Read more

July 27, 2023
2 min read
Save

AI doubles diagnostic accuracy of skin conditions among patients with skin of color

You've successfully added to your alerts. You will receive an email when new content is published.

Click Here to Manage Email Alerts

We were unable to process your request. Please try again later. If you continue to have this issue please contact customerservice@slackinc.com.

Key takeaways:

  • Overall, artificial intelligence (AI) diagnostic accuracy among patients with skin of color was 86.5%.
  • AI was most successful in diagnosing non-neoplastic conditions (90.98%).

Artificial intelligence was highly successful in diagnosing skin disease among patients with Fitzpatrick skin types IV to VI, according to a study.

Justine G. Schneider, BS, of the research department at Moy-Fincher-Chipps Facial Plastics & Dermatology and of the dermatology department at The Ohio State University College of Medicine, and colleagues reported that diagnostic accuracy of cutaneous and subcutaneous pathology is lower among patients with darker skin types, referencing one study that found diagnostic accuracy to be 44.3% in this population compared with 50.5% accuracy in patients with intermediate skin.

Telehealth
AI was highly successful in diagnosing skin disease among patients with Fitzpatrick skin types IV to VI. Image: Adobe Stock.

“The use of [artificial intelligence (AI)], especially that equipped to accurately diagnose conditions in patients with [skin of color (SOC)], may be a means of improving the diagnostic performance of providers assessing and managing dermatologic conditions,” Schneider and colleagues wrote.

In this study, Schneider and colleagues evaluated the effectiveness of AI in identifying and classifying cutaneous disease in patients with Fitzpatrick skin types IV to VI.

The researchers obtained 163 nonstandardized clinical photographs of skin disease manifestations from patients with skin of color that could be categorized into benign-neoplastic, malignant-neoplastic or non-neoplastic classes.

Each photo was uploaded to a custom-built AI software called Triage Inc. following a diagnosis from a specialist. The specialists’ diagnoses were confirmed through biopsies or responses to treatment.

The AI software offered a Top 1 diagnosis and a Top 2 diagnosis. Results showed that 86.5% of the Top 1 diagnoses and 93.25% of the Top 2 diagnoses were accurate, implying a “high degree of overall accuracy,” according to the researchers.

The AI was most successful in diagnosing non-neoplastic conditions (90.98%), and was also highly accurate in diagnosing malignant-neoplastic conditions (77.78%) and moderately accurate in diagnosing benign-neoplastic conditions (69.57%).

Overall, AI’s accuracy in diagnosing skin conditions in Fitzpatrick skin types IV to VI was 86.5% — a marked improvement from the previously reported 44.3%.

The uneven distribution of disease classes (75% non-neoplastic, 14% neoplastic-benign, 11% neoplastic-malignant) was a study limitation, according to the authors.

“While the development of AI in identifying skin conditions has advanced significantly over the past several years, opportunity for further refinement remains, especially for SOC,” Schneider and colleagues concluded.

Reference: