A new artificial intelligence (AI) model can predict cancer patients' risk of mortality from a single facial photograph, according to research published in The Lancet Digital Health. In a study involving 3,820 cancer patients, 64 of the 122 patients with the highest risk scores passed away within 90 days of their photographs being taken.
Developed by Ray Mak, a radiation oncologist at Mass General Brigham Cancer Institute, and his colleagues, the model aims to make clinical assessments of patients' health and frailty more objective. Doctors often evaluate how healthy or seriously ill a patient appears when deciding on treatment plans, including surgery, chemotherapy, and radiation.
However, such assessments can be subjective, researchers noted. The model may also detect visible changes linked to health factors such as smoking and heavy alcohol consumption.
For instance, smokers and heavy drinkers were more likely to appear biologically older than their actual age. The algorithm also accounted for common forms of plastic surgery, although it tended to rate individuals who underwent rhinoplasty as slightly older in post-surgery photographs.
Experts caution that the technology is not yet ready for clinical use, and its potential role in treatment decisions raises significant ethical concerns. The study highlights a growing intersection between machine learning and oncology.
By analyzing subtle facial features that might escape the human eye, the AI provides a data-driven perspective on patient prognosis. This could eventually assist medical professionals in prioritizing palliative care or adjusting the intensity of aggressive treatments.
Despite the promising results, the research team emphasizes that the AI is intended to supplement, not replace, clinical judgment. The ethical implications of using facial recognition for mortality prediction remain a subject of intense debate within the medical community.
Future iterations of the model will likely focus on increasing the diversity of the training data to ensure accuracy across different demographics. As the technology evolves, regulatory bodies will need to establish guidelines for its implementation in hospital settings.
Ultimately, the goal is to improve patient outcomes by identifying those who may be at the highest risk of rapid decline. While the current findings are preliminary, they represent a significant step forward in the application of digital health tools for personalized cancer care.