AI Reads X-Rays Smarter by Using Age, Sex, and Bone Info

When a doctor looks at an X-ray, they don't just see a picture. They know the patient's age, sex, and which bone was hurt. That context matters — a hairline fracture in an elderly person looks different than one in a child. Researchers wanted to teach AI to use that same context, so they built a system that combines the X-ray image with basic patient info to classify fractures.

The system was tested on 1,493 X-rays from Bangladesh. It used a standard image analysis tool plus a simpler network that processes patient details like age, sex, bone type, and which side of the body was X-rayed. The key innovation was a "consistency gate" — a safety check that ignores patient info if the X-ray itself clearly shows a different bone than what the records say. This prevents bad data from confusing the AI.

Results were promising. The AI correctly classified fractures more often than image-only systems, improving its accuracy score from about 0.57 to 0.60. More importantly, it stayed accurate when patient records were scrambled or missing. In a stress test where metadata was shuffled, the better design lost only 0.02 accuracy points instead of 0.06 — a big difference in real-world settings where records are often messy.

The bigger picture: hospitals in low-resource settings often have incomplete patient records. If AI can work reliably with missing or even wrong information, it becomes much more useful in real clinics, not just in perfect research conditions. The authors note the results still need testing in other countries and real hospital settings before we see this in everyday use.