AI Tool Predicts Cancer Survival from Patients’ Photos
Researchers have created an artificial intelligence system capable of estimating cancer patients’ chances of survival based on just a photograph of their face. Developed by experts at Mass General Brigham in Massachusetts, this AI tool is designed to assess biological age and predict survival outcomes for those diagnosed with cancer.
This breakthrough represents one of the most extensive demonstrations of how analyzing a selfie can be utilized to evaluate health and potentially inform medical decisions. The AI was trained using data from over 40 million facial images spanning various ethnicities and age groups.
Details of this advancement were shared in the Lancet Digital Health journal. The system can determine a person’s biological age—reflecting appearance rather than chronological age—and assess survival probabilities for cancer patients using an ordinary facial photo.
Medical professionals have developed two interrelated tools from this technology. FAHR-FaceAge estimates a patient’s biological age and predicts cancer survival, while FAHR-FaceSurvival directly predicts survival from facial images without relying on biological age as a middle step.
Even though both FAHR-FaceAge and FAHR-FaceSurvival stem from the same model, they provide distinct insights regarding a patient’s health. They utilize different facial features to make their respective predictions. For instance, FAHR-FaceAge focuses on features like the lines extending from the nostrils to the mouth corners, the forehead above the eyes, and the temples, whereas FAHR-FaceSurvival examines the area beneath the eyes and around the nasal bridge.
Co-senior author Hugo Aerts noted that “FAHR-FaceSurvival takes a step forward in facial health recognition AI by learning survival risks directly from cancer patients’ photographs.” This is, to our knowledge, the first instance where an AI model has been able to predict cancer survival based solely on a facial image.
In a study involving cancer patients who received radiotherapy in the US, those whose biological age, as estimated by FAHR-FaceAge, was five or more years older than their actual age showed a 21% increased risk of death over an average four-year follow-up period. Conversely, individuals who appeared five or more years younger had a 17% reduced risk.
According to Aerts, the key innovation lies in the fact that FAHR-Face learns a universal representation of the face from a vast dataset, which can then be applied to various health outcomes, including biological age and survival predictions. This foundational model creates a scalable method for developing multiple facial health indicators rather than requiring independent training for each model.
The study results indicate that estimates of biological age correlate closely with physical health, which may influence survival after certain treatments.
This advancement could facilitate AI facial analysis in helping doctors navigate challenging treatment decisions for patients. It holds the potential to enhance the prediction of risks following major surgeries or other cancer treatments.
Dr. Raymond Mak, a radiation oncologist at Mass General Brigham, emphasized the necessity for improved methods to evaluate physical resilience, especially as the population ages and the need for intensive treatments increases. Their aim is to make health assessments based on facial images more quantitative, consistent, and scalable to support personalized treatment decisions while still honoring physician expertise.
Researchers found that lifestyle choices significantly impacted perceived facial aging. For instance, excessive alcohol consumption, drug use, and smoking were all linked to appearing biologically older in the AI’s analysis of photos.
Mak mentioned another interesting observation: FAHR-FaceAge displayed remarkable consistency despite common variations in real-world photos concerning pose, lighting, expression, and makeup. It also remained stable across numerous cosmetic surgery procedures, except for a minor change noticed after rhinoplasty.
Although doctors acknowledge the potential for easily accessible assessments based on facial images to aid in personalized treatment strategies, they also stressed the need for further research before such tools can be implemented widely.
