The 105-study review finds acne AI is focused almost entirely on diagnosis, yet rarely reports results by skin color. That gap leaves open equity questions for people using acne apps, especially those with darker skin. Acne AI means software that analyzes skin photos to identify acne or grade severity. The review screened PubMed, Cochrane and Scopus and included 105 acne-AI studies, according to the International Journal of Dermatology review.
Medical information disclaimer: This article is for general educational purposes only and does not provide medical advice, diagnosis, or treatment. Always consult a physician or other qualified health professional about symptoms, medications, tests, or treatment decisions.
Table of Contents
- What the 105 studies actually tested
- Why skin-tone performance is usually unclear
- Who is most affected by the gap
- How to use an acne app more safely
What the 105 studies actually tested
Most studies addressed diagnosis only. The review found 96.2% focused on diagnosis only, 9.5% on management only, and 5.7% on both, according to the International Journal of Dermatology review. Most models were image-based.
They comprised 76.2% deep learning, 9.5% classical machine learning, and 11.4% ensemble models. That focus matters for app users. A tool trained to label a photo may not help choose treatment, track change, or flag another skin condition.
Why skin-tone performance is usually unclear
Only 14 of 105 studies reported patient skin color by ethnicity and only 4 included this data. So performance by skin tone usually cannot be judged from the acne studies themselves. This is a reporting gap, not proof that every app fails.
Without tone-specific testing, a reader cannot tell whether accuracy holds across light, medium, and dark skin. Training libraries add to the concern. Reviews note common datasets are dominated by lighter Fitzpatrick types, with underrepresentation of non-White skin types documented by Guo and colleagues.
Who is most affected by the gap
People with darker skin face the largest uncertainty. Daneshjou and colleagues tested three dermatology AI models on the diverse DDI image set and found ROC-AUC was higher for Fitzpatrick I-II than V-VI, for example ModelDerm 0.64 versus 0.55, according to the 2022 Stanford Medicine and Science Advances study. ROC-AUC is a score for how well a model separates one condition from another.
A lower score means less reliable distinction. A related summary reports DeepDerm sensitivity fell from about 69% on Fitzpatrick I-II to about 23% on Fitzpatrick V-VI. Consumer apps have also shown accuracy varying by Fitzpatrick group in testing on diverse-tone images.
How to use an acne app more safely
Check what the app discloses before uploading photos. Look for training skin tones, validation by tone, and limits for acne versus other rashes or spots.
Bring the app result, original photos, and product history to the visit. That record helps the clinician judge the app suggestion quickly.
- Ask whether the app reports accuracy by skin tone or Fitzpatrick group
- Look for clear guidance that output is preliminary, not a diagnosis
- Save well-lit photos and note symptoms, duration, and changes
- Seek dermatologist review for uncertain, painful, spreading, or dark-skin lesions, consistent with the npj Digital Medicine survey on skin-tone criteria
You Might Also Like
- Artificial Intelligence for Acne Diagnosis in a 2026 Scoping Review: Skin-Tone and Equity Questions for People Using Acne Apps
- Skin-Tone Equity Gaps in 2026 Acne AI Research: How People Using Acne Apps Can Avoid AI Hype
- Acne AI Validation and Clinical Safety Questions in 2026: Skin-Tone and Equity Questions for People Using Acne Apps