AI in Dermatology: What It Can, and Can’t, Do for Patients

AI and Dermatology: Why I Am Both Enthusiastic and Careful

I have been publishing research on artificial intelligence in dermatology for over a decade, earlier than most clinicians in this field and well before AI became a mainstream healthcare conversation.

My background is unusual: undergraduate training at MIT in genomics, a master’s in public policy from Harvard Kennedy School with a focus on health technology, and a clinical career as a board-certified dermatologist and Mohs surgeon.

I have watched the AI-in-dermatology space develop from academic curiosity to commercial product to patient-facing tool. And I have opinions.

The short version: AI tools in dermatology are real. They have genuine clinical utility in specific contexts. They are also overhyped and misapplied in others.

Patients deserve an honest, nuanced explanation of what these tools can and cannot do. Not breathless startup language. Not blanket dismissal. Just a clear explanation of where the science actually is.

Where AI Has Demonstrated Real Clinical Value

AI in dermatology is not one thing. There is a major difference between a regulated clinical tool used by physicians, a research algorithm trained on high-quality image datasets, an AI scribe used for documentation, and a consumer app that analyzes a selfie.

Some AI tools are promising. Some are useful today. Some are not ready for medical decision-making.

Skin Lesion Analysis and Melanoma Detection

The most studied use of AI in dermatology is skin lesion classification, especially for melanoma detection.

A landmark 2017 paper in Nature showed that a convolutional neural network could classify skin lesions at a level comparable to board-certified dermatologists in specific image-based tasks. That study helped launch the modern conversation around AI and skin cancer detection.

Since then, multiple studies have shown that AI can perform very well when analyzing high-quality clinical or dermoscopic images under controlled conditions.

But this is the critical nuance: classifying a well-lit image in a research dataset is not the same as diagnosing a patient in real life.

A dermatologist does not only look at one image. We ask whether the spot is new or changing. We examine the rest of the skin. We consider personal and family history, immune status, medications, prior biopsies, sun exposure, skin tone, symptoms, and subtle clinical patterns. We may use dermoscopy. We may biopsy.

AI can assist with image interpretation. It cannot replace the full clinical context.

The Problem With Smartphone Photos

Performance can drop when AI systems are used on smartphone photos taken by patients at home.

Lighting varies. Focus varies. Angles vary. Skin tone capture varies. Scale is often unclear. The lesion may be partially obscured by hair, shadow, makeup, bleeding, or inflammation.

This matters because many consumer AI skin apps operate in exactly this real-world environment.

The performance of AI under controlled clinical conditions is not the same as its performance “in the wild.”

That is why I do not want patients relying on an app to decide whether a changing mole is safe.

Teledermatology and Triage

AI-assisted triage in teledermatology is one of the more promising applications.

In a triage setting, AI may help sort which lesions or rashes need urgent evaluation, which can wait, and which should be routed to the right type of clinician. This can be especially useful in areas where dermatologist access is limited.

The goal is not for AI to independently diagnose every patient. The goal is to help patients get to the right care faster.

That is a meaningful public health application, especially for rural communities, underserved areas, and health systems with long dermatology wait times.

Clinical Documentation and Workflow

One of the most immediately useful applications of AI in dermatology is not diagnosis. It is documentation.

AI scribing tools use ambient audio and natural language processing to help generate clinical notes from patient encounters. These tools can reduce the time physicians spend documenting and help return more attention to the patient in the room.

This may be less glamorous than AI melanoma detection, but it is extremely practical.

The documentation burden in medicine is enormous. If AI can safely reduce that burden while preserving accuracy, privacy, and clinician oversight, it can improve the patient experience.

That said, AI-generated notes still need physician review. The doctor is responsible for the medical record.

Where I Am Skeptical: Consumer AI Skin Analysis Apps

The app store is full of products that promise to analyze your skin, identify conditions, recommend products, and flag possible skin cancers from a smartphone photo.

As someone who has spent years in this research space, I want to be direct: the clinical validation supporting many consumer-facing AI skin apps is limited.

Some apps may be helpful for awareness. They may remind patients to check their skin. They may encourage someone to seek care. They may provide basic skincare education.

But they should not be used as a substitute for a dermatologist’s evaluation of a new, changing, bleeding, painful, itching, or non-healing lesion.

Several studies of consumer-facing apps have shown variable accuracy. Some tools miss concerning lesions. Others overcall benign lesions and increase unnecessary anxiety and healthcare visits.

Both problems matter. A false negative can delay care. A false positive can create panic and unnecessary procedures.

AI and Skin of Color

This is one of the most important issues in dermatology AI.

AI systems are only as good as the data used to train and test them. If training datasets overrepresent lighter skin tones, the algorithm may perform worse on darker skin.

