Deep Learning for Identification of Skin Diseases: Emphasizing on Arsenic-Induced Skin Conditions

Skin diseases such as arsenic-induced lesions, melanoma, psoriasis, and eczema present significant diagnostic challenges due to their visually similar characteristics. Accurate classification is critical, especially for arsenic-related conditions, which are often difficult to differentiate from other skin diseases. This study aims to develop a deep learning-based classification model that effectively identifies arsenic-related skin conditions alongside other common skin diseases. Using a custom dataset, we evaluated the performance of pre-trained models, including VGG16, MobileNetV2, InceptionV3, and ConvNeXtBase. The results show that while VGG16 and InceptionV3 achieved competitive accuracies of 85% and 82%, respectively, our proposed model (ARS-CNNSA) significantly outperformed them, achieving 91% accuracy and 90% F1-score. The model’s architecture integrates Conv2D layers, and self-attention mechanisms which enhance its ability to capture complex features and avoid overfitting. These findings suggest that our approach can improve diagnostic accuracy, particularly in regions with high arsenic exposure, and may serve as a foundation for future AI-driven tools in dermatology by using less amount of hyperparameters of the deep learning model.