Skin disease detection and diagnosis using CNN: a systematic review

This review (2020–2025) examines recent progress in Convolutional Neural Networks (CNNs) for detecting and diagnosing skin diseases. Given the increasing need for accessible dermatological services and the limited number of specialists, CNNs have become effective tools for automating diagnoses through image analysis. They have shown expert-level accuracy in recognizing conditions such as melanoma, psoriasis, and eczema, utilizing datasets like ISIC and HAM10000. Notable architectures such as ResNet, MobileNet, and VGGNet, in addition to custom models, have achieved impressive results.Significant innovations include transfer learning, ensemble strategies, multistage systems, and enhanced data augmentation techniques to tackle class imbalance and the scarcity of annotated data. Lightweight CNN designs tailored for mobile and edge devices facilitate real-time applications at the point of care, increasing accessibility in underprivileged areas. Nevertheless, challenges persist in managing variability in skin tones, enhancing the interpretability of models, and ensuring compliance with ethical and legal standards in clinical applications. By merging technical advancements with social and ethical concerns, this review emphasizes both the potential and the obligations of utilizing CNNs in dermatology. It offers an outlook on how CNN-based solutions can foster early detection, boost diagnostic efficiency, and ultimately close the gaps in global healthcare delivery.