AI Algorithm FaceAge That Can See Your Biological Age and Predict Cancer Outcomes

Updated: May 09 2025 06:29

AI Summary: Researchers have developed an AI tool called FaceAge that estimates a person's biological age from facial photographs. This tool revealed that cancer patients appear, on average, about five years older than their chronological age, and this difference strongly correlates with survival outcomes, even outperforming human clinicians in predicting short-term life expectancy in palliative care settings. The study suggests that FaceAge could provide an objective measure of biological aging, potentially improving treatment planning for cancer and having broader implications as a visible biomarker for overall health.

Your face may reveal more about your health than you realize. While we've long understood that some people appear older or younger than their actual age, researchers at Mass General Brigham have now quantified this observation with AI and discovered its profound implications for cancer treatment and survival outcomes.



In a recent study titled "Decoding biological age from face photographs using deep learning", researchers unveiled FaceAge, a deep learning algorithm capable of analyzing facial photographs to estimate a person's biological age, which can differ significantly from their chronological age. The technology, developed by Dr. Hugo Aerts and his team at Mass General Brigham's Artificial Intelligence in Medicine program, demonstrated remarkable ability to predict cancer survival outcomes across multiple types of the disease.

We can use artificial intelligence to estimate a person's biological age from face pictures, and our study shows that information can be clinically meaningful. This work demonstrates that a photo like a simple selfie contains important information that could help to inform clinical decision-making and care plans for patients and clinicians.

Among the study's most striking findings: patients with cancer appeared significantly older than their actual age—on average, about five years older than their chronological age. This gap between how old someone looks versus how old they actually are correlated strongly with survival outcomes.

"How old someone looks compared to their chronological age really matters," explained Dr. Aerts. "Individuals with FaceAges that are younger than their chronological ages do significantly better after cancer therapy."

The research team trained their algorithm on nearly 60,000 photographs of presumed healthy individuals from public datasets, then tested it on over 6,000 cancer patients from medical centers in the Netherlands and the United States. The tool outperformed human clinicians in predicting short-term life expectancy for patients receiving palliative radiotherapy.


When physicians see patients, they often make intuitive assessments about overall health and vitality based on appearance. These subjective impressions, combined with other clinical factors, inform treatment decisions. However, such assessments are inherently inconsistent and influenced by personal biases.

Dr. Ray Mak, co-senior author of the study, noted, "This opens the door to a whole new realm of biomarker discovery from photographs. As we increasingly think of different chronic diseases as diseases of aging, it becomes even more important to be able to accurately predict an individual's aging trajectory."

The researchers suggest FaceAge could help standardize and quantify what has traditionally been a subjective element of the clinical exam, providing an objective measure that could improve treatment planning, especially in complex cases where the benefits and risks of aggressive therapies must be carefully weighed.



FaceAge employs a two-stage deep learning process. First, a convolutional neural network locates and processes the face within the photograph. Then, a second network called Inception-ResNet v1 analyzes facial features to generate an age estimate.

What exactly is the algorithm seeing? While the researchers haven't specified which facial features contribute most to the predictions, previous studies on facial aging have identified several key indicators:

• Fine lines and wrinkles around the eyes and mouth
• Changes in facial volume and fat distribution
• Skin texture and pigmentation differences
• Subtle shifts in facial symmetry and proportions

The technology performed particularly well for individuals aged 60 and older—the demographic most relevant to cancer care—with an average error of just 4.09 years.



Perhaps the most immediate practical application for FaceAge lies in helping physicians make appropriate treatment decisions for patients with advanced cancer. When treating individuals with limited life expectancy, doctors must carefully balance potential benefits against side effects and quality of life concerns.

In one experiment, ten clinicians (including oncologists, palliative care physicians, and researchers) were asked to predict whether 100 patients with metastatic cancer would survive six months based on:

• Face photographs alone
• Photographs plus clinical chart information
• Photographs, charts, and FaceAge risk model predictions

The results were revealing. With photographs alone, clinicians' predictions were only slightly better than a coin flip. Adding clinical information improved accuracy significantly, but incorporating FaceAge predictions boosted performance even further—to the point where the combined approach matched the standalone FaceAge algorithm.

"This could be especially valuable in palliative care settings," noted Dr. Dennis Bontempi, one of the study's lead authors. "Accurately predicting short-term life expectancy helps ensure patients receive appropriate interventions that maximize quality of life without unnecessary treatments."



While the current research focused on cancer outcomes, the implications extend far beyond oncology. The researchers found evidence linking FaceAge estimates to molecular mechanisms of cellular senescence—the biological process of cellular aging and deterioration.

Through gene-based analysis, they discovered significant associations between FaceAge predictions and known senescence genes, correlations that weren't present when using chronological age alone. This suggests FaceAge may serve as a visible biomarker for deeper biological aging processes occurring throughout the body.

"I hope we can ultimately use this technology as an early detection system in a variety of applications, within a strong regulatory and ethical framework, to help save lives," said Dr. Mak.


Despite the promising results, significant work remains before FaceAge could be implemented in routine clinical care. The research team acknowledges several limitations and areas for further investigation:

• Testing across more diverse populations and settings
• Tracking how FaceAge estimates change over the course of treatment
• Evaluating how cosmetic procedures, makeup, and other appearance alterations affect predictions
• Exploring potential biases related to race, ethnicity, and gender
• Addressing privacy concerns regarding facial recognition technology in healthcare

The team plans to expand their research to patients with different cancer stages and to explore applications in other chronic diseases associated with accelerated aging.


FaceAge represents part of a broader trend toward using AI to extract clinically relevant information from everyday data sources. From voice analysis detecting Parkinson's disease to mobile phone usage patterns signaling cognitive decline, AI is increasingly finding hidden health signals in unexpected places.

What makes FaceAge particularly notable is its accessibility and non-invasiveness. Unlike expensive genetic tests or complex imaging procedures, facial photographs are simple to obtain and analyze. This positions the technology as potentially valuable not just in specialized academic medical centers but in community clinics and even remote settings with limited resources.

While FaceAge isn't yet available for clinical use, the research highlights the increasingly complex relationship between appearance, biological age, and health outcomes. It also underscores how factors beyond chronological age—including genetics, lifestyle choices, and disease states—affect how quickly our bodies age.

Dr. Osbert Zalay, another lead author on the study, suggests the findings should encourage a more nuanced view of aging:

Chronological age is just a number that tells you how long you've been alive. Biological age tries to capture how your body is actually functioning—and that's much more relevant to your health.

Research Paper: Decoding biological age from face photographs using deep learning