AI Tool Uses Face Photos to Estimate Cancer Patients' Survival Odds
The tool, trained on millions of healthy faces and 35,000 cancer patient photos, is far from hospital use, the researchers say
By AXY Health Desk ·
Key points
- US researchers report two experimental AI tools that estimate biological age and predict cancer survival from a single face photo.
- The FaceSurvival tool gives the most weight to the area under the eyes and around the bridge of the nose, according to the study.
- The tool was trained on millions of healthy faces and 35,000 photos of cancer patients, mostly from the US and the Netherlands.
- The researchers say the tool is a long way from being suitable for hospital use and is designed to support, not replace, doctors.
- The findings are preliminary and were published in The Lancet Digital Health.
US researchers say an artificial intelligence tool may be able to help doctors judge how ill a patient is by analysing a single photograph of their face. The preliminary findings, published in the journal The Lancet Digital Health, describe two AI tools that estimate biological age and predict cancer survival from a face photo. The researchers stress that the technology is a long way from being suitable for use in hospitals, but say their goal is to help doctors recommend the right treatments for their patients.
What the researchers built
The study describes two separate AI systems. The first estimates biological age, which is a measure of how old the body appears to be in terms of its physical condition, rather than the number of years a person has been alive. The second predicts survival in people with cancer. The researchers say both tools work from one photograph of the face, and that the features each one relies on differ. The cancer-survival tool is called FaceSurvival in the research.
The researchers describe the work as a preliminary step rather than a finished product. The tool has a patent application pending, which means the underlying technology is being protected while further development continues. The team has presented the work as research into whether faces contain useful signals about health, not as a diagnostic device that clinicians can use today.
How the tool reads a face
According to the researchers, the FaceSurvival tool gives the most weight to the area under the eyes and the area around the bridge of the nose when estimating how long a patient might have left to live. The scientists say these regions had the greatest influence on the tool's predictions. In the published illustrations, the areas that mattered most for survival and the areas that mattered most for age are highlighted on different faces, suggesting the two tools focus on partly different features.
It is worth explaining why a researcher would look at the face at all. Human faces change with age, illness, fatigue, nutrition and the effects of treatment. Clinicians have long paid attention to a patient's general appearance when judging how unwell they seem, but that assessment is subjective and can vary from one doctor to another. An AI system that measures visual patterns consistently, across thousands of images, offers a different kind of tool. Whether those patterns reflect real underlying biology, or are picking up something else in the photographs, is the central question that more research would need to answer.
The patients and data behind the study
The researchers say the tool was trained on millions of images of healthy faces, as well as 35,000 photographs of people with cancer. The metrics they applied covered many different types of cancer, including cancers of varying severity and cancers that had spread to the bones or the brain. This breadth matters because survival prospects differ greatly between cancer types and stages, and a tool that works only for one type would have limited use in a busy oncology clinic.
Most of the cancer patients were from the United States and the Netherlands, and most were in their 50s or 60s. Some were children, and others were in their 90s. The age range is significant because frailty, the general loss of physical reserve that makes a person more vulnerable to stress and illness, varies widely across life stages. However, the source material does not give a breakdown of how the tool performed for each group, so readers should not assume it works equally well for all ages or populations.
Why doctors assess how ill patients are
Doctors already examine patients and carry out a range of checks and tests to judge how unwell they are and what their prognosis, meaning the likely course of an illness, might be. Those assessments shape major decisions. Before recommending intensive treatments such as radiation, surgery or systemic therapy, which is treatment that travels through the whole body, such as chemotherapy, doctors need to know whether a patient is strong enough to tolerate it. A treatment that is too aggressive for a frail patient can cause serious harm, while under-treating someone who could benefit can mean missed chances to control disease.
Lead researcher Dr Raymond Mak, from Mass General Brigham in Boston, linked the work to demographic change. He said: “As the population ages, physicians need better ways to estimate physical reserve and frailty before recommending intensive treatments such as radiation, surgery or systemic therapy.” Physical reserve is a term for the body's capacity to withstand stress, and assessing it accurately is a long-standing challenge in medicine, particularly for older patients who are more likely to be diagnosed with cancer.
Dr Mak said the aim was to make face-photograph-based assessment “more quantitative, consistent and scalable so it can support personalised treatment decisions rather than replace physician judgment.” The phrase suggests that the researchers see the tool as one input among many, to be weighed alongside a clinician's examination, medical history and test results.
Limits and the road ahead
The most important caveat is that the research is preliminary. The source material states that the tool is a long way from being suitable for use in hospitals. Published early-stage studies typically need to be repeated and tested on new groups of patients before doctors can rely on them. Analysts following AI in medicine often ask whether a system trained on a particular set of images will perform as well on patients from different countries, ethnic backgrounds, camera types and lighting conditions. The source does not address these questions in detail, so they remain open.
There are also practical and ethical questions. A photograph of the face is easy to capture, and that simplicity is part of the appeal. Yet it raises questions about privacy, consent and how such images would be stored and used. If a tool like this were ever used in practice, patients would need to understand what it measures, how it influences their care and what its limits are. Experts commonly argue that AI tools in clinical settings should support, not override, the judgment of trained clinicians, and the researchers' own description of the tool as designed to help doctors rather than replace them reflects that view.
What could happen next
The next steps would likely involve independent testing, validation on new patient groups and comparison with the methods doctors already use. A key test would be whether the tool adds useful information beyond what a clinician already learns from a patient's records and examination. Researchers would also need to show that its estimates are consistent over time, for example when the same person is photographed at different points during treatment.
For now, the study is best understood as an early signal rather than a clinical advance. It suggests that the face may carry measurable information about biological age and cancer outcomes, and it gives researchers a new way to test those links. Whether that information can be turned into a reliable and fair tool that improves treatment decisions will depend on the results of larger and more diverse studies. Readers should watch for peer-reviewed follow-up work, regulatory discussion of AI medical tools, and any independent assessment of the patent-pending system.
Questions and answers
Can an AI tool really tell how ill someone is from a photo?
The researchers say an AI tool may be able to help doctors judge how ill a patient is from a single face photo. The study is preliminary, and the researchers say the tool is a long way from being suitable for hospital use.
Which part of the face does the tool focus on?
According to the study, the FaceSurvival tool suggests the area under the eyes and around the bridge of the nose are the most important parts of the face when estimating survival.
What data was the AI tool trained on?
The tool was trained on millions of healthy faces and 35,000 photos of cancer patients. Most cancer patients were from the US and the Netherlands, with ages ranging from children to people in their 90s.
Will this tool replace doctors?
No, according to the researchers. The tool is designed to help doctors with their assessments, and the lead researcher said it should support personalised treatment decisions rather than replace physician judgment.
Topics: artificial intelligence in medicine, cancer survival prediction, facial analysis, Mass General Brigham, The Lancet Digital Health, frailty and biological age, Raymond Mak
Sources: BBC Health
https://axynewsglobal.com/article/ai-tool-uses-face-photos-to-estimate-cancer-patients-survival-odds