Size from your modality mix, not from a vendor's per-bed rule
Storage quotes for PACS are usually built on a per-bed or per-study rule of thumb supplied by whoever is selling the array. That rule is almost always wrong for a specific hospital, because study size varies by an order of magnitude across modalities. A computed radiography chest study and a thin-slice CT angiogram are both one study in your volume count and nothing alike on disk. Two hospitals with identical study counts can differ by a factor of five in annual storage growth purely on case mix.
The only defensible method is to measure. Pull twelve months of completed studies from your RIS grouped by modality and body part, then pull the actual stored size of a representative sample of each group from the archive rather than trusting published averages. Multiply, sum, and you have a real figure for last year. Do this before you talk to anyone about capacity, because the conversation changes entirely once you can state your own number.
Then apply growth honestly. Growth in imaging storage comes from three separate places that hospitals tend to merge into one percentage: more studies, more slices per study as equipment is replaced, and new modalities or protocols. The second is the one that surprises people. Replacing a sixteen-slice CT with a modern scanner can multiply the data produced per examination without changing the examination count at all, so an equipment replacement plan is also a storage plan.

Inputs your sizing model needs before it means anything
- Twelve months of study counts grouped by modality and protocol
- Measured stored size per group, sampled from your own archive
- Planned equipment replacements and their expected data output
- New services or protocols committed for the coming period
- Your actual retention obligation, resolved rather than assumed
What retention genuinely obliges you to keep
Retention is where planning goes wrong most expensively, because hospitals either keep everything forever out of anxiety or apply a rule imported from a jurisdiction that does not govern them. Imported guidance is a particular trap here: retention periods written for other countries appear at the top of search results and have no application in India. Your obligation comes from Indian medical records requirements, professional conduct regulations, the terms of any accreditation you hold, and the limitation periods relevant to litigation and to your medico-legal exposure.
Resolve it properly and in writing rather than by inference. That means your medical records officer, your legal adviser and your clinical leadership agreeing a retention schedule, differentiated where the underlying rules differentiate — records connected to medico-legal cases, to minors, and to ongoing proceedings are commonly treated differently from routine studies. Write down the reasoning as well as the period, because in three years nobody will remember why the number was chosen and the schedule will quietly drift back to keeping everything.
Once written, the schedule has to be enforceable by your archive. A retention policy that no system implements is a document, not a control. Check that your PACS can actually express the rules you have written, apply them per study class rather than globally, and produce evidence of what was deleted and when. If it cannot, that is a procurement requirement for your next cycle and an interim manual process in the meantime.
Tiering the archive so the money follows the access pattern
Imaging access is heavily front-loaded. The overwhelming majority of reads on a study happen in the days and weeks after acquisition, after which access falls away to occasional comparison against a new examination. Storing a five-year-old study on the same fast tier as this morning's trauma CT is paying premium rates for data almost nobody touches, and it is the single largest avoidable cost in most imaging estates.
A conventional arrangement is a fast online tier holding recent studies and anything with a scheduled follow-up, a cheaper nearline tier holding the bulk of the history, and an archive tier for material retained only to satisfy the retention schedule. What matters more than the number of tiers is that the movement between them is automatic and rule-driven. Manual tiering is done enthusiastically for two months and then never again.
Two design points get missed. First, prior-study retrieval has to be prefetched against the appointment schedule, not fetched on demand, or your radiologists will experience tiering as the day their comparisons became slow. Second, retrieval cost from a cold tier is a real operating cost in cloud arrangements, and a prefetch policy that is too aggressive can quietly cost more than the storage it saved. Model retrieval, not only storage.

