Skin Disease Diagnosis: A Comprehensive Survey of Models, Datasets, and Clinical Applications

The early and accurate diagnosis of skin diseases, particularly malignant melanoma, remains a critical challenge in dermatology. Dermoscopy, a non-invasive imaging technique, significantly enhances visual inspection but demands expert interpretation, often limited by interobserver variability. The advent of machine learning (ML) and deep learning (DL) has revolutionized automated dermoscopic analysis by enabling scalable, objective, and high-precision diagnostic pipelines. This survey comprehensively reviews the evolution of AI-driven dermatological diagnostics, spanning classical ML classifiers, convolutional neural networks, transformer based models, and hybrid architectures. We systematically examine widely adopted dermoscopic image datasets such as ISIC, HAM10000, and PH2, highlighting their structure, scope, and impact on model development. In addition, the paper explores critical preprocessing and feature engineering strategies, including image enhancement, lesion segmentation, and multimodal fusion. Evaluation protocols encompassing accuracy, sensitivity, specificity, and area under the receiver operating characteristic curve are discussed alongside comparative results across architectures and tasks. The practical integration of AI in clinical settings is investigated, supported by real-world applications and interpretability techniques like Gradient-Weighted Class Activation Mapping and Local Interpretable Model-agnostic Explanations to foster clinical trust. Key limitations, such as dataset imbalance, generalization to diverse populations, and lack of transparency, are analyzed under open challenges. Finally, future research trajectories including federated learning, lightweight deployment ready models, and multimodal diagnostic systems are outlined. This survey serves as a foundational reference for advancing AI-assisted dermoscopic diagnostics, bridging algorithmic development and clinical translation.