Can AI predict cancer survival better than a doctor? I have my doubts (www.theguardian.com)

Researchers from Mass General Brigham have developed an AI model called FAHR-Face that analyzes facial images to predict cancer survival and biological age. Trained on over 40 million images, FAHR-Face employs two components: FAHR-FaceAge assesses biological age based on specific facial features, while FAHR-FaceSurvival predicts survival probabilities by focusing on areas of the face historically linked with mortality. Initial findings indicate that AI estimates correlate with survival in 18 out of 28 cancer types, highlighting its potential utility in oncology. Despite this promise, significant concerns persist regarding the model's reliability and applicability. Its performance varies across different cancer types and demographic groups, with evidence suggesting racial bias—its predictive accuracy diminished for Black patients. Additionally, the study emphasizes that while a selfie may offer clues about health, it is limited by various factors such as lighting and cosmetic enhancements. The implications of this research suggest that while AI could supplement existing clinical assessment methods, it currently falls short of providing a definitive tool for predicting patient survival in practice.