Roxana Daneshjou, MD, PhD, is a board-certified dermatologist and physician-scientist who serves at Stanford University as an assistant professor in both the Department of Biomedical Data Science and the Department of Dermatology.

Roxana Daneshjou, MD, PhD

Daneshjou said diversifying who develops AI technology is essential to catching bias before algorithms and data sets reach patients.

She called for equity testing to happen during AI development, not only after products launch.

Daneshjou said she couldn't yet endorse specific camera systems for skin of color imaging, though both Apple and Google appear to be investing in improvement.

Roxana Daneshjou, MD, PhD, assistant professor of biomedical data science and 

 at Stanford University, outlined concrete steps the medical and technology fields can take to close AI equity gaps for patients with skin of color.

Her discussion of these steps continues a conversation with Morayo Adisa, MD, medical director of Dermatology Physicians SC in Chicago and Kenilworth, Illinois, which took place during the 

Skin of Color Savvy: The Art and Science of Treating Patients of Color

What Dermatologists Can Do to Make AI More Equitable

What Can Be Done to Improve AI Equity in Dermatology?

Daneshjou said progress starts with diversifying who is involved in developing artificial intelligence technology, noting that representation is essential to recognizing when data sets or algorithms fail to reflect diverse populations. She credited organizations such as SOCS for prioritizing this work and said expanding diverse data sets remains a key priority. She also pointed to growing reliance on chatbots for medical information, stressing that researchers must identify and publicize flaws in how these tools respond to health questions relevant to patients with skin of color.

Daneshjou referenced her own prior research documenting racially biased chatbot responses to medical questions, noting that public attention after its publication contributed to later improvements from technology companies. She said she would prefer equity testing built into development rather than addressed only after products reach market.

Are Some Cameras or AI Systems Better for Skin of Color?

Asked whether any camera systems or algorithms currently outperform others for imaging skin of color, Daneshjou said she was unaware of empirical data supporting specific claims and was unwilling to endorse any product without direct testing. She noted that both Apple and Google have taken steps toward improving how their technology captures diverse skin tones, based on conversations with research teams at each company.

Daneshjou also referenced her past criticism of an early Google dermatology algorithm with limited representation of Fitzpatrick skin types V and VI, saying she remains willing to challenge large tech companies when warranted, while acknowledging both are investing in improvement.

Editor’s note: This episode was summarized with the help of AI tools.

Kaundinya T, Kundu RV. Diversity of Skin Images in Medical Texts: Recommendations for Student Advocacy in Medical Education. J Med Educ Curric Dev. 2021 Jun 11;8:23821205211025855. doi: 10.1177/23821205211025855. PMID: 34179498; PMCID: PMC8202324.

Adisa M, Daneshjou R. Skin of Color Savvy: Tech Check—AI, Telederm, and the Bias Question. HCPLive. July 13, 2026. Accessed August 10, 2026. 

https://www.hcplive.com/view/skin-color-savvy-tech-check-ai-telederm-bias-question

Videos

Roxana Daneshjou, MD, PhD, outlines steps to close AI equity gaps in dermatology, from diverse data sets to camera testing for skin of color.

Research shows AI systems used in hospitals have allocated fewer resources to Black patients by relying on health care spending as a flawed proxy for illness severity.

A 2022 study found dermatology AI algorithms performed significantly worse at detecting skin cancer on brown and Black skin compared with White skin.

Daneshjou emphasized that AI tools often reflect and reinforce existing structural inequities in health care rather than correcting them.

 at Stanford University, spoke in a recent podcast about how artificial intelligence (AI) systems can replicate structural racism embedded in health care.

Offering real-world examples in a conversation with Morayo Adisa, MD, medical director of Dermatology Physicians SC in Chicago and Kenilworth, Illinois, Daneshjou was featured alongside Adisa on the latest episode of 

Does Artificial Intelligence Worsen Health Care Disparities?

Daneshjou, who splits her time between clinical dermatology practice and running an AI-focused research lab at Stanford, said her work centers on evaluating whether artificial intelligence performs equitably across patient populations, spanning both imaging-based systems and large language models. She said findings across these studies consistently show that artificial intelligence tools often reinforce existing inequities within health care rather than correcting them.

How Does AI Bias Show Up in Real Health Care Systems?

To illustrate the issue, Daneshjou pointed to a widely cited study led by Ziad Obermeyer, MD, published in Science, which evaluated an algorithm already in use across multiple hospital systems to determine which patients received additional support following discharge.

The analysis found that Black patients needed to be considerably sicker than White patients to receive the same level of allocated resources. Daneshjou explained that the algorithm relied on health care spending as a proxy for illness severity, a flawed measure given that Black patients have historically spent less on health care due to systemic barriers in access, housing, and socioeconomic status, despite often being sicker.

What Did Research Reveal About AI and Skin Cancer Detection?

