Medical image cloud storage audits expose 44 TB of cold data

AdvancedUNO
13 Jun, 2026
Medical image cloud storage audits expose 44 TB of cold data

10 min read

The Reality of Hybrid Imaging Archives

  • The Operational Friction: On-premise storage arrays are buckling under exponential DICOM data growth, but full cloud lift-and-shift migrations stall due to high egress fees and clinical latency demands.
  • The Pragmatic Fix: A hybrid, tiered storage model that keeps active clinical studies on local SSDs while silently offloading historical archives to cold cloud object storage.
  • The Immediate Action: Query your PACS database to isolate studies older than 24 months, mapping their actual retrieval rates to draft an automated lifecycle policy.

The Production Reality of Medical Image Cloud Storage

Medical image cloud storage is often sold as a frictionless upgrade, but the actual migration of enterprise PACS is a messy, half-finished transition.

At 2:15 a.m. in a busy metropolitan emergency department, the system failure is rarely one of medical knowledge. A resident knows exactly how to read a CT angiogram to rule out an aortic dissection. The failure is almost always one of infrastructure. The resident clicks the patient's record, and the screen remains white. On the local storage area network (SAN), a decade of un-tiered historical chest films has filled the physical disk arrays to 99.4% capacity. The local DICOM router is thrashing, attempting to serialize a massive 1.2 GB multi-phase acquisition while simultaneously purging older studies to free up blocks. By the time the first slice renders, eleven minutes have passed.

Enterprise imaging software vendors frequently pitch cloud-native PACS as an immediate cure for this physical capacity crisis. In sales presentations, the transition is depicted as a swift, wholesale move to the cloud that instantly lowers total cost of ownership. In production, however, health systems find themselves operating in a hybrid state. They are caught between legacy on-premise hardware that cannot be turned off and cloud buckets that accumulate unexpected monthly costs. The transition is not a sudden leap but a slow, uneven migration where clinical safety and network bandwidth dictate what moves to the cloud and what must remain local.

The Hidden Plumbing of Tiered Cloud Objects

To understand why wholesale cloud migrations stall, one must look at how clinical data actually behaves. Attempting to run a modern radiology department on an un-tiered legacy SAN is like trying to run a major metropolitan library by keeping every single book ever written stacked directly on the front checkout desk.

The vast majority of stored medical images are never looked at again. In a representative clinical system, once an imaging study passes the 90-day mark, the probability of a physician retrieving it drops below 10%. By the time a study is two years old, it is essentially dormant. Yet, regulatory requirements and patient care continuity demand that these files remain accessible for years, sometimes decades.

The Architecture of a Silent Tiering Policy

A clear example of this storage reality can be seen in the production data of King Hamad University Hospital (KHUH) and the Bahrain Oncology Center. Managing a 600-bed facility, the hospital accumulated approximately one million medical image studies by January 2022. This translated to 476 million individual files totaling 44 TB of data, with the volume expanding by 1 TB of new data every single month.

When engineering teams analyzed the retrieval patterns of this massive archive, they uncovered a stark disparity. Out of those one million studies, only 94,000 older studies (spanning 2011 to 2019) were retrieved during the entire year of 2021. Those 94,000 retrieved studies represented just 3 TB of data. In other words, only about 7% of the historical archive was ever touched. The remaining 93% of the data sat completely idle, occupying expensive, high-performance local storage.

King Hamad University Hospital PACS Data Access Patterns (2021)
Inactive Historical Studies (2011-2019) — 93%Retrieved Active Studies — 7%

Figures compiled from the sources cited below.

Instead of executing an expensive, risky lift-and-shift of their entire PACS, the hospital built a long-term storage solution on AWS. They kept their existing on-premise PACS interface intact for active diagnostic workflows but routed the cold, historical 93% of their data to AWS. This hybrid approach allowed them to reduce their storage costs by 40% without disrupting the daily work of their clinical staff.

"In clinical production, the most elegant cloud architecture is completely worthless if a trauma surgeon has to wait more than four seconds for a CT scan to render."

A Four-Stage Blueprint for Hybrid PACS Migration

Transitioning to a hybrid cloud PACS requires a systematic approach that prioritizes network bandwidth and clinical safety.

