Cancer AI Technology Uses Face Photos as a Prognostic Tool
In This Article
- Clinicians use several factors when assessing cancer patients’ health to determine the best treatment option
- Investigators at Mass General Brigham found that people with cancer had a FaceAge about five years older than people without. And the older a person looked, the worse their predicted survival outcomes
Clinicians use several factors when assessing the health of patients with cancer to determine the best treatment option. Now, there’s another potential method to add to the toolbox: FaceAge. Powered by an artificial intelligence (AI) algorithm, FaceAge analyzes facial photos to predict a person’s biological age. More importantly, it predicts survival in people with cancer.
Subscribe to the latest updates from Oncology Advances in Motion
Investigators at Mass General Brigham found that people with cancer had a FaceAge about five years older than people without. And the older a person looked, the worse their predicted survival outcomes. These findings were recently published in The Lancet Digital Health.
Developing the FaceAge algorithm
To create the algorithm, researchers uploaded photos taken from publicly available databases. There were photos of more than 56,000 healthy people ages 60 and older. Researchers assumed the chronological age and biological age were a close match and validated it with clinical cohorts.
Once the FaceAge tool was ready, investigators took data collected from over 6,000 patients of two radiation oncology clinics. The data included photos of patients’ faces, their chronological ages, and when they died.
“[The AI tool] first localizes the face, and then it looks at different facial characteristics,” says co-senior and corresponding author Hugo Aerts, PhD, director of the Artificial Intelligence in Medicine (AIM) Program at Mass General Brigham. “Deep learning is really good at quantifying thousands of metrics in the face and using that for training a model to optimize this age prediction.”
Figure 1
FaceAge graphical depiction of an AI-generated man.
AI indicators of aging
FaceAge assesses aging differently than people do. “Being bald or gray matters less, for example, especially in older individuals,” Dr. Aerts says.
FaceAge focuses on the nasolabial folds (known as smile lines) and the temporalis muscle on the side of the head. These two areas align with the literature on aging, says co-senior author Ray Mak, MD, a faculty member in the AIM Program.
Nasolabial folds become more prominent as skin loses collagen and elastin and gravity pulls skin downward. Meanwhile, the temporalis is an area where people lose muscle mass as they get older. So, it appears the algorithm is paying attention to the right areas of the face, Dr. Mak says.
FaceAge as a tool for cancer prognosis
Drs. Aerts and Mak stress that AI tools like FaceAge aren’t meant to replace other tests clinicians use to assess patient health. However, AI can support subjective assessments like eyeballing.
“It's hard, right? You just look at somebody's face to decide if they're going to die or not. That’s not an easy thing for anybody to do, even an experienced doctor,” Dr. Mak says.
Researchers tested the accuracy of eyeballing by showing photographs of 100 patients with terminal cancer to several doctors. The doctors were asked to estimate whether the patient was likely to die within six months. Their estimates were slightly more accurate than flipping a coin.
Then, doctors were given clinical information about the patient and asked to estimate again. Their accuracy only improved slightly.
But when the physicians were given FaceAge information, their predictions were accurate 8 out of 10 times.
FaceAge offers clinicians another data point they can use to assess the patient’s health and decide on next steps. “We're not saying that FaceAge is going to replace human judgment, but at least it can provide a number to back up clinical intuition,” Dr. Mak says.
AI in the clinic: Next steps
“This isn't a technology that you can put into the clinic tomorrow just yet,” Dr. Mak says. “There's still a lot of work ahead.”
The research team is now applying their work across more patient settings. While the FaceAge algorithm is currently trained for patients with cancer, it can potentially be used in many specialties.
“Anytime we use chronological age to make a decision in medicine, you could use this algorithm instead,” Dr. Mak says.
Investigators are also testing this technology to predict diseases, general health status, and lifespan.
“We have a new tool, and we don’t know exactly how people are going to use this in 10 or 20 years,” Dr. Aerts says. “But there’s a lot of potential, and that is something we’re very excited about.”
Read the full study in The Lancet Digital Health
Learn about the Artificial Intelligence in Medicine (AIM) Program