Facial AI: Could a Photo Help Guide Cancer Treatment?

How could facial AI help assess cancer patients?

The researchers adapted their foundation model into FAHR-FaceAge, which estimates biological age from appearance, and FAHR-FaceSurvival, which estimates survival risk directly. According to the team's announcement, the development and validation study appeared in The Lancet Digital Health. These outputs are statistical estimates, rather than direct measurements of every aspect of health.

The two tools captured partly different information. Combining their scores improved survival-model fit in an external cancer cohort. That supports further investigation, while leaving open whether the additional information changes treatment decisions beneficially. [Mass General Brigham research announcement](https://news.massgeneralbrigham.org/en/ai-facial-recognition-tools-assess-age-cancer-survival-risk).

How do doctors assess whether cancer medicines are suitable?

For older adults, established assessment already extends beyond chronological age. ASCO's guideline recommends evaluating physical function, cognition, emotional health, other illnesses, medication burden, nutrition and social support. Patients aged 65 and older receiving systemic treatment who have identified impairments should receive management guided by those findings. This applies to chemotherapy, targeted therapy and immunotherapy. [ASCO guideline update](https://ascopubs.org/doi/10.1200/JCO.23.00933).

These assessments can lead to concrete interventions, such as medication review, nutritional support or additional help at home. Randomized trials underlying the guideline found that assessment-guided care can reduce serious chemotherapy toxicity. A possible future role for facial AI would be to flag patients needing closer evaluation; whether it adds useful information beyond these established assessments remains a research question.

What evidence is needed before facial AI can guide treatment?

Related research provides additional evidence of an association between facial aging and prognosis. A Nature Communications study analyzed serial photographs from 2,276 patients receiving radiation therapy. Faster AI-estimated facial aging was associated with worse overall survival after adjustment for several clinical factors. Because this was an observational analysis, it could not establish that facial aging caused poorer outcomes or that changing treatment based on the score would help. [Nature Communications study](https://www.nature.com/articles/s41467-025-66758-w).

The clinical implication is that prediction and treatment selection require separate evaluation. A score associated with mortality does not, by itself, identify which medicine will work or whether a lower dose is safer. A useful next step would be prospective comparison of usual assessment with assessment supplemented by facial AI, measuring treatment toxicity, quality of life and survival. Testing performance across patient populations would also help establish where the tool can be used reliably.