  1. Audit the Archive and Map Retrieval Horizons: Query your PACS database to extract the date of study, modality type, and last-accessed timestamp. Group this data by department to identify the exact point where retrieval probability drops below 5%, which typically occurs between 180 and 360 days post-acquisition.
  2. Deploy an On-Premise Edge DICOM Router: Install an open-source or enterprise DICOM router (such as Orthanc or dcm4chee) on-premise to act as an intelligent gateway. Configure this router to intercept all incoming studies from modalities like CT, MRI, and ultrasound, caching them locally on high-speed SSDs for immediate clinical review.
  3. Establish Automated Cloud Tiering Rules: Set up an object storage bucket in a cloud environment, such as AWS S3 or Azure Blob Storage. Configure lifecycle policies to automatically transition studies from S3 Standard to S3 Infrequent Access at 90 days, and then to a cold archive tier like Glacier Instant Retrieval at 360 days, ensuring the metadata remains searchable.
  4. Integrate HL7 Prefetching Triggers: Connect your cloud storage broker to the hospital's HL7 feed, listening specifically for ADT (Admit, Discharge, Transfer) and SIU (Scheduling Information Unsolicited) messages. When a patient is scheduled for an appointment or admitted to the emergency department, the broker automatically triggers a prefetch request to pull their historical imaging files from cold cloud storage back to the local cache before the clinician even opens the chart.

Choosing Your Storage Architecture: Enterprise SaaS versus Hybrid Object Tiering

  • Cloud-Native SaaS PACS (e.g., Change Healthcare Stratus Imaging, Visage 7 Cloud): This approach offers fast feature deployment, advanced 3D rendering performed server-side, and built-in machine learning workflows (such as Nvidia and King's College London privacy-preserving federated learning models). The trade-off is complete vendor lock-in, high monthly subscription fees, and an absolute dependence on high-speed WAN connectivity.
  • Hybrid Object Tiering (e.g., AWS S3 with on-premise gateway, Azure Blob Cold Tier): This model leaves your existing PACS software completely unchanged while silently offloading historical archives to low-cost cloud buckets, delivering immediate cost reductions of up to 40%. The catch is that it requires internal engineering resources to write and maintain the routing scripts and manage the local cache.
  • Open-Source DICOM Archives (e.g., Orthanc, dcm4chee, PACS Integration Tools): These tools provide ultimate customization, zero licensing fees, and direct control over your database schema, making them excellent for research networks and specialized clinics. The catch is the lack of formal enterprise support SLAs, which places the entire burden of high availability and HIPAA compliance on your in-house DevOps team.

Why Do Specialized Ophthalmology PACS Workloads Fail in the Cloud?

While radiology has standard methods for managing large image files, specialized clinical departments often struggle during cloud migrations because their data structures are fundamentally different.

The global ophthalmology PACS market is expanding rapidly, driven by an aging population and the rising prevalence of diabetes-related eye diseases like diabetic retinopathy and age-related macular degeneration. Ophthalmic imaging relies heavily on specialized modalities such as fundus cameras, optical coherence tomography (OCT), and visual field analyzers. These devices generate thousands of small, highly detailed files rather than a single, sequential series of large slices like a CT scan.

When IT teams attempt to migrate ophthalmology PACS to the cloud using the same methods they use for radiology, they run into three major technical roadblocks:

  • The Small-File Latency Tax: An OCT study often consists of hundreds of individual high-resolution B-scans. If the cloud storage architecture is not optimized for high-frequency, low-payload requests, the latency overhead of establishing a secure connection for each individual file can cause the clinician's viewer to freeze.
  • Metadata Fragmentation: Ophthalmic devices from different manufacturers often write proprietary metadata into non-standard DICOM fields. When these files are uploaded to generic cloud object storage without an indexing tool like AWS HealthLake to parse them, the proprietary metadata becomes unsearchable, breaking clinical search filters.
  • Bandwidth Choke Points in Community Clinics: Unlike centralized radiology departments in major hospitals, ophthalmology clinics are often distributed across smaller, suburban satellite offices. These locations rarely have dedicated fiber connections, meaning that uploading heavy raw OCT volumes directly to the cloud can saturate local office networks and disrupt other clinical systems.

