How AI Spots Fractures Humans Overlook

A person slipped on the wet floor and hurt their wrist. Naturally, the wrist hurts, so they go to the emergency room to get an X-ray. The doctor doesn't see anything broken and sends them home with a splint and ibuprofen.

But two months later, the patient still can't turn a doorknob.

It's been too long, and they decide to see a different doctor who orders another scan. That scan clearly shows a tiny hairline fracture that was nowhere to be found on the first X-ray. Or was it just that the first doctor was lazy and incompetent?

The issue here isn't a 'bad' doctor but the limitations of human vision. Some fractures are simply too hard to see on standard X-rays because bones overlap, and cracks can be thinner than paper. And when you add the chaos of the emergency room to that, it's inevitable that some things will get missed.

2-9% fractures are missed on the initial radiographic interpretation in ERs. – National Institutes of Health

Luckily, it seems that this is yet another area where AI can help.

What AI Sees That Humans Don't

Being in pain is enough to make you frustrated, but imagine being in pain for weeks or months because someone didn't see a fracture on time.

There’s a estimated annual 5% diagnostic errors rate in the U.S. – Agency for Health Research and Quality

Imagine living in a small town like Elizabeth and going to Chicago for a check-up because you feel pain in your hip.

That's a 4-hour drive (in one way!), only to be told there's nothing wrong with you. And even if you were able to schedule a free hip fracture case review in Chicago, that's another day you need to lose on a medical appointment, and you can't even be sure the doctors will find out what's actually wrong.

Now, that all changes with AI.

Yes, you'd probably still need to go to a bigger city to get the best service possible, but AI would spot the issue your doctor couldn't.

Why? Because it learned to do that.

Researchers use thousands of X-rays, some of which have fractures, and some of which don't. They feed them into the program, and over time, that program starts to notice the tiny differences at the pixel level that even an experienced radiologist might miss.

AI tools which have been trained on imaging datasets have shown performance levels comparable to (or exceeding) radiologists. – Stanford University of Medicine

A human will use their experience and eyesight to look for what's wrong. AI, however, uses pure statistical pattern detection across every single case it's ever seen.

In this context, even a tiny change can mean a world of difference.

A fracture, for example, can show up as a barely visible change in bone density, or a slight asymmetry between two sides of an image. This isn't something you'd spot right away, especially if you're in a hurry.

Let's circle back to that hip fracture that was mentioned earlier because that's the perfect example of this problem. When a bone cracks, but stays in its place (a non-displaced fracture), it often looks 100% normal on the X-ray if it's in the early stage.

During early states occult/hidden hip fractures can be radiographically invisible; further imagining (e.g., MRI) might be required in such cases. – National Library of Medicine

The patient is in pain, but the image doesn't show why.

This is exactly where you'd want to use AI. It can scan every corner of the image with the same level of attention. And even if it's the 45th scan of the day, and there are 20 more patients in the waiting room, AI won't get tired or distracted.

With all that being said, it's important to know that this technology has its limits. AI doesn't understand context – not the way humans do – and it has no knowledge of the patient's history. All it can do is look at an X-ray and spot the crack, but it has no idea whether that crack is caused by osteoporosis or a fall from the ladder.

In other words, it absolutely can't replace a medical professional; it's a tool, nothing more.

How AI Helps in Day-to-Day Radiology

Here's what using AI in radiology looks like.

Catching Things During Busy Shifts

Imagine having 40 scans, and you have to go through all of them by noon.

It's impossible to stay laser-focused on all of them, but AI can do it without issues.

An increase in diagnostic errors in clinical settings is closely linked to fatigue and heavy workloads (burnout). – Harvard Medical School

And not only will AI spot issues like a champ, but it will also bump the most urgent cases to the top of the list.

Being a Second Set of Eyes

The AI can't exactly replace a radiologist, and nobody in this field thinks it'll ever be able to do that. But it makes an excellent assistant that kind of taps you on the shoulder and says, "Hey, look closer here."

This software can point out areas that are suspicious, and it will even tell you how confident it is that the issue is a fracture.

Keeping Things Consistent

You could have 5 radiologists look at the same X-ray and come to 5 different conclusions. That's completely normal because that's how human judgment works. But AI has the exact same rules for every single scan it sees, so it's perfectly consistent every time.

That's very important because it directly affects the quality of care patients receive.

Conclusion

Don't mistake AI for being magic. It's not.

But while the doctor can miss a fracture on the X-ray, AI usually won't. There are still some kinks to work out with it, of course, because this technology is very new, but it's also extremely promising.

The real power comes from radiologists and AI working together. The machine can see the patterns that look suspicious, and the radiologists can decide if that pattern is actually problematic or not, and how to treat the patient.

The only results worth the effort come from partnership.