AI in Dermatology: Augmenting Diagnosis, Prognosis, and Clinical Workflow

Conclusions

• AI is becoming routine in dermatology, especially for image analysis, and its use is rising quickly among clinicians.
• For standardized skin imaging tasks like dermoscopy and total body photography, AI can match or sometimes exceed average dermatologist performance, particularly when image quality and capture conditions are controlled.
• AI works best as a clinician-in-the-loop tool, where it supports rather than replaces dermatologists and can modestly improve diagnostic accuracy and decision-making.
• Skin-specific foundation models trained on dermatology data outperform general-purpose models, especially for focused tasks such as melanoma detection and explanation.
• Multi-modal models that combine imaging, pathology, and clinical/EHR data perform better than single-modality approaches for many dermatology problems.
• For melanoma recurrence prediction, AI models using EHR and histopathology data outperform AJCC staging alone and can identify low-, intermediate-, and high-risk groups more effectively.
• The strongest prognostic systems are those that integrate clinical, imaging, and pathology information, since each modality contributes independent predictive value.
• AI can support tumor board-style reasoning and treatment selection with high agreement to expert oncologist recommendations, suggesting a path to more standardized and democratized expert guidance.
• The most immediate and widespread benefits of AI in practice come from operations management, especially ambient scribing, patient intake, scheduling, and clinical text generation.
• AI tools can meaningfully reduce documentation burden, after-hours work, and administrative costs, which may also help reduce clinician burnout and improve workflow efficiency.