Dermatology AI is transitioning from a phase of isolated computer vision benchmarks to a complex, multimodal paradigm that prioritizes clinical workflow integration, demographic equity, and administrative automation. While early diagnostic algorithms focused strictly on pixel-level classification of dermoscopic images, current clinical realities demand tools that synthesize patient metadata, address systemic demographic biases, and reduce the administrative burden on overstretched clinicians.
This shift is driven by the realization that standalone image classifiers, no matter how accurate in retrospective reader studies, fail to account for the nuances of real-world clinical practice. By moving toward multimodal architectures and robust administrative tools, the industry is laying the groundwork for sustainable, equitable AI deployment in both specialized and primary care settings.
Notable papers
• Dermoscopy-Based AI Risk Scoring Enhanced Experienced Dermatologists’ Decision-Making: a Large Retrospective Reader Study
This large retrospective reader study demonstrated that a deep learning-generated AI risk score, based on dermoscopic images and metadata, significantly improved experienced dermatologists’ diagnostic accuracy and management strategies for suspicious skin lesions.
Brand take: Genuinely useful clinically, as it demonstrates how AI acts as an effective decision-support tool rather than an autonomous replacement.
• Interpretable multimodal fusion for skin lesion classification using dermoscopic images and patient metadata
This paper presents an interpretable multimodal fusion model that combines dermoscopic images with clinical patient metadata, successfully addressing severe class imbalances and high inter-class visual similarities.
Brand take: Genuinely useful clinically, proving that image-only models are no longer sufficient for real-world dermatology diagnostics.
• Patient Acceptability, Perceptions and Concerns Regarding Artificial Intelligence Scribes in Outpatient Dermatology Clinics
A cross-sectional survey revealed that 82.1% of outpatient dermatology patients reported comfort with ambient AI scribes, though over half still expressed privacy and data security concerns.
Brand take: Underrated, as administrative AI will likely achieve widespread clinical adoption far faster than diagnostic AI.
• AI‐assisted diagnosis of nail unit melanoma and melanonychia using a clinical deep learning model
This study validated a deep learning model designed to distinguish rare, potentially fatal nail unit melanoma from benign melanonychia using clinical images, offering a noninvasive alternative to high-risk biopsies.
Brand take: Genuinely useful clinically, targeting a high-stakes, niche diagnostic dilemma where biopsy-induced nail dystrophy is a major concern.
• 0075 Diagnostic accuracy of artificial intelligence tools for psoriasis recognition in skin of color: A cross-model evaluation
A cross-model evaluation revealed significant performance disparities in psoriasis recognition when testing established AI models on skin of color, highlighting systemic training gaps.
Brand take: Genuinely useful clinically, highlighting the urgent need for diverse training datasets to prevent diagnostic inequities.
• AI without representation is just inequity at scale: on the exportation of unrepresentative artificial intelligence models to the Global South
This paper argues that simply increasing demographic representation in training data is insufficient to mitigate bias when exporting unrepresentative models internationally.
Brand take: Underrated, as it challenges the simplistic ‘more data’ narrative and demands structural changes in model deployment.
• 0277 A dermatology-specific model card framework for transparent and responsible clinical deployment of artificial intelligence (AI) dermatology tools
This study proposes a standardized model card framework tailored specifically to dermatology to ensure transparent, responsible, and standardized clinical deployment of AI tools.
Brand take: Genuinely useful clinically, providing a practical blueprint for clinical AI product teams to document model limitations.
Products, deals & funding
• Procode AI Series A Funding
Procode AI secured $10 million in Series A funding to expand its AI-powered revenue cycle management platform, which currently serves over 350 plastic surgery and dermatology providers.
Brand take: Genuinely useful clinically, as administrative automation addresses immediate operational bottlenecks and clinician burnout far more effectively than diagnostic algorithms.
• Yesilscience – New Skin Cancer AI Beats Nineteen Dermatologists
A new multi-endpoint AI framework outperformed 19 specialists in diagnosing skin cancer, though its hidden failure modes emphasize the ongoing risks of replacement-level deployment.
Brand take: Overhyped, because outperforming clinicians in controlled reader studies rarely translates directly to safe, autonomous real-world performance.
• Yesilscience – AI reduces gender bias in skin cancer diagnosis
An algorithmic framework that forces models to focus on actual skin lesions instead of demographic noise successfully reduced gender bias in skin cancer diagnosis.
Brand take: Underrated, as removing non-clinical demographic cues is a highly practical way to engineer equity into computer vision.
• Yesilscience – AI cuts unnecessary tests for rare skin cancer
A deep learning tool designed to distinguish early-stage mycosis fungoides from benign eczema mimics can prevent defensive, expensive laboratory testing.
Brand take: Genuinely useful clinically, as it directly targets defensive medicine and high-cost diagnostic pathways.
Regulatory & clinical adoption
• DermaSensor Clinical Review
A clinical review analyzed the clinical evidence and regulatory implications of DermaSensor, the first FDA-authorized AI-enabled device for skin cancer detection by non-specialists.
Brand take: Genuinely useful clinically, as it shifts the triage bottleneck away from specialized clinics and empowers primary care providers.
• SkinRAI Consortium Launch
The British Association of Dermatologists (BAD) launched the Skin and Responsible Artificial Intelligence (SkinRAI) Consortium to independently evaluate the safety, effectiveness, and equity of dermatology AI tools before NHS-wide adoption.
Brand take: Underrated, as independent, pre-adoption clinical validation is the only way to prevent algorithmic drift in public healthcare systems.
• WHX Bangkok 2026 Conference
The conference featured dedicated sessions exploring the integration of AI dermatology, advanced imaging, and genetic testing into aesthetic clinical practice.
Brand take: Overhyped, as the marketing of AI in aesthetics currently outpaces the underlying clinical evidence.
Trends & what to watch
Dermatology AI is moving away from the ‘AI vs. Dermatologist’ paradigm and toward collaborative, multimodal clinical decision support. The transition is driven by the realization that image-only models are highly susceptible to demographic bias and lack the context required for complex clinical decision-making. Over the next 1 to 3 months, expect to see clinical AI product teams prioritize multimodal architectures that fuse clinical photography with structured electronic health record data, such as patient history and lesion location.
Concurrently, administrative AI tools—specifically ambient scribes and revenue cycle management platforms—are experiencing rapid real-world adoption. While diagnostic AI remains heavily scrutinized by regulatory bodies and consortia like SkinRAI, workflow automation tools face fewer regulatory hurdles and offer immediate financial returns for clinics. Clinicians and investors should monitor how these administrative tools manage patient privacy concerns, as early survey data indicates that over half of patients remain apprehensive despite high overall comfort levels.
Bottom line
The future of dermatology AI lies not in autonomous diagnostic replacement, but in multimodal decision support and administrative workflow automation that directly address clinician burnout and demographic bias.