Can AI detect certain diseases by looking at images of skin ?

AI can already detect certain skin diseases from images with performance that matches or exceeds dermatologists in controlled studies, especially for common conditions like melanoma, psoriasis, and eczema. Deep convolutional neural networks trained on large datasets of labeled clinical and smartphone-captured images achieve high sensitivity and specificity, and several regulatory-cleared tools are available for use by healthcare professionals. However, real-world accuracy can vary with image quality, skin tone, lighting, and rare or atypical presentations, requiring clinician oversight. Ongoing research focuses on improving generalization across diverse populations and integrating multimodal data such as dermoscopy and patient history.

— Enriched May 13, 2026 · Source: World Health Organization — https://www.who.int/publications/i/item/9789240150842

AI can already assist physicians by screening images of skin lesions with deep-learning models that approach dermatologist-level performance on curated datasets. For example, convolutional neural networks have achieved sensitivities above 90% and specificities above 80% for detecting melanoma in dermoscopic images under experimental conditions. However, real-world deployment faces challenges such as variable lighting, lower-resolution cameras, and skin-tone bias that can degrade accuracy. Generalization across devices, populations, and clinical settings remains an active research area.

— Enriched May 13, 2026 · Source: Nature Medicine — https://www.nature.com/articles/s41591-020-0942-0