# Exploring the Diagnostic Capability of Artificial Intelligence in Dermatology for Darker Skin Tones: A Narrative Review

## TL;DR

This paper reviews how AI performs in diagnosing skin conditions in darker skin tones and highlights the need for diverse datasets to improve accuracy.

## Contribution

The paper provides a focused narrative review on AI diagnostic performance in darker skin tones and the impact of diverse training data.

## Key findings

- AI models show lower accuracy in diagnosing skin conditions in darker skin tones (Fitzpatrick IV-VI).
- Training AI with diverse datasets improves its diagnostic accuracy for darker skin tones.
- Artificially pigmented images can help enhance AI accuracy in darker skin tones.

## Abstract

As the use of artificial intelligence (AI) as a diagnostic tool increases in the field of dermatology, there has been a growing need to diversify datasets to improve its diagnostic capability in darker skin tones. Currently, AI is not as effective as a diagnostic tool in darker skin tones (Fitzpatrick IV-VI) as it has been in lighter skin-toned (Fitzpatrick I-III) populations. This narrative review will provide a summary of the recent data and advancements made within the area. Medline and PubMed databases were searched with the following search terms: dermato* AND (skin tone or race or skin colour or ethnicity or race or Fitzpatrick) AND (ai or artificial intelligence). Texts were filtered for full text and English language from 2020 to 2025. Results including patients under 18 years of age were excluded, which resulted in 52 papers. After scanning through titles and abstracts, a total of eight papers remained that were relevant to the review. AI models have demonstrated lower accuracy in recognising cutaneous pathology in darker skin tones in the majority of studies. When looking at the results after training the models with diverse datasets, there was an overall improvement in the accuracy of AI to recognise pathology in Fitzpatrick skin tone IV-VI. Several studies also showed that there is some benefit to training AI with artificially pigmented images to improve its accuracy. AI has significant potential to enhance dermatology by improving diagnostic accuracy, reducing variability, and improving efficiency. Expanding datasets further appears to be of benefit in improving accuracy in darker skin tones. Further studies with larger sample sizes are needed to analyse other reasons the algorithms have lower accuracy in darker skin tones and how this could be mitigated.