The Lancet Digital Health publishes the FAHR-Face study: using facial photographs to estimate biological age and survival risk in cancer patients
The Lancet Digital Health published a study in which the team developed the FAHR-Face model, trained on more than 40 million facial images, to estimate biological age and survival risk in cancer patients from facial features. After evaluation on data from about 43,000 cancer patients, 52% of the 122 patients classified by the model as high risk died within 90 days, while only 2% of the 975 patients classified as lowest risk died in the same period. The study notes that the method has a low barrier to entry, but causality has not been established, and how much additional predictive value it offers compared with existing clinical methods remains undetermined.
Why read this
AI score 62
The study trained a model on more than 40,000 facial images and provides specific mortality comparisons for risk stratification of cancer patients, offering insight into the boundaries of facial images as a health screening signal.
Topics
Body · AI translation
Recently, an interesting study was published in the journal The Lancet Digital Health. In this study, researchers developed an artificial intelligence model called FAHR-Face (Foundation of AI for Health Recognition). The model was trained on more than 40 million facial images and was adapted for two key tasks. The two models capture different types of information:
The first AI model estimates a person's biological age by analyzing facial features. This may differ from chronological age, but it helps to understand their overall health status and physical resilience.
The second model estimates the survival risk of cancer patients based on facial photographs. The areas around the eyes and the bridge of the nose appear to be particularly important for making these predictions.
How reliable are the results?
Overall, the researchers were granted access to data from approximately 43,000 cancer patients to evaluate the AI model. They then conducted another test on 3,820 of these patients. The model classified 122 of them into the high-risk group. Of these patients, 52% died within 90 days.
The model classified 975 patients into the lowest-risk group. During the same period, only 2% of these patients died.
Significance
A major advantage of this approach is its low technical and economic threshold. Anyone can take a simple facial photograph in a few seconds.
However, a clear causal relationship has not yet been established. AI can identify statistical patterns, but it cannot automatically explain the biological processes that give rise to these patterns. In addition, it remains unclear whether these findings apply to patients with other diseases.
Conclusion
AI appears to be able to extract information related to biological age and the mortality risk of cancer patients from a single facial photograph. The study shows that this association persists even after accounting for clinical prognostic factors. However, it remains undetermined how much additional predictive value facial photographs can provide compared with existing clinical methods.
Anyone who wants to try using facial AI to estimate their own biological age can register on this website and participate in this "public study." Users can upload 3 to 10 facial photographs, either all at once or daily. The system provides a personal biological age score for each photograph, as well as an overall score. Privacy is protected: the photographs are not shared and are automatically deleted after a certain period of time. Other information will be anonymized and provided to other researchers.