The Operational Pitfalls That Drain Migration Budgets

  • Ignoring the Egress Fee Trap: Cloud storage providers make it cheap and easy to upload data, but they charge fees when you pull that data back out. If your prefetching logic is poorly configured—causing the system to repeatedly pull large historical MRI series down to local workstations—your monthly cloud bill can quickly exceed the cost of the old on-premise hardware.
  • Neglecting Diagnostic Viewer Validation: Under FDA regulations, diagnostic viewers used by radiologists must be validated for clinical use. If you migrate your PACS to a cloud-native storage tier without verifying that the viewer software maintains its FDA Class II cleared rendering speeds over a standard WAN connection, you risk falling out of regulatory compliance.
  • Treating HIPAA Security as a Cloud-Only Responsibility: Many healthcare organizations assume that signing a Business Associate Agreement (BAA) with a cloud provider like AWS or IBM means their data is automatically secure. In practice, if your on-premise DICOM router is misconfigured with default passwords or open ports, your entire archive remains vulnerable to ransomware, regardless of how secure the cloud bucket is.

Frequently Asked Questions

What happens to diagnostic viewer performance when our local WAN link drops to 50 Mbps during a cloud-native PACS session?

If you are using a progressive-loading viewer, the initial image slices will render quickly at a lower resolution, sharpening as the remaining data streams in. However, if your viewer requires the entire DICOM file to download locally before rendering, a drop to 50 Mbps will cause noticeable lag, stretching a standard 1.2 GB CT scan load time to over three minutes and stalling clinical workflows.

How do we maintain HIPAA compliance and FDA Class II diagnostic clearance when routing DICOM files through an intermediate AWS S3 bucket?

You must ensure that all data is encrypted both in transit (using TLS 1.3) and at rest (using AES-256 with customer-managed keys via AWS KMS). Additionally, you must sign a Business Associate Agreement (BAA) with AWS and verify that your DICOM routing software is certified by its manufacturer to run on cloud infrastructure without altering the pixel data used for diagnostic decisions.

Why do ophthalmology OCT images take twice as long to render in our cloud PACS compared to standard CT slices?

This is due to the difference in file structure. A CT scan is typically stored as a single, multi-frame DICOM file or a few large files, which can be streamed efficiently. An OCT study often consists of hundreds of individual, single-frame files. This requires the viewer to make hundreds of individual network requests, creating a latency bottleneck that slows down the overall render time.

How do we handle DICOM UID mutations when migrating historical archives from an on-premise vendor to a cloud-native system?

You should never allow your migration tools to mutate the original Study Instance UID, Series Instance UID, or SOP Instance UID. If a legacy PACS vendor used non-standard UIDs that cause collisions in the cloud, you must deploy a metadata translation broker that maintains a secure, indexed mapping table, preserving the original UIDs for clinical audit trails and regulatory compliance.

The most reliable way to handle a PACS storage crisis is to stop treating the cloud as an all-or-nothing destination. Begin by setting up an on-premise DICOM router with a local cache, and configure automated lifecycle rules to move studies older than one year to cold cloud storage. This hybrid approach keeps your active clinical workflows running at local speeds while immediately relieving the pressure on your physical storage arrays.

Engineering References & Signals

This guide is synthesized directly from active engineering signals and the reporting within the Source Data above.

  • Change Healthcare Stratus Imaging: Cloud-native SaaS PACS solution for radiology, offering advanced image analysis for CT and MRI scans, integrated with AWS HealthLake and collaborating with Nvidia and King's College London [1].
  • King Hamad University Hospital (KHUH) AWS Migration: A 600-bed hospital managing 44 TB of data across 476 million files, utilizing AWS to build a long-term storage solution that reduced storage costs by 40% by tiering inactive historical studies [2].
  • Ophthalmology PACS Market Dynamics: Specialized imaging segment managing fundus cameras, OCT, and visual field analyzers, facing increased demand due to diabetic retinopathy and age-related macular degeneration [3].

Related from this blog

Sources

Next Post Previous Post
No Comment
Add Comment
comment url