FaceAge AI: New Biomarker Could Predict Cancer Prognosis from Facial Photos

Researchers have developed a tool that measures biological aging from facial photographs, potentially offering a non-invasive window into cancer outcomes.

The Concept

Scientists at Mass General Brigham have created FaceAge, an artificial intelligence deep learning tool that estimates biological age from a single facial photograph. Now, a new study published in Nature Communications introduces a groundbreaking extension: the Face Aging Rate (FAR).

FAR measures changes in biological age between two photographs taken at different time points, and researchers believe it may serve as a non-invasive biomarker for cancer prognosis.

Key Findings

In a study analyzing 2,279 cancer patients who had facial photographs taken at two different time points during treatment, the median FAR revealed that patients' facial aging outpaced their chronological aging by 40%.

Two Critical Metrics

The study introduced two primary measurements:

• Face Aging Rate (FAR): The rate of biological aging over time, calculated as the change in FaceAge between two photographs divided by the time interval between them.
• FaceAge Deviation (FAD): The difference between biological age estimate from a single photo and chronological age.

How FaceAge Works

Previous research established the foundation for this work:

• In an earlier study, cancer patients appeared approximately five years older than their chronological age according to FaceAge, with older estimates correlating with worse survival outcomes.
• Another study of over 24,500 cancer patients aged 60 and older found that 65% had a FaceAge older than their chronological age. Those with a FaceAge 10 or more years older had significantly worse survival.

Mortality Risk by Time Interval

The study divided patients into groups based on the time interval between photographs, revealing a clear pattern:

Higher FAR was associated with lower overall survival probability, and this association was strongest when the interval between photographs was two years or more. Notably, FAR outperformed single timepoint measurements in predicting survival over longer intervals.

Combined Assessment

Patients with both high FAD and high FAR had poorer survival, but FAR more stably predicted survival over longer intervals. The authors propose integrating FAR with baseline FAD for more nuanced health assessment.

Possible Mechanisms

Researchers suggest FAR may reflect dynamic biological aging, including disease progression and treatment effects. The approach is described as noninvasive, easy to obtain, cost-effective, and repeatable—making it a potentially valuable clinical tool.

Study Limitations

The research team acknowledged several constraints:

• The sample was limited in ethnic, racial, and age diversity
• No data was available on disease progression or treatment causality
• Potential bias due to photograph timing and missing confounders
• Requires validation in prospective studies

Future Research

Further studies are needed to evaluate FaceAge and FAR in more diverse populations. The research team is conducting ongoing and planned prospective trials to investigate outcomes in patients with different cancers and other diseases.

Public Access

The team has launched a web portal (faceage.bwh.harvard.edu) where the public can submit photos for FaceAge assessment and participate in research.