IT Home On October 20, according to Bloomberg, researchers have developed an artificial intelligence tool that can predict the survival chances of cancer patients based solely on a facial photo. This tool can complement a doctor’s clinical experience and is expected to provide a reference for formulating treatment plans, especially in situations where medical information is limited.
This new model, after training, can estimate a person’s age and predict survival outcomes for cancer patients. The research team used over 40 million face images and photos of more than 30,000 cancer patients to train the model, many of whom have already passed away. A study published in The Lancet Digital Health shows that the tool generates risk scores for patients, and those with the highest scores are more likely to die within 90 days.
Research scientists said that this type of research result is unprecedented. Raymond Mak, a radiologist at the Brigham Cancer Center at Massachusetts General Hospital and one of the paper authors, said that doctors can assess the condition of patients based on their physical appearance during training, but it is “usually difficult for doctors to accurately estimate the remaining survival period of patients at the end of life.”
Mark said in an interview: “We believe that with artificial intelligence, just a facial photo can capture certain physiological reserves and the risk of death in the human body.”
The error in this AI tool’s age estimation is approximately 5 years, and it can reliably divide patients into two groups: high-risk and low-risk. There are significant differences in the risk of death between the two groups. The model works effectively among people of different ages, races, genders, and various types of cancer.
IT Home notes that more and more AI tools are conducting trials in the field of cancer detection and treatment: some assist in diagnosis by analyzing images, while others analyze genes and lifestyle risk factors to predict an individual’s risk of cancer.
Although the experimental results seem promising, Miroslaw Bober, a professor of video processing and machine learning at Surrey University, believes that this tool cannot be used in clinical settings yet. He did not participate in this research.
“Predictive results are not equal to explaining the causes.” Bobel said, “The study did not prove that these predictions can provide additional clinically valuable information beyond doctors’ assessments of a patient’s overall health status. The study itself did not measure such assessments by doctors.” He added that this tool still needs more testing compared with doctors’ subjective judgments, across different hospitals and patient groups.
Mark, who developed the tool with the team, said that the model’s excellent prediction of lifespan was very surprising. The team found that the algorithm mainly focused on the bridge of the nose and the area below the eyes during evaluation. After researching, Mark discovered that Hippocrates described it as early external signs of impending death: sunken eye sockets, collapsed temples, and a thin nose tip.
Mark said that the tool will not be used to reject or limit patients from receiving treatment. “We know that, when dealing with dying patients, doctors sometimes provide excessive treatment.” In his view, it is simply “an additional reference for doctors.”