AI predicts cancer survival from selfies but shows racial bias — SkimNews
Get the Health newsletter
Daily health & science — research, biotech, public health, the studies worth knowing. Free.
- Mass General Brigham researchers developed an AI model called FAHR-Face trained on over 40 million facial images to estimate biological age and predict cancer survival.
- FAHR-FaceSurvival was trained on nearly 35,000 photos of cancer patients and focuses on facial regions linked to mortality, such as under the eyes and the nose bridge.
- Dr Raymond Mak noted the facial features used by the AI align with those described by Hippocrates in 400BC as signs of muscle wasting and impending death.
- The Lancet Digital Health published the study showing AI-estimated facial age was significantly associated with survival in 18 of 28 cancer types.
- Prof Miroslaw Bober highlighted that the facial-age measure did not significantly predict survival for Black patients after adjusting for other factors, raising concerns about racial bias.
- Devi Sridhar questioned the reliability of selfies due to variables like lighting, expression, and cosmetic treatments such as botox and GLP-1 drugs that alter facial appearance.
- The researchers found alcohol, drugs, and smoking were the top three factors accelerating facial aging, a result Sridhar noted did not require AI to discern.
Why it matters: The AI model could influence how patient prognosis is assessed if validated, but its current racial bias and dependence on variable selfie conditions mean it risks reinforcing disparities or misleading clinicians. Until tested against real-world clinical judgment, its utility remains uncertain, especially as cosmetic interventions increasingly decouple facial appearance from health.
Ask SkimNews
