Medical data storage is where the bytes live and how long they stay: a chart row, a DICOM series, a genomic BAM, a backup you can actually restore. It is not a ranked list of clouds. It is not a PACS product page. It is not who owns the rules.
If you meant the lifecycle umbrella (capture → store → retire) → healthcare data management. If you meant who owns the rules → data governance in healthcare. If you meant PACS as the imaging library → PACS and DICOM. If you meant the archive that outlives one vendor → what is a VNA. If you meant moving a copy → HIPAA-compliant data transfer. If you meant security products → healthcare data security solutions.
PYCAD builds custom web DICOM viewers and imaging models. It does not sell object storage, a PACS, or a backup appliance. The old “top 10 solutions 2025” ranking on this URL (AWS, Azure, Salesforce Health Cloud, IBM Watson Health, Oracle…) was vendor theater. Salesforce is a CRM. IBM Watson Health was sold. Dropped.
Imaging is the size problem
An EHR row is small. A CT series is not. A chest CT is hundreds of slices; a multiphase abdomen or a cardiac cine is more; a digital-pathology slide is gigapixels. Genomics is a different large file. Wearable streams are small per sample and endless. Storage design starts with which of those you actually keep, not with a brand.
| Kind | What it is | Why the store cares |
|---|---|---|
| Chart | Notes, meds, labs, orders — the EHR | Must come back in a visit-length query. Size is not the problem. Identity and access are. |
| Image | DICOM series on a PACS / VNA. Sometimes a WSI | Terabytes per year in a busy department. Yesterday’s trauma CT has to open in seconds. Last year’s CT can wait 20. |
| Genome / research | BAM / CRAM, derived features, trial locks | Write-once, read-rarely. Different retention and a different lock than the chart. |
| Device / wearable stream | Waveforms, CGM, implant telemetry | High write rate. Most of it is not a diagnostic archive. Keep what the protocol named. |
A chart without the study is an incomplete record. That is why imaging storage is not an afterthought on the data-management page, and why this URL exists: the bytes, the tiers, the restore — not the lifecycle survey.
Three tiers, not ten logos
Hospitals already do this, whether they name it or not. Fast disk for what is being read. Cheaper disk for what might be read this year. Object or tape for what the statute says you keep.
| Tier | Job | Typical home | Fail mode |
|---|---|---|---|
| Hot | Today’s worklist. Open in seconds | PACS workbench, local cache, a ward viewer | Putting last decade’s studies on the same spindles. The worklist slows down. |
| Warm | Priors you will want this year. Open in tens of seconds | Nearline, a VNA, the same vendor’s “archive” shelf | A proprietary store you cannot leave. Neutrality is the VNA page. |
| Cold | Retention and disaster recovery. Open in minutes | Object store (on-prem or a BAA cloud), tape, an off-site copy | “We have S3” with no restore test. A bucket is not a recovery plan. |
Lifecycle policy is a number: how many days hot, how many months warm, how many years cold. Radiology priors are not the same number as a billing extract. Do not copy a consumer “intelligent tiering” slide and call it a clinical policy.
On-prem, cloud, hybrid — as a store, not a PACS SKU
The PACS page already has the deployment table (LAN speed vs OpEx vs two environments). This page is the storage consequence of that choice.
- On-prem. You buy the capacity and the people. Latency is a cable. Sovereignty is a room. Growth is a purchase order. Fits a shop that already runs a data centre and a rule that the pixels do not leave the building.
- Cloud object (with a BAA). You buy the request and the egress. Someone else patches the disk. A trauma CT that has to traverse a WAN is a latency number, not a feature. Egress on a bulk prior pull is a bill. AWS / Azure / GCP are the three hospitals actually use for this. They are not a ranked list and they are not “healthcare storage products.” They are object stores plus a BAA and a region.
- Hybrid. Hot studies stay local. Priors and DR sit in the object store. Two clocks to keep honest: the cache and the bucket. This is the usual adult answer, not a compromise.
There is no default winner. The model is a latency number, an egress number, and a statute — not a brand. The PACS-as-library walk (C-STORE, C-FIND, DICOMweb) stays on PACS and DICOM.
Retention, immutability, a restore you have tried
Keeping a study is not the same as being able to open it in five years. Private tags, a vendor database, a compression you cannot decode — those are how archives die while the disks are still spinning.
- Retention. Statute and medical-record policy, not a vendor default. Paediatric studies outlive adult ones in a lot of jurisdictions. Write the number down.
- Immutability / WORM. Ransomware that encrypts the only copy is a storage failure. Object-lock or an air-gapped copy is the control. A snapshot on the same array is not.
- Restore test. Restore a named study to a named viewer on a named day. “We have backups” is a sentence. The test is the programme.
- Format. Standard DICOM (and, for the chart, something the next EHR can read). A proprietary brick is a migration you will pay twice.
Moving a copy to a specialist is a transfer job, not a storage job — HIPAA-compliant data transfer. Who may keep the study is governance — data governance.
What this page is not
- Not a “top solutions 2025” ranking. Dropped AWS-as-#1, Azure Healthcare APIs, Google Cloud Healthcare API, IBM Watson Health, Salesforce Health Cloud, Oracle Health, Dell EMC, Hyland, Pure, Philips HealthSuite as a trophy table. Some of those are real products. None of them is “the” medical storage solution, and several are not storage.
- Not 5650. The umbrella (capture / store / retire, four components) stays on healthcare data management.
- Not 7929 / 7839. PACS language and VNA neutrality stay on those URLs.
- Not a PYCAD storage product.
If the missing piece is a viewer on studies that already have a home, that is the imaging piece. Case studies.