Practice management
How much storage does your imaging center need (and what will it cost in 10 years)
The question almost always arrives phrased the same way: “how many terabytes do I need?” And it is almost always the wrong question, because it assumes there is a final number. There is not. An image archive does not fill up: it grows every month and cannot be emptied, because the medical record has to be retained.
The useful question is a different one: how many terabytes will you produce per year, and what happens to that figure in ten years. That can be calculated, and it can be calculated from data you already have.
What a study actually weighs
The first thing you need is the size per modality. These are typical ranges, not constants: the same machine can double the size of a study depending on the protocol, slice thickness and how many reconstructions are saved.
| Modality | Typical range | What drives the size |
|---|---|---|
| Radiography (DR/CR) | 10 – 30 MB | Number of projections |
| Ultrasound | 30 – 200 MB | Saved cine clips |
| 2D mammography | 40 – 100 MB | Additional projections |
| Tomosynthesis (DBT) | 400 MB – 3 GB | Reconstructed volume |
| Computed tomography (CT) | 100 – 500 MB | Thin slices, contrast phases |
| Magnetic resonance (MR) | 50 – 250 MB | Number of sequences |
| PET-CT | 100 – 500 MB | Hybrid study, two series |
Note the asymmetry: between a radiograph and a tomosynthesis study there is a factor of one hundred. That is why the number of studies, on its own, tells you nothing about storage.
The mix matters more than the volume
Take a medium-complexity center with 3,000 studies per month. With a typical mix, the calculation works out like this:
| Modality | Studies/month | Volume/month |
|---|---|---|
| Radiography (45%) | 1,350 | ~27 GB |
| Ultrasound (25%) | 750 | ~60 GB |
| Computed tomography (15%) | 450 | ~135 GB |
| Magnetic resonance (8%) | 240 | ~36 GB |
| Mammography (7%) | 210 | ~31 GB |
| Total | 3,000 | ~289 GB |
That is about 3.5 TB per year. But the interesting figure is not the total, it is how it is distributed.
Radiography is 45% of the studies and only 9% of the bytes. CT is 15% of the studies and 47% of the bytes. Put differently: if the center doubles its radiography volume, the archive barely notices; if it adds a CT scanner or starts doing tomosynthesis, the projection has to be redone from scratch.
This is why two centers with the same number of studies can have archives that differ by an order of magnitude. When you request a storage quote, the vendor should be asking about your mix. If they only ask how many studies you do per month, they are guessing.
A 10-year projection is not a multiplication
This is where most budgets break. The temptation is to multiply: 3.5 TB × 10 = 35 TB. That figure always comes up short, for two reasons that act at the same time.
The first is obvious: study volume grows. The second is less obvious and weighs just as much: the same study gets bigger every year. A newer machine reconstructs thinner slices, saves more series and offers protocols that did not exist before. A chest CT today does not take up what one took up five years ago, even if the patient and the indication are the same.
Adding both effects, an 8% compound annual growth rate is conservative for an active center. Under that assumption, the 3.5 TB of the first year accumulate like this:
- Flat projection (no growth): 35 TB
- Projection with 8% compound annual growth: ~50 TB
Forty percent more. And that 40% does not show up in year ten: it shows up spread across the years, squeezing the budget a little more than forecast every single year. If you are going to size your own infrastructure, that is exactly the margin you need contracted from the start, because buying disk halfway through always costs more than buying it up front.
What reduces the figure without touching the image
Terabytes produced and terabytes stored do not have to be the same number.
Lossless compression. It typically reduces size between 2:1 and 3:1 depending on the modality, and the image the radiologist sees is bit-for-bit identical to the one that came off the machine. This is not the same as lossy compression, which does degrade the image and has no place in a diagnostic archive. Applied to the 50 TB in the example, effective storage drops to a range of 17 to 25 TB.
Deduplication. What is already stored does not get stored again. It matters more than it seems at centers with repeated studies and in series that share data.
With both, the conversation moves: it is no longer about how much the center produces, but about how much ends up being billed.
Why the cost should not grow at the pace of the archive
Even though the archive only grows, the cost does not have to follow the same straight line. It depends on an architectural decision: whether every study is charged the same, or charged according to what it actually demands.
A study from six years ago does not need to open in milliseconds, but it does have to be available. With automatic tiering, studies move on their own between a fast tier, a standard tier and a cold archive based on how long ago they were accessed. The result is that the old study — which is most of the archive — costs a fraction of a recent one.
The practical difference is that cost per study falls over time instead of staying flat. Without tiering, your 2036 bill is your 2026 bill multiplied by fourteen.
The costs that are not on the storage line
Two items that tend to show up after signing:
Downloading has a cost. Extracting images in bulk from the cloud generates traffic, and that traffic is billed. It is not fine print or an arbitrary charge from the provider: it is how the infrastructure works. What you should demand is that it be sized in the proposal from the start, rather than discovering it the day you need a large extraction.
Migrating the legacy archive is a separate project. If you are coming from another PACS, the terabytes you already have must be brought over, verified study by study, and confirmed complete. That carries its own cost, which does not belong on the monthly storage line.
How to calculate yours
With four pieces of data already in your system:
- Studies per month over the last year, broken down by modality. Not the total: the breakdown.
- The real average size of your studies, not the one in the table above. Your PACS can give you the average size per modality under your protocols, which is the only number that actually applies.
- Your growth rate, comparing this year against last. If you have three years of history, better.
- What is coming: a new machine, a new site, a large payer contract. Any of the three invalidates a projection made without accounting for it.
With that you have a projection you can defend before a committee, which is a different thing from having a number.
At NOVA Imaging we build this projection from each center’s real production figures before proposing anything, because a ten-year archive budget built on someone else’s averages is no use to anyone. The RIS/PACS platform page has the detail on how compression, deduplication and automatic tiering work, and cloud, on-premise or hybrid covers which of the three models suits you depending on where you want that archive to live.