Could a Selfie Predict Your Cancer Prognosis?

What if we could predict cancer survival outcomes from a simple photograph?

Researchers from Mass General Brigham have made significant strides in this direction. Their innovative approach involves the development of a deep learning algorithm called FaceAge, which analyzes a photo of a person’s face to estimate their biological age and predict survival outcomes for cancer patients. Their results are published in The Lancet Digital Health.

Estimating biological age and cancer prognosis by a “selfie”

Researchers have long utilized artificial intelligence (AI) in healthcare, but its use in predicting cancer outcomes through facial photos is a relatively new frontier. Unlike other AI tools that focus on specific disease indicators or genomic data, this model leverages visual characteristics in facial features, offering a non-invasive and easily accessible way to gather valuable information. This AI model analyzes facial features to predict biological age, potentially providing insights into cancer prognosis.

Biological age – the state of one’s body at a cellular or molecular level – differs significantly from chronological age. While chronological age is simply the number of years a person has lived, biological age can be influenced by various factors such as lifestyle, environmental exposures and genetic makeup.

Understanding biological age is crucial in cancer research as it is a stronger predictor of cancer outcomes than chronological age. As the body ages, cellular damage accumulates, impacting immune response and tissue regeneration, which significantly influences cancer development and progression.

A patient's appearance can offer physicians valuable insights into their health, guiding treatment alongside chronological age and biological measures. However, biases about age may influence decisions, underscoring the need for objective, predictive tools in care.

With that goal in mind, Mass General Brigham researchers leveraged deep learning and facial recognition technologies to train FaceAge.

"We can use AI to estimate a person’s biological age from face pictures, and our study shows that information can be clinically meaningful,” said co-senior and corresponding author Hugo Aerts, PhD, director of the Artificial Intelligence in Medicine (AIM) program at Mass General Brigham.

“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. How old someone looks compared to their chronological age really matters – individuals with FaceAges that are younger than their chronological ages do significantly better after cancer therapy,” he continued.

Cancer patients show older FaceAge, linked to reduced survival

The FaceAge tool was trained on 58,851 photos of presumed healthy individuals from public datasets. Aerts and team then tested the algorithm in a cohort of 6,196 cancer patients from two centers, using photographs routinely taken at the start of radiotherapy treatment.

Results showed that cancer patients appear significantly older than those without cancer, with their average FaceAge being approximately five years older than their chronological age. Older FaceAge was associated with worse survival outcomes in the cancer patient cohort, especially in individuals who appeared older than 85 years – even after adjusting for chronological age, sex and cancer type.

An important consideration in cancer treatment is estimating end-of-life survival time, a factor that is often challenging to determine. To address this, Aerts and his team asked 10 clinicians and researchers to predict short-term life expectancy based on 100 photos of patients undergoing palliative radiotherapy. Even after clinical context, such as the patient’s chronological age and cancer status, the clinician’s predictions were only slightly better than a coin flip. Yet when clinicians were also provided with the patient’s FaceAge information, their predictions improved significantly.

Research continues to allow for real-world applications

Break Through the NGS Interpretation Bottleneck To Improve Patient Outcomes

This eBook explores how to enhance throughput while maintaining high-quality variant analysis and reporting.
View eBook

Advertisement

As with many emerging AI tools, additional research is essential before this technology can be confidently integrated into routine clinical practice. Currently, the research team is focused on evaluating the tool’s ability to predict various diseases, assess overall health status and estimate lifespan.

In the next phase, follow-up studies will aim to broaden the scope of this work by expanding across multiple hospitals, incorporating diverse patient populations at different stages of cancer and monitoring how FaceAge predictions evolve over time. Furthermore, the team plans to validate the tool’s accuracy by comparing its results against data sets from plastic surgery and makeup applications, offering a more comprehensive understanding of its potential real-world applications.

“This opens the door to a whole new realm of biomarker discovery from photographs, and its potential goes far beyond cancer care or predicting age,” said co-senior author Ray Mak, MD, a faculty member in the AIM program at Mass General Brigham.

He concluded: “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. 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.”

Reference: Bontempi D, Zalay O, Bitterman DS, et al. FaceAge, a deep learning system to estimate biological age from face photographs to improve prognostication: a model development and validation study. Lancet Digit Health. 2025. doi: 10.1016/j.landig.2025.03.002

This article is a rework of a press release issued by Mass General Brigham. Material has been edited for length and content.