New AI tool predicts cancer survival rates using facial photos

Researchers have developed an AI tool capable of predicting cancer patients' survival rates using just a single facial photograph. The tool, trained on over 40 million facial images and data from more than 30,000 cancer patients, generates risk scores that correlate with the likelihood of death within 90 days. Raymond Mak, a radiologist at Brigham and Women's Hospital, noted that while doctors traditionally assess patients' appearance to gauge their condition, they often struggle to accurately predict survival times. The AI model, which estimates age with a 5-year margin of error, can reliably categorize patients into high- and low-risk groups, with significant differences in mortality rates between the two. The study, published in The Lancet Digital Health, highlights the tool's potential to supplement clinical judgment, particularly in resource-limited settings. Mak explained that the algorithm primarily focuses on the nasal bridge and under-eye areas, which align with historical observations by Hippocrates about signs of impending death. However, Miroslaw Bober, a professor at the University of Surrey, cautioned that the tool is not yet ready for clinical use, emphasizing that the study did not demonstrate its ability to provide clinically valuable information beyond doctors' assessments. The researchers aim to refine the tool through further testing across diverse patient populations and hospital settings.