AR Nearly Halved Beginners’ Needle Error. One Number Should Stop You

What happened

Putting a drip in someone’s arm sounds simple. When the vein is hard to find, a nurse or doctor uses an ultrasound probe to see it. That is where the trouble starts.

The ultrasound picture appears on a screen beside the patient. The needle is in your hand, over here. So you keep looking away from your hands to the screen, then back. Your eyes are in one place and your hands are in another. Researchers call this dividing attention, and it is the reason the job is harder than it looks.

Augmented reality offers an obvious fix. Put the ultrasound picture where the arm is, floating in place, so your eyes and your hands look at the same spot. Researchers call that “in situ”, meaning in position. The team compared it against the normal way: a monitor.

The test

The team compared in-situ AR against a monitor-based display for ultrasound-guided drip insertion. They measured two things that matter: how accurately people hit the target, and how often the needle punctured tissue it should not have.

They reported results separately for novices. That is a good choice. A tool that helps beginners and does nothing for experts is still a useful tool, but only if you know which group you are buying it for.

The result

For novices, spatial accuracy rose from 50.6 percent, plus or minus 8.2, to 65.9 percent, plus or minus 13.6. Mean targeting error fell from 1.26 millimetres to 0.79 millimetres. That is a large improvement in a job measured in millimetres.

Tissue perforations in novices fell from 29.4 to 3.0.

Now look at the numbers attached to those perforation figures. The 29.4 comes with a spread of plus or minus 45.6. The 3.0 comes with a spread of plus or minus 2.7.

Why does that matter so much? This is because a spread larger than the average tells you the average is not describing anybody. If you perforate 29 times on average but the spread is 45, then some people in that group barely perforated at all and at least one person perforated a huge number of times. The 29.4 is not a typical performance. It is a mixture of very different people, flattened into one figure. A SPREAD BIGGER THAN THE AVERAGE IS A WARNING, NOT A RESULT.

The AR figure does not have that problem. Three, plus or minus 2.7, describes a group that behaved consistently. So the honest version of this finding is not “AR cut perforations by a factor of ten”. The honest version is: without AR, novice performance was wildly unpredictable, and with AR it became predictable and low.

That is arguably a better result than the one the big number suggests. Consistency is what you want from a medical procedure.

What it means

The accuracy gain is real and the direction is clear. But notice what the headline number hides, and what it reveals when you read the spread instead of the average.

For anyone buying this kind of equipment, the lesson generalises. Ask for the spread, not just the average. A vendor who shows you only means is showing you the half of the data that flatters them. Therefore, “plus or minus” is the most important part of any claim.

Business ideas from this paper

1. Spread-and-variance reporting for clinical device trials. Most internal device evaluations report averages only. Offer to re-analyse a hospital’s own trial data and report variability. Buyer: hospital quality and safety teams. First test: ask three quality managers whether their last device evaluation reported standard deviations. Blank looks are the opportunity.

2. Ultrasound-guided cannulation training for hard-stick patients. The study says beginners are erratic, not uniformly bad. Erratic means trainable. Build a short, measured course aimed specifically at difficult veins. Buyer: nursing schools, agency staffing firms and dialysis units. First test: run one paid session and measure each trainee’s spread, not their average.

3. A plain-language claim checker for medical device buyers. Procurement officers read vendor one-pagers full of averages. Sell a one-page review of a vendor’s published evidence, in plain words. Buyer: procurement teams at mid-size hospitals and clinic chains. First test: do one free for a clinic you know and ask whether it changed the decision.

How sure can you be?

Moderately sure about the direction. Much less sure about the size.

The strong point is that this is peer reviewed and published in a serious venue, which many studies in this field are not. The accuracy improvement for novices is consistent across two different measures, which makes it more believable.

The limits are important. The abstract does not state how many people took part, so the figures above come without a group size, and a small group would make all of them shakier. The enormous spread on the perforation figure means that single comparison should not be quoted as a ten-fold improvement by anyone. And a study of novices on a controlled task is not the same as a ward at 2am. Watch for an independent group repeating this with a stated number of participants and with experienced staff as well as beginners.

Do this today

Find any claim with an average in it — a product spec, a job advert, a health statistic. Look for the spread. If there isn’t one, you have been shown half the data.

Source: In Situ vs. Monitor-Based: Evaluating Augmented Reality for Ultrasound-Guided Peripheral Venous Catheter Insertion, IEEE Transactions on Visualization and Computer Graphics, October 2026. 10.1109/TVCG.2026.3747499 · doi.org (peer reviewed).

Just Out Tech explains new research in plain language. This article was drafted with AI assistance and checked by a human against the original source.