TL;DR
Turn quiet afternoons into listening time
- Thousands of audiobooks, podcasts and originals
- Listen on your phone, tablet or Echo — also offline
- Cancel anytime
Researchers report that FAHR-Face, an AI model trained on more than 40 million facial photographs, found links between facial age and survival outcomes among cancer patients. The findings suggest photographs could offer a noninvasive source of prognostic information, but they do not establish that the model can guide care; further validation is needed.
Researchers have developed FAHR-Face, an artificial intelligence model that analyzes facial photographs for information associated with cancer survival. In a study published in The Lancet Digital Health, patients whose faces the model estimated to look older than their chronological age had poorer survival outcomes, suggesting photographs could offer a noninvasive prognostic signal; the findings require further validation before clinical use.
The researchers pretrained the model on more than 40 million consumer-grade facial photographs, using a method that learns patterns from images without needing a health label for every photograph. They then adapted it for two tasks: estimating facial age, called FAHR-FaceAge, and predicting mortality risk directly from photographs of cancer patients, called FAHR-FaceSurvival.
According to the study in The Lancet Digital Health, FAHR-FaceAge was evaluated in two clinical groups: 38,183 patients treated at Harvard-affiliated radiotherapy centers in the United States and 4,892 cancer patients who received radiotherapy in the Netherlands. In public image datasets, the age-estimation model had a mean absolute error of 5.1 years, meaning its estimates differed from chronological age by that amount on average.
In the study’s survival analysis, researchers compared estimated facial age with chronological age and found that patients whose faces appeared older tended to have poorer survival. At 90 days, mortality was 21.3% in the group whose estimated facial age was more than 20 years older than chronological age, compared with 6.4% in the group whose estimated facial age was 10 years younger. The researchers adjusted survival analyses for clinical prognostic factors, but these associations do not show that facial appearance causes worse outcomes.
What Facial Images Could Add
If the findings hold up in further studies, photographs could offer a low-cost, noninvasive source of information alongside existing clinical assessments. A model that detects patterns not readily captured by routine measures might help researchers study aging and health or identify patients who warrant closer assessment.
For now, the study establishes an association, not a clinical tool. It does not show that using the model improves treatment decisions or survival, and a photograph-based estimate cannot replace established prognostic information. Any future use would require evidence that performance is reliable across patient groups and settings, and that the predictions are useful in actual care.
As an affiliate, we earn on qualifying purchases.
From FaceAge to Survival Prediction
The study builds on earlier work by the same authors involving FaceAge, a deep-learning model that estimates biological age from facial photographs. The authors reported that higher facial-age estimates were associated with poorer survival among cancer patients. That earlier model had been developed mainly using White patients aged 60 and older, limiting how broadly its results could be applied.
FAHR-Face expands the image training base and separates two questions: how old a person appears and whether facial features are associated with survival risk. The study authors also examined performance across demographic and cancer subgroups and tested age-estimation robustness to factors including lighting, makeup, facial pose and cosmetic surgery. These evaluations address sources of image variation, but do not by themselves establish readiness for clinical deployment.
What Validation Must Establish
The study reports links between facial estimates and outcomes, but it remains unclear how well the survival model will perform across broader and more diverse patient populations, cancer types, and care settings. The researchers report that FAHR-FaceSurvival was fine-tuned using 34,363 cancer photographs; evaluation included 3,820 remaining patients from that dataset and another independent patient dataset. The supplied report does not give a single overall accuracy measure for its survival predictions.
It is also not established whether the model adds useful information beyond the clinical factors already used to assess prognosis, or whether its predictions would change care in ways that benefit patients. The researchers say further validation is needed before clinical translation. Potential effects of image quality and demographic differences also need careful assessment.
Further Testing Before Clinical Use
The next step is independent validation in additional patient groups and clinical settings, with attention to how the model performs across demographic and cancer subgroups. Researchers would also need to establish whether its predictions add value beyond standard clinical prognostic factors and remain reliable under ordinary image-taking conditions.
Until those questions are answered, FAHR-Face is a research model. The study published in The Lancet Digital Health does not announce a clinical rollout or a change to cancer treatment; it points to a possible research direction that requires further testing.
Key Questions
What is FAHR-Face?
FAHR-Face is an AI model developed to analyze facial photographs. Its two versions estimate apparent biological age or predict survival risk among cancer patients.
Does looking older mean a cancer patient will have a shorter life?
No. The study found an association between older facial-age estimates and poorer survival outcomes. It does not establish that appearance causes a patient’s outcome or predict what will happen to any individual.
How accurate was the model?
For age estimation on public image datasets, FAHR-FaceAge had a mean absolute error of 5.1 years. That figure measures age-estimation error, not the accuracy of the cancer survival predictions.
Can doctors use this model now?
The study does not establish FAHR-Face as ready for clinical use. The authors say the findings need further validation before clinical translation.
Source: rss
Fall Picks
fall essentials
As an affiliate, we earn on qualifying purchases.
