What's Happening?
Researchers at Mass General Brigham, a Harvard-affiliated hospital network in Boston, have developed two artificial intelligence (AI) models that can predict a cancer patient's risk of death and estimate biological age from a single facial photograph.
Published in The Lancet Digital Health, these models, FAHR-FaceAge and FAHR-FaceSurvival, are built upon a foundation model called FAHR-Face, trained on over 40 million facial photographs. The FAHR-FaceAge model estimates how physiologically old a person appears, with an average error of 5.1 years. The FAHR-FaceSurvival model, trained on approximately 34,000 photographs of cancer patients undergoing radiation therapy, showed that patients with the highest risk scores had more than three times the mortality risk over roughly four years compared to those with the lowest scores. The study also found that each additional decade of apparent facial aging was associated with a 14% increase in mortality risk, even after accounting for other factors. The researchers noted that the AI models focused on facial regions similar to those described by Hippocrates over 2,000 years ago in relation to a patient approaching death.
Why It's Important?
This development is significant as it introduces a potentially cheap, noninvasive method for assessing a patient's physical reserves and frailty, particularly valuable for an aging population where conventional assessments can be challenging. The ability to predict mortality risk from a facial photograph could support personalized treatment decisions, helping doctors determine appropriate aggressive treatments like radiation, surgery, or medication. The technology could open a new field of medicine, making health-related information collection more accessible through common devices like smartphones or webcams. While the researchers emphasize that the systems are intended to support, not replace, medical decisions, the potential for early, non-invasive risk assessment could lead to more timely and tailored interventions, improving patient outcomes and potentially reducing healthcare costs associated with complex diagnostic procedures. However, experts caution that further research is needed to understand the underlying physiological changes the AI detects and to ensure the technology improves patient care beyond what physicians already assess.
What's Next?
Mass General Brigham has filed a provisional U.S. patent application for the next generation of its algorithms. Future research aims to determine if similar facial biomarkers can provide useful information about healthy individuals and other diseases. A separate clinical trial is underway to compare FaceAge assessments against physical strength and functional reserve tests in older lung cancer patients. Researchers also need to conduct further testing to ensure accuracy across diverse populations, including different skin tones, facial characteristics, and age groups, as the current test population for the survival model was predominantly white. The next step involves comparing the strongest predictions doctors can make with existing clinical information against predictions that incorporate facial analysis to ascertain the technology's added value. While a promising proof of concept, the technology is not yet ready for clinical use, requiring more validation and understanding of its mechanisms.
Beyond the Headlines
The ethical implications of AI-driven health assessments, particularly those based on facial appearance, are profound. Concerns about potential racial bias are significant, as algorithms trained on facial photographs could inadvertently incorporate historical inequalities in cancer care if certain populations have received poorer treatment. This could lead to biased predictions and exacerbate existing health disparities. The study's acknowledgment that the age model distinguished risk levels more effectively among white and Asian patients than among Black and Hispanic patients highlights this concern. Furthermore, the technology raises questions about data privacy and the potential for misuse of facial data, even if the researchers state that page content leaving the device is handled ephemerally. The rediscovery of Hippocrates' ancient observations through modern AI also points to a fascinating convergence of historical medical insight and cutting-edge technology, suggesting that some aspects of human health assessment are deeply rooted in observable physical cues, now quantifiable by AI.
