• General Dermatology
  • Eczema
  • Alopecia
  • Aesthetics
  • Vitiligo
  • COVID-19
  • Actinic Keratosis
  • Precision Medicine and Biologics
  • Rare Disease
  • Wound Care
  • Rosacea
  • Psoriasis
  • Psoriatic Arthritis
  • Atopic Dermatitis
  • Melasma
  • NP and PA
  • Skin Cancer
  • Hidradenitis Suppurativa
  • Drug Watch
  • Pigmentary Disorders
  • Acne
  • Pediatric Dermatology
  • Practice Management

Visual Racism and Racial Bias Persist in Dermoscopy Imaging

News
Article

In one study, less than 5% of images from a HAM10000 database came from patients with darker skin types.

Racial bias and visual racism are pervasive in several aspects of medical education, clinical trials, and imaging databases1,2—and this bias also translates into dermoscopy imaging, according to a recent exploratory examination published in the Journal of the European Academy of Dermatology and Venereology.3

Up close angle of woman's leg and hand
Image Credit: © Юля Бурмистрова - stock.adobe.com

Background and Methods

While diversity initiatives have increased across health care education and settings, further studies have demonstrated that studies of ethnic and racial minorities experience higher rates of negative experiences, higher attrition, and differential attainment.1 Pronounced disparities also persist in medical literature and databases, with prior research finding as few as 4% and as much 18% of images within medical texts and images are inclusive of patients with darker skin types.2

As such, researchers sought to investigate the occurrence of skin of color in the Harvard Dataverse repository, or HAM10000, database. The repository is utilized in multidisciplinary settings and includes more than 10,000 images of actinic keratoses and intraepithelial carcinoma/Bowen's disease, basal cell carcinoma (BCC), benign keratosis-like lesions, dermatofibroma, melanoma, melanocytic nevi, and vascular lesions.

Researchers utilized corner pixel sampling to accurately identify patients' original skin tone outside of the depicted lesion(s) and median RGB to balance out variables such as shadows or lighting differences. Additionally, the Gray World Algorithm was used to adjust the color balance and neutralize color biases.

Furthermore, researchers assessed skin lesions and conditions found in the HAM10000, comparing those in the database with common skin lesions found in patients with dark skin types.

Findings

The study found that less than 5% of photographs included in the database were representative of patients with darker skin tones, with greater than 90% of sampled images being inclusive of fair skin tones.

Aggregation of conditions such as nevi, melanoma, and BCC, which tend to be significantly more common in patients with lighter skin tones, demonstrated that these conditions accounted for 83.2% of images in the databases. Conditions that significantly affect Black patients, such as squamous cell carcinoma and dermatosis papulosa nigra, were notably missing from the database.

The photographs included in the review also demonstrated inadequate representation of the diverse appearances of nevi in darker skin. In addition, the dataset lacked insights into acral lentiginous melanoma and BCC, which also significantly affect Black and skin of color patients.

Conclusions

"This research highlights an imbalance in the HAM10000 data set's representation of skin lesions, particularly noting a lack of diversity in images of individuals with darker skin tones," wrote study authors Morales-Forero et al. "Compounded by the complete lack of metadata or any variables that would permit a detailed analysis of dermatological manifestations across various ethnic groups, this disparity poses a considerable barrier to comprehensively understanding and addressing the dermatological needs of a racially diverse population."

A limitation of the study, as noted by authors, is the potential of the sampling method to mistakenly identify shadows or lighting discrepancies.

"Visual racism is an issue that needs to be addressed in medical sources of information and education," wrote authors. "Image databases and AI models need to be nourished with information, including all skin types, to guarantee equal access to opportunities."

References

  1. Joseph OR, Flint SW, Raymond-Williams R, Awadzi R, Johnson J. Understanding healthcare students' experiences of racial bias: A narrative review of the role of implicit bias and potential interventions in educational settings. Int J Environ Res Public Health. 2021;18(23):12771. Published December 3, 2021. doi:10.3390/ijerph182312771
  2. Kaundinya T, Kundu RV. Diversity of skin images in medical texts: Recommendations for student advocacy in medical education. J Med Educ Curric Dev. 2021;8:23821205211025855. Published June 11, 2021. doi:10.1177/23821205211025855
  3. Morales-Forero A, Jaime LR, Gil-Quiñones SR, Barrera Montañez MY, Bassetto S, Coatanea E. An insight into racial bias in dermoscopy repositories: A HAM10000 data set analysis. J Eur Acad Dermatol Venereol. Published online June 14, 2024. https://doi.org/10.1002/jvc2.477
Recent Videos
© 2024 MJH Life Sciences

All rights reserved.