Highlights
FaceAge is an AI system that predicts cancer patients’ length of survival by estimating their biological age from a face photo.
Patients who look five years older than their actual age tend to have worse outcomes, even after adjusting for clinical factors.
FaceAge improved doctors’ accuracy in predicting six-month survival for terminally ill patients.
It is often a heart-stopping moment for patients when they hear their doctor’s prognosis that they have cancer.
For late-stage cancers, especially, the question that often arises is, “How long do I have to live?”
Doctors typically rely on experience and medical tests to make their best educated guess. Depending on the prediction, a series of treatments are recommended.
However, hospital researchers affiliated with Harvard Medical School are taking the guesswork out of this crucial prognosis.
They developed FaceAge, an artificial intelligence system that analyzes a photograph to more accurately estimate a patient’s biological instead of chronological age. For example, a healthy 75-year-old person may have the physiological traits of some 60 years old.
“We found that, on average, patients with cancer look approximately five years older than their chronological age and have a statistically higher FaceAge compared with clinical cohorts of patients without cancer who are treated for conditions that are benign or precancerous,” the researchers wrote in their paper, which was published in The Lancet Digital Health.
By correctly assessing the body’s true age, survival predictions become more accurate, which determines what treatments to give patients, among other measures of medical care.
“We showed that survival prediction performance of clinicians improved when FaceAge risk model predictions were made available,” the paper said.
FaceAge, which was developed using deep learning techniques, showed that patients who look older than their actual age are more likely to have worse outcomes, even after controlling for traditional clinical risk factors.
“Looking older was correlated with worse overall survival…,” according to the authors, which included Canadian and European researchers.
For patients, FaceAge represents a future in which one photograph could provide personalized insights into health, risk and treatment decisions that can accompany lab tests and medical scans.
For healthcare providers, FaceAge can complement their clinical judgment, which is especially crucial when treating seriously ill patients.
This is particularly relevant for cancer, where the narrow window of survival often forces doctors to make difficult decisions about aggressive treatments based on their own prognosis.
While AI is increasingly being used in medical settings, it cannot replace the crucial role that physicians and other caregivers play, healthcare experts told PYMNTS.
However, AI tools can be an important complement to ensure the patient gets a seamless digital experience, according to the PYMNTS Intelligence report “The Digital Healthcare Gap: Streamlining The Patient Journey.”
How FaceAge Works
The FaceAge AI model was trained on nearly 59,000 images of healthy individuals aged 60 or older and tested on 6,200 cancer patients from the United States and the Netherlands. Using a two-stage neural network system, the algorithm detects a face in a photo, extracts key features and generates an estimated biological age, per the paper.
The tool was better at predicting the length of survival than looking at the patients’ chronological age among three groups: those receiving curative radiotherapy, those with thoracic cancers, and those receiving palliative care for metastatic disease.
Patients with cancer had a FaceAge that was five years older on average than their actual age, a statistically significant gap.
FaceAge also improved survival predictions for terminally ill patients. When used alongside the TEACHH clinical model, a tool used to estimate life expectancy in patients undergoing palliative radiotherapy, FaceAge boosted the model’s accuracy in predictions, the paper said.
Physicians also performed better at predicting six-month survivals when aided by FaceAge, according to the paper. This could have a major impact on treatment decisions, helping clinicians weigh the pros and cons of therapy in patients nearing the end of life.
Despite its promise, the system raises important ethical considerations. The researchers acknowledged risks, including potential misuse by insurers or advertisers and racial or socioeconomic bias in the model.
Although FaceAge showed minimal bias across ethnic groups in preliminary testing, the researchers called for further validation using more diverse datasets and careful regulatory oversight, saying in the paper that “further assessments of bias in performance across different populations will be essential.”
While FaceAge is not yet ready for routine clinical use, its success marks a step toward integrating AI-based biomarkers into healthcare. It suggests that something as simple as a patient’s face may soon hold the key to more precise, humane and personalized care.
Going forward, the researchers said in the paper that testing with larger groups and further research are needed to “establish whether the findings extend to patients with other diseases.”
For all PYMNTS AI coverage, subscribe to the daily AI Newsletter.