Daneshjou also referenced her own 2022 research evaluating dermatology-focused artificial intelligence algorithms designed to detect skin cancer. Although none of the algorithms studied were in clinical use at the time, they were being discussed as candidates for future deployment.

Her team found that these systems performed significantly worse at identifying skin cancer on brown and Black skin compared with White skin, raising early concerns about scaling such technology without addressing performance gaps across skin tones.

Adisa M, Daneshjou R. Skin of Color Savvy: Tech Check—AI, Telederm, and the Bias Question. HCPLive. July 13, 2026. Accessed August 7, 2026. 

A Stanford dermatologist explains how AI systems can reinforce racial bias in health care and underperform on skin cancer detection in darker skin.

Skin of Color Savvy: Tech Check—AI, Telederm, and the Bias Question

Skin of Color Savvy: The Art and Science of Treating

, Morayo Adisa, MD, medical director of Dermatology Physicians SC in Chicago and Kenilworth, Illinois, sat down with Roxana Daneshjou, MD, PhD, assistant professor of biomedical data science and dermatology at Stanford University, to examine one of the most rapidly evolving topics in medicine: the intersection of artificial intelligence, 

, and equitable care for patients with skin of color.

AI systems can perpetuate existing healthcare disparities if trained on biased or nonrepresentative datasets.

Improving dermatology education and diversifying AI development are critical to achieving equitable care.

Teledermatology expands access but must address challenges including camera bias, digital access, and patient trust.

, featured a discussion exploring both the promise and limitations of artificial intelligence (AI) in dermatology, with Daneshjou emphasizing that while AI has the potential to improve healthcare delivery, it also risks perpetuating longstanding disparities if not developed and evaluated thoughtfully.

Daneshjou explained that many AI systems inherit biases already present within healthcare because they are trained on historical data that reflect existing inequities. She highlighted research demonstrating that algorithms have, in some cases, allocated healthcare resources inequitably by relying on spending data rather than true disease burden, inadvertently disadvantaging Black patients. Within dermatology specifically, she discussed her own work evaluating skin cancer detection algorithms, which found significantly poorer performance when identifying malignancies on darker skin tones compared with lighter skin.

A central theme of the episode was that AI cannot solve problems rooted in deficiencies within the healthcare system itself. Daneshjou noted that dermatology education has historically underrepresented skin of color in textbooks, educational resources, and training materials. As a result, clinicians may diagnose conditions later in patients with darker skin, limiting the availability of high-quality, early-stage clinical images needed to train more equitable AI systems. Improving AI performance, she argued, must begin with improving dermatology education and increasing representation across all aspects of clinical training.

The conversation also examined practical steps dermatologists can take to improve the future of AI. Daneshjou advocated for diversifying image datasets, ensuring broader representation among researchers and technology developers, and routinely evaluating algorithms for fairness before widespread implementation. She stressed that equity should be considered during the development process rather than only after products reach the market.

Adisa and Daneshjou also discussed the expanding role of teledermatology, acknowledging that virtual care has increased access for many patients while introducing new challenges. Although teledermatology can improve access for individuals with transportation or mobility limitations, Daneshjou cautioned that disparities persist through unequal internet access, language barriers, and differences in smartphone camera technology. She noted that camera systems have historically been calibrated using lighter skin tones, potentially reducing image quality and diagnostic accuracy for patients with darker skin during store-and-forward teledermatology encounters.

Patient trust emerged as another important focus. Daneshjou emphasized that patients should be informed whenever AI is incorporated into their care, whether through documentation tools, clinical decision support, or other technologies. She advocated for explicit patient consent surrounding AI use, particularly given evidence that current systems may not perform equally across diverse populations.

The discussion also touched on AI scribes, which have become increasingly common in clinical practice. Although available within her own institution, Daneshjou explained why she has chosen not to adopt the technology, citing concerns about patient privacy, inaccurate documentation, hallucinated clinical information, and the importance of personally crafting assessment and treatment plans as part of her own clinical reasoning process.

Looking toward the future, Daneshjou acknowledged that AI will almost certainly become more deeply integrated into healthcare but stressed that ongoing advocacy, rigorous evaluation, and greater diversity among developers will be essential to ensuring these technologies improve rather than worsen health disparities. She also identified textured hair and hair disorders as an area where AI research remains notably sparse, presenting an opportunity for future innovation in skin of color dermatology.

The episode concluded with a call for continued collaboration among clinicians, researchers, technology developers, and organizations such as the Skin of Color Society to ensure that advances in AI and digital health are guided by principles of equity, transparency, and representation, ultimately benefiting all patients regardless of skin tone.

To learn more about SOCS’s programs and initiatives, visit 

Podcasts

In this podcast episode, 2 featured experts discuss AI, teledermatology, and strategies to reduce bias in skin of color dermatologic care.

Roxana Daneshjou, MD, PhD

Articles