Can a simple selfie reveal clues about cancer? New study finds surprising answers in patients’ faces

When you meet someone new, how do you determine how old they are? You see their facial features and body language and make a prediction in your mind. Apart from this, only your birthdate can tell someone how old you are. If there are two people in front of you, would you be able to determine their health? You can guess it but most of the time you won’t be right

Do you think anybody can determine how much your body can endure? Two patients of the same age can have very different levels of health, which makes treatment planning more complicated than checking a birth date. Various parameters need to be measured, but it is still challenging to determine if a body would be able to endure cancer or recover from it.

According to a report by CNN, some researchers at Mass General Brigham are studying, if a facial photograph can determine about the person’s health. They have made an AI model, which has now found a link between a person’s face and if they would be able to survive cancer. While the results are something to hold on, it is still a long way to tell if a selfie can do a cancer diagnosis.

What did researchers develop?

The team at Brigham has built an AI model. It is a foundation model named FAHR-Face, which is a short form for Foundation Artificial Intelligence for Health Recognition-Face. It was trained on more than 40 million facial photographs.

Researchers have built this model for two purposes. The First task of FAHR-FaceAge is to estimate age from facial features. The second task is to examine the facial features with survival outcomes in cancer patients.

The survival model was fine-tuned using 34,363 photographs of patients with cancer. Both tools were clinically evaluated using two independent cancer patient datasets. The research paper has been published in The Lancet Digital Health.

What could a photograph reveal about cancer survival?

The models identified differences between groups of patients with different outcomes. The model developed by the researchers ran multiple profiles of patients and identified differences between their diagnoses.

According to one of the analysis mentioned in the CNN report, in a batch of 122 patients classified in the highest-risk group, 52.5% died within 90 days. Among 975 patients in the lowest-risk group, about 2% died during that period.

While they found these results from their AI model, it does not mean that the system can accurately predict how long a person can live. Those figures describe outcomes across groups. They do not mean the system can tell an individual exactly how long they will live.


The study also found that patients in the highest-risk group had more than three times the risk of dying compared with those in the lowest-risk group, even after accounting for other medical factors.

Does FaceAge measure your biological age?

There is a difference in what FaceAge could find. While Chronological age means how old you are since birth, your biological age reflects changes in your body functions over time.

FAHR-FaceAge estimates age through appearance. It does not directly examine cellular ageing or measure the condition of every organ.

As a caution against dependency on AI models, University of Surrey professor Miroslaw Bober said that one should not treat an appearance-based estimate as a direct measurement of underlying biological ageing.

How might this influence cancer treatment?

Doctors already consider a patient’s general health when discussing treatment. According to the researchers, this system involving analysing the facial features can make part of that assessment more consistent.

The research has not shown that using these predictions to choose treatment improves patient outcomes. Nor should an algorithm’s risk score, on its own, determine whether someone receives treatment.

What needs testing before doctors can rely on it?

To trust any research or study, experts want prospective studies, which can help patients in any way rather than just assessing any record. Emory University’s Dr. Wilbur Lam said performance needs further evaluation across patient subgroups.

Bober also questioned whether the technology adds useful information beyond doctors’ assessments of patients’ health and fitness. Those assessments were not measured in the study, he noted.

While the AI model could help determine patients’ health, researchers still need to analyse how accurate the predictions on age and health are. They must show when the tool helps clinicians and whether it improves care.

.