Decisions that make or break a tiered archive
- Which study classes are pinned to the fast tier regardless of age
- How priors are prefetched from the booking list rather than on demand
- Whether tier movement is automatic or depends on someone remembering
- Retrieval volumes and their cost, modelled alongside storage cost
- How a tier failure degrades: slow, read-only, or unavailable
Cloud, on-premise and the honest comparison between them
The cloud-versus-on-premise question for imaging is not the same as it is for a database, because imaging moves large objects and the network is usually the binding constraint rather than the storage itself. A hospital with a reliable high-capacity link and a workflow that tolerates a small retrieval delay can run a cloud archive comfortably. A hospital whose connectivity drops during monsoon and whose radiologists report from home over consumer broadband will experience the same architecture very differently.
Compare the two over a full replacement cycle rather than on year one. On-premise carries the array, the refresh at end of life, power, cooling, floor space, the second copy at another site, and the staff time to run it. Cloud carries subscription, egress and retrieval charges, the connectivity you have to upgrade to make it viable, and the migration cost when you leave. The commonest error is comparing a cloud subscription against on-premise hardware alone, omitting the refresh and the disaster-recovery copy, which flatters on-premise substantially.
Whichever way you go, treat data residency as a constraint on the shortlist rather than a detail to settle afterwards. Where patient data may be stored is governed by obligations that sit outside the imaging decision, and a provider whose region options do not satisfy them is not a cheaper option, it is not an option. Establish that before you evaluate anything else about them.
The exit problem nobody costs at purchase
Imaging archives are the stickiest systems a hospital owns. Years of studies accumulate in one vendor's environment, and the cost of moving them is invisible until the day you decide to. Migration is slow, has to be verified study by study rather than trusted, tends to lose non-image objects such as key-image notes and presentation states, and competes for the same network and staff you need for daily operations.
The protections are contractual and have to be secured at purchase, when you have leverage, rather than at renewal, when you have none. Establish that the data is yours, that it will be provided in standard form on request, that there is no per-study or per-gigabyte extraction charge, and that assistance during transition is included rather than quoted at the time. A vendor unwilling to write those terms is telling you something useful about the relationship you are about to enter.
Test the export before you need it. Ask for a sample extraction of a few hundred studies including some with annotations and prior comparisons, and try to load them somewhere else. Hospitals discover the gaps in their export path at the worst possible moment, and a two-day test at implementation is far cheaper than finding out mid-migration that presentation states do not survive the trip.
“We had budgeted the new archive to the rupee and had not budgeted a single day for getting fourteen years of studies out of the old one. That migration took nine months and it was the whole project.”
Protecting the copy you would actually recover from
A single archive, however resilient the array, is one site away from total loss. Imaging is also increasingly a target rather than a bystander, and an encrypted or deleted archive with no independent copy is not recoverable by any amount of goodwill. The distinction that matters is between redundancy, which protects against hardware failure, and an independent copy, which protects against everything else including a mistake made by your own team.
Set recovery objectives for imaging deliberately rather than inheriting whatever the array does by default. Ask how much imaging data the hospital could tolerate losing and how long radiology could operate without archive access, and be honest that the answer for a busy department is short. Those two answers drive the copy frequency and the recovery architecture, and they usually reveal that the arrangement in place was never designed against a stated target at all.
Then test recovery rather than assuming it. Restore a sample of studies from the independent copy on a defined schedule, time it, and record the result. Backups that have never been restored are a belief rather than a control, and imaging backups are particularly prone to silent failure because their size means problems surface only at the end of a long job that nobody watches.

Questions worth answering before you call the archive protected
- How much recent imaging could be lost without clinical harm
- How long radiology can operate with no archive access at all
- Whether the second copy is genuinely independent of the first
- When a restore was last performed and how long it took
- Who is authorised to delete at scale, and what stops a mistake
Reviewing the plan on a cycle instead of at purchase
The reason imaging storage runs out unexpectedly is that sizing is treated as a procurement activity rather than an operational one. The model built at purchase is correct on the day it is built and drifts from that point onward, as protocols change, equipment is replaced and services are added, and nobody revisits it until an alert fires at eighty-five per cent.
A modest quarterly review is enough to prevent that. Compare actual consumption against the model, restate the projected exhaustion date, and note any change in equipment or protocol that will alter the trajectory. Fifteen minutes, minuted, with the projected date carried into the capital planning conversation. The point is not precision; it is that the exhaustion date is a known number that moves visibly rather than a surprise.
Report it somewhere management already looks. Storage consumption alongside modality utilisation and reporting turnaround gives a coherent picture of the imaging service rather than an isolated IT metric, and a platform such as HealUDoc can carry those figures into the same operational reporting the rest of the hospital is reviewed against. Capacity that is discussed quarterly gets funded in a planning cycle; capacity that is discussed at eighty-five per cent gets funded in an emergency, at a worse price.


