An artificial intelligence model developed in the United States analyzes facial photographs of cancer patients to estimate biological age and identify features associated with mortality risk over the subsequent months, according to findings published in The Lancet Digital Health.
AI Estimates Mortality Risks Through Facial Features
- What the AI does: The software evaluates facial features from standard photographs to estimate biological age and flag elevated mortality risks in cancer patients.
- Where it looks: The model focuses heavily on specific facial landmarks, including the bridge of the nose, the area beneath the eyes, and nasolabial folds (smile lines).
- Current limitations: Researchers emphasize that the tool remains a prototype and cannot yet replace standard clinical evaluations or guide treatment decisions.
How the Algorithm Evaluates Facial Data
The artificial intelligence system was trained on a massive dataset comprising 40 million facial photographs to recognize underlying biological patterns without an initial specific task. Following that initial training phase, the model was directed to estimate biological age—defined as an individual’s physiological condition relative to their chronological age—which yielded encouraging preliminary results. Researchers subsequently tested the framework using facial photographs of 3,820 cancer patients to assign individual risk scores. More than half of the patients whom the system categorized as high risk died within three months of the photograph being captured. In a separate test involving over 38,000 patients, individuals whose biological age exceeded their chronological age exhibited a higher risk of mortality.
Despite these correlations, the precise mechanics driving the software remain opaque. When assessing mortality risk, the algorithm concentrates primarily on the bridge of the nose and the infraorbital region directly beneath the eyes. When calculating biological age, the model focuses on nasolabial folds, which are the creases extending from the edge of the nose to the corners of the mouth. As is common with deep learning architectures whose decision-making processes are difficult to interpret, researchers cannot yet explain why the system prioritizes these specific anatomical zones over others.
AI Lacks Proven Superiority over Clinical Evaluations
While clinicians routinely make informal prognostic assessments by observing patients during consultations, the new AI system has not yet been proven superior to standard clinical evaluations. Independent critics note that the research has not demonstrated whether the software adds meaningful prognostic value beyond what is already gathered through conventional diagnostic testing and physician observation. Technical vulnerabilities persist. The biological age estimations can fluctuate depending on lighting conditions, photographic angles, and subject positioning during image capture.
Demographic bias also limits the generalizability of the current data. The experimental cohorts were restricted to patients residing near Boston in the United States and Maastricht in the Netherlands, resulting in a heavy demographic skew toward white individuals. The research group intends to conduct subsequent trials to diversify the patient baseline.
Inaccuracies May Lead to Insurance or Employment Exploitation
Critics warn that diagnostic inaccuracies could lead to the underestimation of a patient’s care requirements. Over the long term, predictive facial software could theoretically be exploited by health insurance providers to alter coverage eligibility or by prospective employers to evaluate candidate health status based on algorithmic estimations.
Research into artificial intelligence applications in healthcare continues to attract substantial financial investment. However, validation requires more than statistical correlation; algorithms must prove their reliability across diverse patient populations, demonstrate added value beyond existing clinical metrics, and assist physicians in optimizing therapeutic decision-making.
