A photograph of a patient’s face could one day help doctors assess their outlook after a cancer diagnosis. Researchers are developing artificial intelligence tools to look for information in facial images that could complement medical assessments and help guide treatment. arXiv (source, opens in new tab)
arXivFoundation Artificial Intelligence Models for Health Recognition Using Face Photographs (FAHR-Face)Open source (opens in new tab) Mass General Brigham (source, opens in new tab)
Mass General BrighamAI facial health recognition tools help physicians assess age and survival risk to guide personalized treatmentOpen source (opens in new tab)
The idea is to teach a computer to recognise patterns in faces and turn those patterns into a risk score. When the researchers tested their system in two separate groups of cancer patients, higher scores were associated with higher rates of death during the study’s follow-up period.
That suggests a photograph may contain useful clues about a patient’s outlook. It does not mean the computer can diagnose cancer from a face or tell someone how long they will live.
The researchers started by training an AI system on more than 40 million facial images. They then adapted it for cancer patients, using 34,389 photographs to train the survival-risk tool, called FAHR-FaceSurvival. The large initial collection helped the system learn general patterns before it was trained for this more specific task.
The difference between the highest and lowest scores was marked. In the study’s preprint, patients in the highest-scoring quarter had 3.22 times the rate of death during follow-up compared with those in the lowest-scoring quarter. This difference remained after the researchers accounted for clinical factors relevant to survival.
A related tool, FAHR-FaceAge, estimates biological age: a measure of health-related ageing that can differ from a person’s age in years. Here, that estimate comes from patterns in a facial photograph. The gap between the computer’s estimate and a patient’s actual age was also independently associated with survival.
Mass General Brigham, the healthcare organisation reporting the work, says the tools are intended to support doctors’ judgement when tailoring treatment. But the research has not established that using them leads to better treatment decisions or better outcomes for patients. Whether they work reliably in everyday care and among other populations remains an open question.
- arXiv · PreprintFoundation Artificial Intelligence Models for Health Recognition Using Face Photographs (FAHR-Face)arxiv.org · opens in new tab
- Mass General Brigham · InstitutionAI facial health recognition tools help physicians assess age and survival risk to guide personalized treatmentmassgeneralbrigham.org · opens in new tab
