Please login or create an account. If you do not have access to this content, you will be shown a 30 second preview and licensing options.

Presentation

Artificial Intelligence in Dermatology: Diagnostics, Prognostication, and Workflow Improvement

Description

The talk reviewed the evolution and current use of artificial intelligence in dermatology, emphasizing that while AI is not new, recent advances in hardware, software, and foundation models have accelerated its clinical relevance. The speaker explained that dermoscopy and digital photography were early “low-hanging fruit” for convolutional neural networks because of their standardized inputs, and studies have shown AI can match or sometimes exceed dermatologist performance, especially for melanocytic and keratinocytic lesion classification. However, combining AI with dermatologist review has produced only modest gains in some studies. The talk also highlighted approved AI devices for triaging skin lesions, noting that they are designed for high sensitivity but often have limited specificity, leading to false positives. Beyond diagnosis, AI is increasingly used in total body photography, teledermatology, and multimodal foundation models that integrate imaging, histopathology, and clinical data; these models show promise but still leave room for improvement across broader skin disease categories. The speaker described research from his lab using machine learning to improve melanoma prognostication beyond AJCC staging by combining electronic health record, imaging, and histopathology data, including features such as mitotic rate, to better predict recurrence, timing, recurrence type, and survival. Finally, he argued that the most immediate impact of AI will likely be in operations management, such as ambient scribing, clinical intake, scheduling, summarization, and patient communication, which can reduce documentation burden, improve workflow, and lower administrative costs.

View more

Conclusions

AI is becoming highly effective for dermatology image analysis, especially in standardized tasks like dermoscopy and certain digital photography applications.
Dermatology AI can perform comparably to, and sometimes better than, dermatologists in lesion classification, though expert human performance remains important.
Combining AI with dermatologist review can improve diagnostic performance, but the gains are often modest rather than transformative.
Performance is strongest when image inputs are standardized, while heterogeneous real-world photos remain a major challenge for deployment.
Commercial AI devices in dermatology are already available and are designed to prioritize sensitivity and triage, even if specificity is limited.
Total body photography is an important and growing AI-enabled tool for melanoma surveillance and is expanding into inflammatory disease monitoring and teledermatology.
Newer total-body imaging platforms are reducing space and cost barriers while preserving useful lesion-detection quality.
Foundation models trained on large, multimodal dermatology datasets can outperform general-purpose models and better support downstream dermatologic tasks.
Multimodal AI systems can integrate clinical, imaging, and pathology data to improve melanoma prognostication beyond staging alone.
In melanoma recurrence prediction, machine learning models can outperform AJCC staging and meaningfully stratify patients into low-, intermediate-, and high-risk groups.
Features such as mitotic rate remain highly predictive and may be undervalued in traditional staging systems.
AI is likely to have its most immediate practical impact in operations management, where it can reduce documentation burden, improve scheduling, and streamline patient communication.
Ambient scribing and related workflow tools can save clinician time and reduce after-hours work without sacrificing care quality.
Overall, the near-term value of AI in dermatology appears to be greatest across diagnostics, prognostication, and productivity rather than as a full replacement for clinicians.

Li Z, et al, Journal of Clinical Medicine Research 11 (22).
Codella N, et al., ISIC Competition 2018.
Pham et al., Scientific Reports 2021.
Brinker et al., European Journal of Cancer 2019.
Du-Harpur X, et al., Br J Dermatol. 2020 Sep;183(3):423-430.
Zhou, J., et al. Nat Commun 15, 5649 (2024).
Yan, S., et al. Nat Med 31, 2691–2702 (2025).
Wang G, et al. NPJ Precis Oncol. 2022 Oct 31;6(1):79.
Wang and colleagues in NPJ Precis Oncol. 2022 and J Am Acad Dermatol. 2024.
Clinician Experiences With Ambient Scribe Technology to Assist With Documentation Burden and Efficiency. JAMA Network Open.#10.1001/jamanetworkopen.2024.60637