When you twist your ankle or hurt your foot, getting a clear answer from an X-ray is crucial. A new analysis of 14 studies suggests that artificial intelligence (AI) models designed to read these X-rays might be very good at their job. The review found that, when looking at the combined data, these deep learning models correctly identified fractures about 93% of the time and correctly ruled them out about 95% of the time.
The analysis looked at how well different versions of these AI tools performed. Interestingly, models that used multiple X-ray angles didn't do any better than those using just one view. The AI was particularly sharp at spotting fractures in the heel bone (calcaneus). What we don't know from this review is who exactly was in these studies or what the real-world clinical setting was like.
It's important to remember what this review does and doesn't tell us. The results are promising, but they come from combining data from separate research projects. The authors themselves note that more studies with larger groups of patients and real-world testing are needed. Crucially, this analysis didn't compare the AI's performance directly against radiologists, so we can't say if it's better, worse, or just different from a human expert's read.