Skin Checks for People of Colour: Why Fitzpatrick Diversity Matters in AI and Dermatology

The Diagnostic Gap: Skin Cancer in People of Colour

Skin cancer doesn't discriminate by race, but diagnostic tools and clinical training historically have. The majority of dermatology textbooks, training datasets, and clinical images feature lighter skin tones, creating a systematic gap in recognising conditions on darker skin.

The consequences are measurable and stark:

Understanding the Fitzpatrick Scale

The Fitzpatrick Skin Type classification, developed by dermatologist Thomas Fitzpatrick in 1975, remains the most widely used framework for categorising skin phototypes. It classifies skin into six types based on response to ultraviolet radiation:

Limitations of the Fitzpatrick Scale

While widely adopted, the Fitzpatrick scale has significant limitations. It was originally designed to measure UV sensitivity in white patients and doesn't fully capture the diversity of darker skin tones. Researchers are developing more inclusive alternatives, including the Individual Typology Angle (ITA) and the Monk Skin Tone Scale (10-point scale), which provide finer granularity for Fitzpatrick Types IV–VI.

How Skin Conditions Present Differently on Darker Skin

Many skin conditions present with fundamentally different visual characteristics on darker skin tones, making recognition harder for clinicians and AI systems trained primarily on lighter skin:

AI Skin Analysis and the Diversity Problem

The accuracy of AI in dermatology is fundamentally shaped by its training data. Historically, most dermatological image datasets have been drawn from populations with lighter skin tones (Fitzpatrick I–III), creating a measurable performance gap:

• Training data imbalance: Studies have shown that less than 5% of images in major dermatology AI training datasets represented Fitzpatrick Type V–VI skin prior to 2023.
• Performance disparities: A 2022 Stanford study found that AI dermatology models had up to 20% lower accuracy on darker skin tones compared to lighter tones.
• Condition-specific gaps: AI performs worst on conditions that present with subtle colour differences on darker skin—exactly the cases where AI assistance is most needed.

What's Changing in 2024–2026

The AI dermatology community has responded to these documented disparities with concrete action:

• Diversified datasets: Major initiatives (ISIC 2024, DDI dataset, Fitzpatrick17k) now prioritise Fitzpatrick IV–VI representation, with some achieving near-equal distribution across skin types.
• Fairness auditing: Leading platforms now routinely measure and report accuracy stratified by skin type, with minimum performance thresholds across all Fitzpatrick categories.
• Domain adaptation: Transfer learning techniques allow models trained on larger, lighter-skin datasets to be fine-tuned with smaller, diverse datasets—achieving near-parity performance.
• Community-driven data: Platforms like ScanSkinAI leverage global user bases to continuously expand representation from underserved populations.

For the latest evidence on AI accuracy, see our comprehensive 2026 evidence review on AI skin cancer detection.

Protecting Your Skin Health: Practical Guidance

For People of Colour

• Know your risk areas: Check palms, soles, nail beds, and mucous membranes regularly—these are the most common sites for melanoma in darker skin tones.
• Use AI screening: Regular AI screening can help establish a baseline and detect changes over time. Use our AI Skin Condition Checker for instant analysis.
• Look for the "Hutchinson sign": Dark pigment extending from a nail bed onto surrounding skin is a red flag for subungual melanoma—seek immediate professional evaluation.
• Don't skip sunscreen: Melanin provides some UV protection, but not complete protection. Daily SPF 30+ is recommended for all skin types.
• Seek diverse dermatologists: When possible, choose dermatologists experienced with diverse skin tones. Board-certified dermatologists with training in skin of colour are increasingly available.
• Photograph changes: Document any new or changing spots with well-lit photos. Our photo guide provides techniques optimised for all skin tones.

For Healthcare Providers and Employers

• Audit your AI tools: Before deploying AI skin screening, request performance data stratified by Fitzpatrick type. Reject tools that cannot demonstrate equitable accuracy.
• Diversify training: Support and fund diverse dermatological image collection initiatives.
• Educate on diverse presentations: Ensure clinical staff can recognise conditions on all skin tones, not just the textbook presentations.
• Reduce access barriers: Offer AI skin screening as an employee benefit to democratise access. Learn more about AI screening as an employee benefit.

The Role of AI in Closing the Equity Gap

Paradoxically, AI—which has contributed to diagnostic disparities through biased training data—also holds the greatest potential to close the equity gap:

• Accessibility: AI screening removes barriers of cost, location, and insurance that disproportionately affect communities of colour.
• Consistency: Unlike human clinicians who may have unconscious biases, well-trained AI applies consistent analysis regardless of patient demographics.
• Scalability: AI can screen millions of people at zero marginal cost, enabling population-level skin cancer prevention in underserved communities.
• Continuous improvement: As training datasets become more diverse, AI accuracy on darker skin tones improves—a self-correcting cycle that outpaces textbook revision timelines.

Frequently Asked Questions

Related Articles