AI tool predicts cancer survival from selfies

An AI tool named FAHR-FaceSurvival, developed by researchers at Mass General Brigham hospital in Massachusetts, can predict the survival chances of cancer patients based on a selfie. The tool is based on FAHR-Face, an AI model trained on over 40 million facial images, and was improved with 34,363 photos of cancer patients.

Focus on specific facial areas

The tool focuses on the area under the eyes and around the nose, which is associated with mortality. Tested on a sub-cohort of 3,820 patients from the Harvard-RT test dataset, the tool identified 122 patients with a high risk of dying, of whom 52.5% died within 90 days following the evaluation, according to a study published in the Lancet Digital Health journal.

AI also estimates biological age

In addition to FAHR-FaceSurvival, researchers have developed FAHR-FaceAge, a tool capable of estimating a patient's biological age by analyzing the nasogenal furrows, the area just above the orbits, and the temples. These regions are associated with aging.

Testing on various datasets

FAHR-FaceAge was tested on various datasets, including Harvard-RT and APPA-REAL, which consists of individuals presumed to be in good health. Among 1,060 patients whose biological age exceeded their chronological age by more than 20 years, 21% died within 90 days.

Lifestyle factors affect aging

Conversely, approximately 6% of the 4,195 patients whose biological age was at least 10 years younger than their actual age died within 90 days. The researchers note that makeup, surgery, and other cosmetic facial modifications have not affected the performance of FAHR-FaceAge. However, lifestyle factors such as heavy alcohol consumption and smoking are associated with premature aging.

Complementary health insights

The researchers emphasize that FAHR-FaceAge and FAHR-FaceSurvival collect different and complementary information about a patient's health, with their predictions weakly correlated and remaining independently associated with survival when analyzed together. They clarify that their tools are not intended to replace doctors but to assist them by making health assessments from facial photographs more quantitative, consistent, and easily applicable to support personalized therapeutic decisions.

Limitations and accessibility

The researchers highlight the use of "cost-effective and easily obtainable public facial photographs," which expands the clinical application of FAHR-FaceSurvival and helps overcome "limitations related to the size and generalizability of clinical datasets." However, they acknowledge that the quality of images, lighting, and facial expression can affect the accuracy of both tools. Additionally, the cancer patient cohorts used in the study mainly come from centers in high-income countries, potentially restricting the generalization of the results to other diseases, regions, and underrepresented racial groups.