Send It and Move On: Why Batch Reporting Beats Watching the AI Work

A user asked us a good question recently. They send a case to RadPro, then sit and watch the impression come in, and they wanted to know if it could go faster. It can, a little. But the better answer is to stop watching. Send the case, open the next one, and come back later for a quick second look. That is how I use it every day. It saves me a lot of time, and the second look makes my reports better.

Why RadPro is not built to win a race

Plenty of AI reporting tools are tuned to start typing the moment you press the button. That looks fast in a demo. RadPro is tuned for the report you would actually sign. A case goes through several steps: fit a template if you dictated findings only, structure the findings without adding or dropping any, write the impression, then run a quality check for wrong sides, contradictions and findings missing from the impression. Each step gets room to think. On a complex CT that takes longer than a one-shot draft, and on a big pan-scan it is longer than anyone wants to stare at.

So we chose quality and accepted the wait. The trick is to make the wait cost you nothing, and the way to do that is to not be there for it.

How I actually read

I work in batches of five to ten cases. The loop looks like this:

1. Open the study, read it, and dictate the findings the way I always have. No formatting, no impression, just findings.
2. Send it to RadPro and press process.
3. Open the next study straight away. Do not look at the RadPro window.
4. Repeat until the batch is done.
5. Go back to the first case. The report is finished and waiting. Read the impression and the QA list, take a quick second look at the images where something is flagged, paste the report back and sign.

For the whole first half of the batch my head stays in one mode: look, find, say it. I am not editing headings, I am not wording an impression, and I am not proofreading a half-written paragraph that is still moving on the screen. RadPro organizes the report; my job during those minutes is to find things.

Watching the AI is the expensive part

Sitting through a run feels harmless because you are not doing anything. That is the problem. You are idle, then you are reading text as it arrives, then you are back in the images, and each of those is a switch. Switching between two kinds of task has a measurable cost even when the switch is expected, a result that task-switching research has shown for decades. Reading rooms already have plenty of switches we do not choose. A study of resident call shifts found that each extra phone call in the hour before a report raised the chance of a major discrepancy by 12%, and an eye-tracking study found that a phone interruption made radiologists spend longer on the case and look at the dictation screen instead of the images right after it. There is no reason to add a self-made one to every case.

Hopping back is easier than you think

People worry that coming back to a case means rebuilding it in their head. In practice it does not, for two reasons.

First, you leave at a clean break. You have finished the read and said everything you saw, so nothing is half done. Work on interruptions in computing found that being pulled away between subtasks is far less disruptive than being pulled away in the middle of one. The end of a dictation is the best break you will get all day.

Second, the second look is a different job from the first read. You are not searching the whole study again. You are checking a finished report against the images, with a short QA list telling you where to look.

The software does the bookkeeping. Each case gets its own tab in RadPro, so nothing overwrites anything. With RadPro Hub on Windows, one key sends the dictation over, and when you reopen a study in PowerScribe, RadPro switches to that study's tab by itself. Another key pastes the finished report back. Finished reports also stay in My Reports if you close a tab by mistake.

Where the time comes from

The saving is not in any one case. It is that the AI's time and yours now overlap. While you dictate case four, cases one to three are being structured, given an impression and checked. By the time you come back, the slowest part of the run already happened while you were doing something useful. Over a batch of eight, that is several minutes of watching you never do, and you never drop out of the flow of reading to do it.

Two practical notes. Batches of five to ten work for me; much longer and the first case is cold when I return, much shorter and the gain is small. And keep the urgent cases out of the batch: a stroke or a critical result gets read, finished and called on its own. If a study is simple enough that you would not use RadPro at all, a normal chest film for example, dictate it as usual between the others.

The second look is where quality comes from

The quick second scroll is the part I would keep even if RadPro were instant. RadPro is meant to improve your reports as well as your speed, and this is where it does.

On the first read you are hunting. On the second you have a finished report beside the images, a QA list pointing at anything that looks off, and a few cases of distance from your own dictation. A quick scroll through the study with that in hand catches what the first pass let through: the nodule you saw and never dictated, the measurement that belongs on the later series, the side that came out wrong, the impression that does not lead with what the referring doctor needs. You also read what is on the screen instead of what you meant to say, which is how a dropped “no” gets caught.

It takes a minute, often less. Most of the time nothing changes and you sign. When something does, you are glad you looked, and so is the patient.