AI is helping doctors detect diseases, from heart problems in electrocardiograms to autism in children. However, studies show AI can be biased, leading to incorrect diagnoses for some groups. A new study in Science Advances found that an AI model used to read chest X-rays often missed diseases in Black patients and women.
The model, called CheXzero, was trained on nearly 400,000 X-rays but struggled to detect conditions like an enlarged heart in Black women. When tested on different data sets, the problem remained. Researchers believe the AI may be using hidden signals in X-rays that humans can’t see, leading to biased results.
To reduce this bias, researchers tried giving the AI additional information, such as the patient’s age, race, and sex. This improved accuracy for some diseases but not all. Experts say AI needs more diverse training data, and human doctors should always be involved in reviewing AI-generated results.
Source: science
