Archive · 28 Dec 2025
AI in Medical Scans Reveals Hidden Risks
Scientists at Harvard Medical School found that AI models used to analyze medical scans can learn things about patients that doctors can't see, like their age, sex, and race.
This hidden learning can make the AI less accurate for some people, depending on their age, sex, or ethnicity.
This is a big concern because it could lead to unfair treatment and put some patients at risk.
The study shows that we need to be careful about how we use AI in medicine to make sure it's fair and safe for everyone.
AI models trained on pathology slides can infer patient age, sex, and race, even though this information isn't visible to human doctors.
This hidden learning can affect diagnostic accuracy, leading to unequal performance across different demographic groups.
Some AI systems were less accurate for certain cancers depending on the patient's age, sex, or ethnicity.
The study raises serious concerns about bias, fairness, and patient safety in AI-driven medical diagnostics.
Researchers emphasize the need for careful consideration of how AI is used in medicine to ensure fairness and safety for all patients.
- Who
- Researchers from Harvard Medical School
- What
- AI models trained on pathology slides can infer patient age, sex, and race, affecting diagnostic accuracy
- When
- Not specified
- Where
- Harvard Medical School
- Why
- To understand the risks and biases in AI-driven medical diagnostics
- Research Institution
- Harvard Medical School
- AI Training Data
- Pathology slides
- Inferred Patient Information
- Age, sex, and race
- Impact on Diagnostics
- Unequal performance across demographic groups
- Concerns Raised
- Bias, fairness, and patient safety