This is not theoretical. Studies and reviews have found that dermatology AI models can have lower accuracy on darker skin tones and uncommon diseases when datasets are not diverse.

That matters because patients with skin of color are already more likely to have certain skin cancers diagnosed later. A poorly validated AI tool could worsen disparities if it gives false reassurance or fails to recognize disease in darker skin.

Good AI in dermatology must be tested across diverse skin tones, ages, image types, and real-world clinical settings. Equity cannot be an afterthought. It has to be built into the dataset, validation process, and deployment plan.

What Good AI in Dermatology Actually Looks Like

The most promising AI applications in dermatology share several features.

• They are trained on large, diverse, clinically validated datasets.
• They are tested across different skin tones and image conditions.
• They are evaluated prospectively, not only on retrospective image sets.
• They are transparent about limitations.
• They are designed to assist clinicians, not replace them.
• They have clear privacy protections.
• They are regulated appropriately when they are being used for medical decision-making.
• They improve care without increasing harm, anxiety, or unnecessary procedures.
• The best AI tools do not pretend to be doctors. They help doctors and patients make better decisions.

What AI Cannot Do

• AI cannot feel a lesion.
• AI cannot ask a patient why they are worried about one spot.
• AI cannot know that a mole looked different 6 months ago unless that history is provided.
• AI cannot perform a biopsy.
• AI cannot evaluate the full patient the way a clinician can.
• AI cannot replace a dermatologist’s judgment when something does not fit the pattern.

This is especially important in dermatology because context matters. A rash can look identical in two patients but mean completely different things depending on medication exposure, autoimmune history, infection risk, pregnancy status, or immune suppression.

Images are powerful, but they are not the whole diagnosis.

My Recommendation to Patients

Use AI skin tools for what they are good at: awareness, education, tracking, and prompting you to pay attention to your skin.

• Do not use them to rule out skin cancer.
• Do not let a “low-risk” app result stop you from seeing a dermatologist if a spot is new, changing, bleeding, painful, itching, growing, or not healing.
• Do not upload sensitive medical images without understanding the privacy policy.
• Do not assume an AI skincare recommendation understands your rosacea, eczema, acne medication, pregnancy status, allergies, or history of procedures.

If something concerns you, come in.

AI can support dermatology. It should not replace dermatology.

When to See a Dermatologist Instead of Using an App

You should see a dermatologist if you notice:

• A new or changing mole
• A bleeding or non-healing spot
• A dark streak in the nail that is new or changing
• A lesion that looks different from your other spots
• A rash that is painful, spreading, blistering, or associated with fever
• Acne that is scarring or not responding to over-the-counter treatment
• A skin concern in a child, pregnant patient, or immunocompromised patient
• Any spot that your instinct tells you is not right

An app should never talk you out of seeking care for something that worries you.

The Bottom Line

AI in dermatology is exciting, but it needs to be used carefully.

The strongest applications are clinician-supported tools, triage systems, image analysis under controlled conditions, workflow support, and documentation assistance. The weakest applications are overconfident consumer tools that promise diagnosis from a single smartphone photo without context.

The future of AI in dermatology should not be about replacing dermatologists. It should be about expanding access, improving triage, reducing administrative burden, and helping patients get to the right care sooner.

That is the AI future I believe in.

FAQ

Q: Can AI diagnose skin cancer?

A: AI can help analyze skin lesion images in certain settings, but it should not replace a dermatologist’s evaluation. Diagnosis requires clinical context, exam, and sometimes biopsy.

Q: Are AI skin cancer apps accurate?

A: Some apps show promise, but accuracy varies widely. Many consumer-facing tools have limited real-world validation and should not be used to rule out skin cancer.

Q: Can I use an AI app to check a mole?

A: You can use an app for tracking or awareness, but if a mole is new, changing, bleeding, painful, itching, or looks different from your other spots, see a dermatologist.

Q: Is AI dermatology safe for skin of color?

A: It depends on the tool. AI systems must be trained and tested on diverse skin tones. Tools that are not validated in darker skin may be less reliable.

Q: What is AI best used for in dermatology?

A: AI is most promising for clinical decision support, teledermatology triage, lesion tracking, documentation, workflow support, and helping improve access to care.

Q: Can AI recommend skincare?

A: AI can offer general skincare suggestions, but it may not understand your full medical history, skin condition, medications, allergies, pregnancy status, or procedure history.

Q: Will AI replace dermatologists?

A: No. AI may help dermatologists work more efficiently and improve access, but it cannot replace clinical judgment, physical examination, biopsy, or individualized medical care.

Q: What should I do if an AI app says my mole is low risk but I am still worried?

A: See a dermatologist. Your concern matters, and a changing or unusual lesion should be evaluated regardless of what an app says.

Updated June 2026