New Delhi: A photograph of the face could one day help doctors assess how old a person appears biologically and estimate survival risk in cancer patients, according to a new artificial intelligence (AI) model developed by researchers and tested in separate groups of patients.
The model, called Foundation Artificial Intelligence for Health Recognition-Face (FAHR-Face), was trained on more than 40 million facial photographs and adapted for two tasks — estimating biological age, or how old a person appears to be physically, and predicting the risk of death. The findings were published in The Lancet Digital Health in 2026.
When tested on publicly available photographs, the AI model estimated biological age with an average error of about five years. Among 38,183 patients treated at six US centres affiliated with Harvard, those who appeared biologically older than their actual age tended to have poorer survival outcomes.
A second AI model, FAHR-FaceSurvival, was trained on photographs of 34,363 cancer patients and tested on another 3,820. Patients classified in the highest-risk group were more than three times as likely to die as those in the lowest-risk group, even after researchers accounted for other factors, including patients’ clinical characteristics and demographic details.
The researchers also tested the models on 4,892 cancer patients treated in the Netherlands. The results suggested that the AI could help estimate survival risk in patients beyond the original US group.
Dr Mandeep Singh Malhotra, director of surgical oncology at the CK Birla Hospital, Delhi, said the approach was an interesting development, but its reliability in India needed to be established.
“An important consideration is whether this facial recognition AI can deliver equally reliable results in the Indian population, given our distinct epidemiological profile and diverse facial features. This remains to be established,” he said.
Malhotra said the technology could eventually complement precision oncology if combined with genomic testing and other diagnostic techniques, including liquid biopsy. Such an approach, he said, could potentially improve predictions of cancer survival, help detect recurrence earlier and enable timely intervention.
“Rather than viewing it as a standalone technique, its potential may lie in complementing established diagnostic approaches, including genomics and liquid biopsy, as part of a more comprehensive, personalised approach to cancer care,” he said.
The two AI models appeared to identify different aspects of health-related information from facial photographs. Using them together improved the accuracy of survival-risk estimates, suggesting that facial analysis could eventually complement existing tools doctors use to assess patients and plan treatment.
However, the model cannot diagnose cancer or predict an individual’s lifespan. Since the study mainly involved patients in the US and the Netherlands, its findings may not apply to other populations. Lighting, image quality and facial expressions may also affect estimates. Further validation is needed before the technology can be used routinely in clinical practice.
