Researchers at Mass General Brigham in Massachusetts have created an artificial-intelligence system that analyses facial photographs to estimate biological age and to forecast survival for patients with cancer. The model, named FAHR-Face, was trained on more than 40 million images and a separate set of roughly 35 000 cancer-patient photos, according to a study published in The Lancet Digital Health.
FAHR-FaceAge focuses on cues such as forehead shape, nasolabial folds and temple contour to produce a facial-estimated age. A companion model, FAHR-FaceSurvival, concentrates on the region beneath the eyes and the nasal bridge, areas historically linked to wasting and mortality. Dr Raymond Mak, one of the authors, noted that these facial zones echo observations made by Hippocrates over two millennia ago.
The analysis showed that the AI-derived facial age correlated significantly with survival outcomes in 18 of 28 cancer types examined. However, the association disappeared for Black patients after statistical adjustment for other variables. Professor Miroslaw Bober cautioned that the tool still requires validation against clinicians’ judgments and across diverse hospitals and patient populations.
The study relied on retrospective data, identifying statistical links across thousands of cases rather than prospectively forecasting individual outcomes. Translating such associations into reliable, patient-specific predictions remains a substantial leap. Moreover, the authors acknowledge that a single selfie may capture health signals, but the evidence does not yet demonstrate that these signals can guide clinical decision-making.
Selfies are notoriously variable: lighting conditions, camera angle, facial expression and even time of day can alter apparent age. Cosmetic procedures such as Botox, fillers, facelifts, and emerging anti-aging drugs that modify facial fat further complicate the relationship between visible age and underlying health. The researchers tested the age model against some of these factors, yet residual error is inevitable.
Proponents argue that an objective facial metric could complement clinicians’ intuition, reducing reliance on subjective notes such as “appears older than stated age.” Yet the tool must be benchmarked against experienced physicians and proven to improve outcomes before it can be adopted in routine oncology care. Without such validation, its contribution may be limited to academic interest rather than practical decision support.
The project’s website invites users to upload a selfie to receive a biological-age estimate, but the consent form states that each image becomes part of the ongoing dataset. Currently the service provides only the age estimate, and the authors warn against interpreting this as a health diagnostic. Users should therefore treat the output as experimental rather than a definitive assessment of cancer prognosis.