

Skin disorders pose a significant global health risk, impacting millions of individuals and placing a substantial burden on healthcare systems. The accuracy and speed of diagnosis are crucial for effectively managing various conditions. Deep learning models have demonstrated exceptional performance in diverse medical imaging applications, including the categorization of skin diseases. In recent years, the VGG16 deep learning architecture has gained prominence for its ability to extract meaningful features from images. In this study, a VGG16 model has been leveraged to early diagnose skin diseases. This approach involves collecting an extensive dataset comprising images of different skin disorders sourced from an open-source repository "Kaggle". Further, the VGG16 model is then fine-tuned on this collected dataset to learn the distinguishing patterns and characteristics associated with different skin conditions. The evaluation of the model's effectiveness has been done using standard metrics such as precision, recall, F1-score, and accuracy. These metrics assess the model's analytical capabilities in distinguishing between various skin disorders. The proposed deep learning model achieves remarkable accuracy of 90.1%, proving its proficiency in diagnosing a wide range of skin diseases, including those that appear similar. Furthermore, precision, recall, and F1-score have been identified as 0.867, 0.942, and 0.891, respectively. This research contributes to the evolution of computer-aided disease detection, potentially leading to enhanced healthcare outcomes by facilitating early detection and treatment of skin disorders. Nonetheless, continuous refinements and validation on larger, more diverse datasets are imperative to further enhance the model's accuracy and ability to generalize across various conditions.

Topics & Concepts