The idea that a single selfie could predict how long someone with cancer may live sounds like an incredible breakthrough. AI is getting better at spotting patterns in medical data, but can it be used to reliably predict what will happen to an individual patient? I asked a clinical colleague what she thought about the recent headlines.
She found it an interesting development given that doctors have long used facial appearance to assess patient health. For example, she noted that they may write “appears stated age” or “appears older than stated age” as a proxy for general health. Now, researchers from Mass General Brigham, a healthcare institution in Massachusetts, have taken this idea a step further.
They have developed an AI model trained on more than 40 million facial images – called FAHR-Face – to assess biological age and predict survival in cancer patients. For biological age, they developed FAHR-FaceAge, which estimates age from certain facial features such as the forehead, the wrinkles between the nose and mouth and the temples. For cancer survival, they developed FAHR-FaceSurvival, trained using almost 35,000 photos of cancer patients.
This model focused more closely on the area below the eyes and the bridge of the nose – regions of the face “linked with dying”. Dr Raymond Mak, one of the study’s authors, noted that these parts of the face are the same as those referenced by Hippocrates in 400BC as signs of muscle wasting and mortality risk: “the nose sharp, the eyes sunken, the temples fallen in”. The results, published in The Lancet Digital Health this month, are intriguing – but far from conclusive.
Researchers found that AI-estimated facial age was significantly associated with survival in 18 of 28 cancer categories. So, basically, it works for some cancers, but not for others. In addition, Prof Miroslaw Bober noted, “The tool still needs testing against doctors’ own judgments and across different hospitals and patient groups.
For example, the facial-age measure was not significantly associated with survival for Black patients after adjustment for other factors.” That finding adds to the growing concern that AI models can exhibit racial bias based on the data used to train them and underlying assumptions. For example, racial bias has been shown when AI models analyse cardiac magnetic resonance (CMR) scans – used to diagnose and monitor heart disease. Let’s go back to what the published study does and doesn’t tell us.
Extract — continue reading at the source.
