Why can a fracture be missed?
A suspected fracture is one of the most common reasons for a trip to A&E, and an X-ray is the standard test. But in emergency departments, X-rays are often read by clinicians who are not bone specialists.
According to the authors, missed fractures account for between 41% and 80% of reported diagnostic errors in some emergency departments. A fracture that is not seen can delay treatment and leave lasting problems, such as a bone that heals badly or arthritis.
What did the researchers do?
The team asked 18 senior, specialized orthopedic surgeons to label 135,409 X-rays, drawing a box around every fracture they could see. These labels act as a high-quality reference.
They then trained a deep neural network, a type of AI that learns to recognize patterns in images, first on all parts of the body, then on 31,490 wrist X-rays. The AI says whether it sees a fracture and shows where it is on the image.
Finally, 40 emergency clinicians (doctors and physician assistants) were recruited to read wrist X-rays twice: first alone, then with help from the AI.
What were the results?
Without help, the average clinician spotted 80.8% of fractures (this is sensitivity). With the AI, they spotted 91.5%.
Their specificity, meaning their ability to say "no fracture" when there is none, rose from 87.5% to 93.9%. So there were both fewer missed fractures and fewer false alarms.
Overall, the average clinician's rate of reading errors fell by 47%, and almost every clinician improved on both measures.
Key figures
- 135,409X-rays labeled by 18 orthopedic surgeons
- 80.8%fractures spotted by the average clinician without help
- 91.5%fractures spotted by the average clinician with help from the AI
- 47%drop in the rate of reading errors with help from the AI
A fall, an X-ray, then what? With Mima
This AI helped emergency clinicians miss fewer fractures, but only in an experiment. In real life, a fracture can still be missed on the first visit.
Mima helps you note how your pain and wrist movement change day by day, and keep your X-ray reports together. If nothing improves, you can ask: "Should I have another X-ray or a different test?"
What could this change for patients?
After a fall onto the hand, a wrist fracture could be spotted on the first visit to A&E, instead of being found later because the pain does not go away.
The authors argue that AI can pass on the know-how of highly specialized surgeons to the generalist clinicians on the front line, day and night.
What are the limits?
The experiment was done after the fact, on a web interface that mimicked the hospital's imaging software. The authors acknowledge that a prospective study, in a real department with the real software, would be needed.
The study mainly covers the wrist. It does not yet show whether help from the AI improves how patients recover, or how it performs for other bones.
Frequently asked questions
Can AI detect a fracture on an X-ray?
In this study, an AI spots and locates wrist fractures. Above all, it helps emergency clinicians miss far fewer of them.
Does the AI replace the doctor reading the X-ray?
No. Here it is used as an aid: the clinician looks at the X-ray, sees what the AI suggests, then decides.
What if my wrist still hurts after a normal X-ray?
See a doctor again. Some fractures are hard to see at first, and another test may be offered.
The original study
Lindsey R, Daluiski A, Chopra S, et al. “Deep neural network improves fracture detection by clinicians” PNAS, October 22, 2018. DOI : 10.1073/pnas.1806905115
Read the full study in PNASThis article summarizes a scientific study and is not medical advice.