AI skin cancer detection tools are getting better, but only for people with light skin
Imagine you're getting out of the shower one morning and notice a mole on your thigh that you've never seen before. It's reddish-brown, bumpy and surprisingly large. Is it a benign mole, or is it melanoma?
New artificial intelligence tools claim they can accurately detect skin conditions like melanoma, through smartphone apps or clinician software. While promising for broad access to medical expertise, these tools struggle to accurately diagnose darker skin tones. The issue stems from AI's pattern recognition, which associates visual features with diseases, but becomes skewed by the skin color behind the lesions.
Tests showed AI models' accuracy dropped significantly when skin tones were digitally darkened. For conditions like atopic dermatitis, AI may identify lighter pink marks but fail to recognize gray or violet tones on darker skin. This disparity has critical consequences, as skin cancers are harder to spot on pigmented skin, leading to late diagnoses and poorer outcomes for patients of color.
The bias extends to AI chatbots like ChatGPT, where darker skin tones can trigger unnecessary panic or overlooked cancer diagnoses. To achieve equal accuracy across skin tones, AI models need diverse training data. However, ethical and privacy concerns hinder gathering enough images of darker skin tones. Synthetic images generated by generative AI may not accurately represent real skin conditions, risking misleading diagnostic tools.
Written by urgent.news from Medical Xpress's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
This story
This is one outlet's version. Read the fullest account.
