A hospital archive has to keep a study for many years and produce it in under a minute.

The chain in most hospitals runs: modality, PACS, vendor-neutral archive, archive tier, then a write-once copy. Each hop has a distinct job and a distinct failure mode.

• Modality. Generates DICOM objects and pushes them to the PACS. Study size and object count are decided here and cannot be renegotiated later.
• PACS. Serves the radiologist. Its storage is tuned for reading speed and is the most expensive tier in the chain. It is not an archive, though it is frequently used as one until it fills.
• VNA. Holds the durable copy in a form that does not depend on the PACS vendor.
• Archive tier. Bulk capacity, disk or optical or tape, with retrieval measured in seconds to minutes.
• Write-once copy. Retention-locked media or volumes that resist alteration and deletion inside the period.

The common design error is collapsing the middle. Sites that archive directly from PACS to bulk storage end up with an archive whose catalogue lives inside the PACS database, which makes the next PACS replacement a data migration rather than an application swap.

Study size is driven by matrix size, slice count, bit depth and whether the modality produces cine or whole-slide data. The ranges below are illustrative planning figures, not measurements from any specific installation, and real sites vary widely.

Now the worked example. Assume, purely as an illustration, a hospital running 120,000 imaging studies a year with this mix, and take a single planning size per modality from the middle of each range.

• DX: 60,000 studies × 30 MB = 1,800,000 MB = 1.8 TB
• Ultrasound: 25,000 × 90 MB = 2,250,000 MB = 2.25 TB
• MR: 15,000 × 250 MB = 3,750,000 MB = 3.75 TB
• CT: 18,000 × 500 MB = 9,000,000 MB = 9.0 TB
• Digital pathology: 2,000 × 2,000 MB = 4,000,000 MB = 4.0 TB

Annual total: 1.8 + 2.25 + 3.75 + 4.0 + 9.0 = 20.8 TB per year. Over ten years at flat volume that is 208 TB, and with a second protected copy, 416 TB of media.

Two things fall out of it. DX accounts for half the study count and under nine per cent of the volume, so any sizing model built on average study size badly misestimates both ends. And digital pathology, at 1.7 per cent of studies, contributes nearly a fifth of the annual footprint. A pathology programme that scales from 2,000 to 10,000 slides a year adds 16 TB annually on its own, which is close to doubling the whole archive's growth rate. That is a capacity decision that gets made in a clinical department, not in IT.

Retention is measured in years. Recall is measured in seconds. A radiologist reading a follow-up scan wants the comparison study on screen while the current one loads, and a retrieval that takes four minutes changes clinical behaviour: people stop asking for priors.

That tension is resolved with cache policy, not with a faster archive. Three patterns are common:

1. Time-window cache. Keep everything from the last N months on fast storage. Simple, predictable, and wasteful for modalities that are rarely re-read.
2. Prefetch on scheduling. When tomorrow's appointment list is known, pull the relevant priors overnight. This converts an unpredictable random read into a planned sequential one, and it is the single most effective change on most sites.
3. Patient-level pinning. Keep all studies for patients under active treatment resident, regardless of age.

Prefetch is what makes an optical or tape archive tier clinically viable. Without it, deep-tier retrieval latency lands on the radiologist. With it, most of the day's priors are already on disk before anyone sits down. Oncology and orthopaedic follow-up pathways are the strongest candidates for it, because that recall pattern is genuinely predictable.

The reason to run a vendor-neutral archive is not storage efficiency. It is that PACS contracts end. A VNA holds studies in standard DICOM with its own index, so replacing the reading application does not require moving petabytes or renegotiating access to your own history.

It also handles the untidy reality of multi-site estates: differing accession number formats, duplicated patient identifiers between a hospital and a clinic it acquired, and modality worklists that were configured by different people years apart. Tag morphing at the VNA is where that gets normalised. Doing it at the archive tier instead means rewriting objects that should be immutable.

DICOM's value over decades is that it is documented, widely implemented and self-describing: the pixel data carries its own acquisition parameters. An archive holding DICOM objects with intact headers can be read by software that does not exist yet. An archive holding a proprietary compressed derivative usually cannot.

Underneath, the durable copy needs to resist alteration. INCOM StorEasy WORM appliances present as NAS over NFS or SMB while writing to optical WORM media in the background, with MD5 checksums maintained at file level and offline copies held separately from the online cache. Optical media in this class is ISO/UDF, which matters for the same reason DICOM matters: a platter readable by any conforming drive is not dependent on the appliance that wrote it. Vendor documentation differs on capacity ceilings between publications, so we size from a configuration quotation rather than from a brochure figure. The mechanisms are covered on immutable and WORM storage and WORM storage appliances.

For study handover to patients, Rimage RX400 MedX systems integrate as a DICOM destination and write to write-protected USB media rather than discs, with published retention figures of up to ten years for the higher-grade stick and up to three for the base one. Those are manufacturer figures for the media, not a statement about your archive.

Retention periods themselves are a legal question. A multi-year clinical imaging obligation, and paediatric records that may run considerably longer, must be set by your own advisers. We build the enforcement, not the policy.

Comparable long-retention estates are discussed under government records and research and AI workloads. Start at the industries overview, or request an assessment with a modality breakdown from your PACS.