AI tool uses selfies to predict biological age and cancer survival
A deep learning algorithm was presented Thursday (8) in the journal The Lancet Digital Health: FaceAge, which converts a simple foreground photo into a number that more accurately reflects a person's biological age, rather than the date of birth in their medical records.
Trained on tens of thousands of photos, it determined that cancer patients were, on average, five years biologically older than their healthy peers.
The study authors say this could help doctors decide who can safely tolerate harsh treatments and who would do better with a gentler one.
“Our hypothesis is that FaceAge could be used as a biomarker in cancer treatment to measure a patient’s biological age and help the physician make these difficult decisions,” said Raymond Mak, a study co-author and oncologist at Mass Brigham Health, a Harvard-affiliated health system in Boston.
In the hypothetical case of two patients, one who is agile at 75 years old and has a biological age of 65 and the other who is more fragile, aged 60 with a biological age of 70, aggressive radiotherapy may be more appropriate for the first, but dangerous for the second.
The same logic can be applied to decisions related to heart operations, hip replacements or palliative care.
There is growing evidence that humans age at different rates, depending on their genetics, stress, exercise habits and habits such as smoking or drinking alcohol.
While expensive genetic tests can reveal how DNA degrades over time, FaceAge promises to get inside your body with just a selfie.
The model was trained on 58,851 portraits of presumably healthy adults over 60 years of age, extracted from public databases.
It was then tested on 6,196 patients receiving treatment in the US and the Netherlands, with photos taken before radiotherapy. Patients with malignant tumors looked, on average, 4.79 years older biologically than their chronological ages.
Among cancer patients, a higher FaceAge score predicted worse survival, even after accounting for age, sex and tumor type. The odds dropped dramatically for those whose biological age was over 85.
FaceAge determines signs of aging differently than people typically do. For example, gray hair and baldness are less important than subtle changes in facial muscles.
Six doctors were asked to examine photos of the faces of terminally ill cancer patients and predict which of them would die in the next six months. With the FaceAge data in hand, their predictions improved considerably.
The model also confirmed a now-famous internet meme when he estimated the biological age of young American actor Paul Rudd at 43 years old from a photo taken when he was 50.
AI tools have been scrutinized for not paying enough attention to people of color.
Mak said preliminary checks have not revealed significant racial bias in FaceAge's predictions; however, the group is training a second-generation model on 20,000 patients.
They also test how factors such as makeup, cosmetic surgeries and variations in lighting can trick the system.
The ethical debates are evident: an AI that can read biological age from a selfie could be a boon for doctors, but also a temptation for life insurers or companies looking to measure risk.
Knowing that the body is biologically older than previously thought can motivate positive changes in health or sow anxiety, another dilemma on the table.
The researchers plan to open a publicly accessible FaceAge portal, where people can upload their portraits to participate in a research study to validate the algorithm. Commercial versions for doctors will follow, but only after further validation.
ia/arp/atm/nn/mel/jmo/aa
IstoÉ
