Priors on screen

Ask a radiologist what they do first when a follow-up CT opens, and most will tell you the same thing. They look for the old one. The current scan tells you what is there; the prior tells you what has changed, and very often that is the actual question. Yet in many Indian centres the prior is missing, or it arrives several minutes after the radiologist has already started dictating, and the report quietly becomes a description of one scan rather than a comparison of two.

Our co-founder Dr. Vishwanath R S, a practising radiologist, keeps one question near the top of his list for any radiology software vendor: "How fast are priors on screen?" (it is one of the questions in The real total cost of radiology software). It sounds like a performance question. It is really a clinical one. So, let's look first at what a prior does for the read, then at why priors are slow or missing in practice, and finally at what good prefetching looks like.

What a prior actually does for the read

A 6 mm lung nodule on a first CT is a problem to be worked up. The same 6 mm nodule, unchanged on a scan from three years ago, is a different conversation. A sclerotic focus in a vertebra that was there in 2021 is probably a bone island. A new one in a patient with a known primary is not.

The best evidence for this comes from mammography, where comparison has been studied in large numbers.

A reader study in Clinical Radiology, covered in 2024, had the same radiologists read the same cases with and without priors. Access to priors reduced false positives in both dense and non-dense breasts, regardless of reader experience.

The 2011 paper is worth reading in full, because it is honest about the trade-off. Sensitivity in that study was lower in the comparison group, and the authors looked closely at how readers responded to perceived change. A prior is not magic. It is information, and readers still have to use it well. But fewer women recalled for something that was there and stable years ago is a real gain, for the patient and for the centre.

Outside mammography, the logic is the same even where the numbers are thinner: nodule follow-up, post-operative change, treatment response in oncology, a chest X-ray in an ICU patient on day three. In each of these, the report without the prior is a weaker report. The same applies to the prior report, not just the prior images; knowing what the last radiologist measured, and how they described it, is often what makes the comparison quick.

Why priors are slow in practice

If priors matter this much, why do radiologists in so many Indian centres still read without them, or wait for them? In our experience the causes fall into five groups.

1. The prior was done somewhere else. The patient had their last CT at another centre, or at another branch of the same chain that runs a different PACS. Nobody can fetch it, so the radiologist reads "no prior available" and moves on, or the patient is asked to bring a CD next time.

2. The patient ID does not match. Registration typed "Ramesh Kumar" last year and "Ramesh K" this year, with a new MRN because the old one could not be found at the front desk. The PACS is not going to link those two on its own.

3. The names do not match either. One scanner calls the series "CT CHEST PLAIN", another calls it "Thorax 1.0 B70f". A rule that looks for "chest" priors misses half of them. Herman Oosterwijk makes a closely related point about hanging protocols in his May 2026 article What's beyond PACS 3.0?: inconsistent study and series descriptions and body part encoding from other facilities and manufacturers are what break them. (The header attributes involved are the same ones we discuss in DICOM headers and AI routing.)

4. The prior is in slow storage. Many on-premise setups push older studies to a cheaper archive tier. Retrieving them can take minutes, sometimes longer if the archive is a separate box in a server room nobody has looked at since installation. (We compare storage set-ups in Cloud PACS or on-premise.)

5. Nobody asked for it in advance. The PACS fetches the prior only when the radiologist opens the case. Now the radiologist sits and waits, or opens the next case and comes back, and loses the thread.

Multiply that wait across a day's worklist and it is easy to see why people stop bothering. Note that only the last two are performance problems; the first three are data problems, and no amount of faster hardware will fix them.

What good prefetching looks like

Prefetching means the system gets the relevant priors ready before the radiologist asks for them. It is not a new idea. In the same article, Herman lists "intelligent prefetching" of priors among the areas where the industry now expects AI to help. But the basic version should already be standard. A good prefetch set-up is characterised as follows.

1. It triggers early. The moment a study is scheduled or registered, not when it is opened for reporting. A patient booked for an MRI brain at 4 pm should have last year's MRI brain sitting in fast storage by 3:55.

2. It knows what "relevant" means. Same body region, sensible modalities, a sensible time window. For a CT chest follow-up, the last two CT chests and the most recent chest X-ray, not every study the patient has ever had.

3. It does not depend on exact names. Matching on body part, modality and anatomy, with a mapping table for each centre's series naming, works far better than string matching on descriptions.

4. It reconciles patient identity. Matching on more than one identifier, and flagging likely duplicates for someone at the front desk to merge, so the prior is actually found.

5. It reaches across sites. If the patient has been to any centre on the same system, the prior is there.

6. It brings the prior report too. The images without the previous report leave the radiologist re-measuring what someone already measured.

7. It is visible. The worklist should show that a prior exists and is ready, so the radiologist knows before opening the case.

The last point is underrated. "Prior available" on the worklist changes how a radiologist plans the next hour.

Priors across a network

What this means for a centre

If you want a quick sense of where you stand, do this for one week:

1. Count how many reports say "no prior available" when you know the patient has been scanned before.

2. Time how long it takes from opening a case to having the prior on screen. Do it at 11 am and at 9 pm.

3. Pull ten patients with duplicate MRNs and see whether their old scans are linked.

4. Ask your PACS vendor exactly when prefetch is triggered, and what rules decide which priors are fetched.

The answers usually make the case on their own. In conclusion, a prior on screen is not a nicety; it is part of the read, and the evidence from mammography shows what is lost without it. Most of the fix is unglamorous: one patient identity, consistent naming, early prefetch and one archive across sites. Faster software helps, but only after those are in place.

Related reading

• Hanging protocols, explained — why the same naming problems break layouts as well as prior matching.
• Running a diagnostic chain on one system — one patient index and priors across the network.
• DICOM headers and AI routing — the attributes prefetch rules depend on.
• The collaborative DICOM viewer — reviewing a current study and its prior with a colleague.

FAQ

Why do radiologists need previous scans?

Comparison shows what has changed. A stable finding over years is usually benign, while a new or growing one needs action. In screening mammography, studies have found recall rates roughly halve when prior mammograms are available.

What is prior prefetching in PACS?

It is the PACS fetching a patient's relevant previous studies into fast storage before the radiologist opens the current one, usually triggered when the exam is scheduled or registered.

Why are priors missing even when the patient was scanned before?

The usual causes are a different patient ID at registration, a scan done at another centre or on a separate PACS, inconsistent series naming between scanners, or older studies moved to a slow archive.

Can priors from another centre be shown automatically?

Yes, if both centres share one system or exchange studies with a common patient identity. Centres on separate PACS installations usually need a manual request.