(Photo courtesy of Tatiana Maximova via Getty Images)
A new artificial intelligence (AI) model can predict a person's biological age – their physical condition and aging process – based on selfies.
The model, called FaceAge, analyzes how old a face looks compared to its chronological age, or the number of years since birth. FaceAge’s developers say their tool will help doctors formulate more effective treatment strategies for diseases like cancer. However, one independent expert told Live Science that more data is needed to confirm that it actually improves treatment outcomes or quality of life before it can be used in that role.
When treating a cancer patient, one of the first things a doctor does is assess how the patient is doing, Hugo Aerts, director of the AI in Medicine program at Massachusetts General Hospital, said at a May 7 press briefing. “It’s often a very subjective assessment, but it can influence a lot of future decisions” about their care, including the intensity and aggressiveness of the treatment plan, he added. For example, doctors might decide that a patient who looks younger and fitter for their age will do better with aggressive treatment and possibly live longer than a patient who looks older and frail, even if their chronological ages are the same.
FaceAge could simplify that choice by turning doctors’ subjective assessments into a quantitative metric, the study’s authors wrote in a new paper published May 8 in the journal Lancet Digital Health. Quantifying biological age provides additional information that could help doctors choose the right treatment.
Aerts and his team trained the model on more than 58,000 photographs of people aged 60 and over who were considered to be in average health for their age at the time they were taken. In this dataset, the researchers had the model estimate chronological age, assuming that the biological ages of these people were similar, though the scientists noted that this assumption is not always true.
The team then used FaceAge to estimate the ages of more than 6,000 cancer patients. The study found that cancer patients looked, on average, five years older than their chronological age. FaceAge scores also correlated with survival after treatment: the older a patient looked, regardless of their chronological age, the lower their chances of living a long life. In contrast, the team found that chronological age was not a reliable predictor of survival in cancer patients.
FaceAge isn’t ready for use in hospitals or doctors’ offices yet. For one thing, the dataset used to train the model was pulled from IMDb and Wikipedia, which may not be representative of the general population and doesn’t account for factors like plastic surgery, lifestyle differences, or digitally altered images. Further research with larger, more representative datasets is needed to understand how these factors impact FaceAge’s scores, the authors say.
The researchers are continuing to refine the algorithm by adding more training data and testing its performance for conditions other than cancer. They are also studying what factors the model uses to make its predictions. But once it is complete, FaceAge could, for example, help doctors tailor the intensity of cancer treatments like radiation and chemotherapy to individual patients, study co-author Dr. Ray Mack, a radiation oncologist at Massachusetts General Hospital, said during a briefing. A clinical trial for cancer patients comparing FaceAge to more traditional methods of assessing a patient’s condition is set to begin soon, Mack added.
The researchers stressed that ethical guidelines around the use of FaceAge information, such as insurance companies having access to FaceAge scores when making coverage decisions, need to be established before the model is implemented. “This is certainly an issue that needs to be addressed to ensure that these technologies are used solely for the benefit of the patient,” Aerts said at the briefing.
Doctors should also think carefully about when and how they use FaceAge in clinical settings, added Nicola White, a palliative care researcher at University College London who was not involved in the study. “Working with people is very different from working with statistics,” White told Live Science. She noted,
Sourse: www.livescience.com
