Imaging data is a gold mine for those looking to create medical AI tools that accelerate crucial research initiatives. As the number of FDA‑cleared AI models built on imaging data continues to rise year over year, accordingly, imaging now accounts for a substantial share of digital health venture capital funding (85%).
At the center of this growth sits academic medical centers (AMCs), which are expected to support clinical care, multisite research and AI development—all while maintaining strict security and regulatory controls. Yet, much of the imaging infrastructure still in use today was designed for a very different era.
As traditional PACS architectures were optimized for clinical viewing and departmental workflows, instead of the kind of large-scale data reuse needed for AI training and other major research initiatives, CIOs and CTOs face mounting pressure to modernize imaging environments. And they must do so in ways that improve scalability and accessibility without compromising governance, compliance or trust.
In this blog, we’ll explore the challenges CIOs and CTOs face when modernizing medical imaging infrastructure for AI training and multisite research.
The Root Problem: Fragmented, Siloed Imaging Environments
In most AMCs, imaging infrastructure has evolved incrementally over many years, resulting in fragmented environments that are difficult to govern. Multiple PACS instances, sometimes from different vendors, store imaging data in ways that limit interoperability and secondary use.
This results in several systemic challenges:
- PACS metadata is difficult to search and extract at scale, which limits research reuse and long-term value
- Data inconsistency and modality-specific conventions make it hard to create reliable cohorts across systems
- Manual workflows for uploading, transforming and sharing imaging data introduce quality and compliance risks
- Data provenance is often incomplete or becomes unavailable once data leaves the PACS environment
These issues compound over time. Researchers copying imaging data into spreadsheets and shared drives is antithetical to strong data governance. Organizations lose visibility into who accessed what data, how data was modified and whether it was properly de‑identified. This fragmentation directly increases security risk and undermines reproducibility in research and AI development.
What Modern IT Leaders Are Prioritizing
Standards‑Based Architecture
CIOs and CTOs are increasingly relying on standards‑based interoperability to reduce complexity and technical debt. Using the DICOM format provides consistent structure for imaging data and metadata, while newer DICOM extensions support AI‑generated artifacts and derived objects.
This offer several advantages:
- Reduced reliance on fragile, custom integrations
- Greater portability across vendors and systems
- Faster onboarding of new modalities, tools and partners
- A clearer path to multi‑institutional and multi-site research and AI model training
Without aligning overall standards first, each new AI or research initiative the organization takes on becomes a one‑off project rather than a repeatable capability. This stymies the ability to scale research and AI efforts.
Automated Ingest and Automated De‑Identification
We know privacy risks are especially acute in medical research, but many research imaging workflows still rely on manual ingestion and de‑identification processes. This isn’t ideal because incomplete de‑identification of DICOM metadata and burned‑in pixel data can lead to PHI exposure in research datasets.
Automating your ingestion pipelines helps you address these risks by:
- Applying consistent de‑identification rules at scale
- Reducing undocumented human intervention
- Ensuring datasets are research‑ and AI‑ready from the moment they enter the environment
- Supporting evolving schemas without re‑engineering workflows
Automating ingestion not only helps support your compliance efforts with regulations such as HIPAA, GDPR and 21 CFR Part 11, it also helps you scale as imaging volumes grow.
Strong Data Governance and Provenance
Imaging data governance is inherently more complex than standard enterprise data governance. That’s due to the nature of this data—large file sizes, heterogeneous formats, and frequent transformations.
Automating your auditability and provenance tracking is critical for both compliance and scientific integrity. But ensuring reproducibility has become challenging in biomedical research due to poor documentation of data handling and analytical workflows.
Modern imaging data management with built-in governance should emphasize:
- Role‑based access control (RBAC) grounded in least‑privilege principles
- Immutable audit logs tracking access and modification
- End‑to‑end lineage, from ingestion through analysis and AI training
These controls help your organization demonstrate compliance while also improving confidence in downstream usage and AI outputs.
What Modern Imaging Infrastructure Looks Like
Given all of this, here’s what you should strive for in your imaging infrastructure:
- Centralized, vendor‑neutral storage that can connect to individual PACS
- Automated workflows for ingestion, quality control and de‑identification
- Standardized metadata that supports discovery and reuse
- Governance embedded into the data layer, rather than added after the fact
- Built-in security features like role-based access controls
- The ability to scale and perform well as your data volume increases
This approach allows you to balance centralized control with the flexibility needed to enable individual labs and research groups to move quickly without creating unmanaged risk at the enterprise level.
The Business Impact for CIOs and CTOs
Modernizing your imaging infrastructure can deliver measurable value for your organization beyond technical improvement. Institutions that have modernized their imaging infrastructure with Flywheel have reported:
- Lower long‑term integration and maintenance costs
- Faster onboarding of new data sources and research initiatives
- Development of cleaner, standardized datasets for AI development
- Stronger compliance posture and faster audit response
- Reduced duplication of effort through data reuse rather than recollection
In just one key example, a team at the University of Wisconsin-Madison team worked with Flywheel to migrate 50,000 datasets of chest X-rays and PCR test results from various sources to develop an AI model that achieved a diagnosis accuracy of 94%, outperforming experienced thoracic radiologists by nine percentage points . This data came from five hospitals and the National Institutes of Health (NIH), combining original DICOM-formatted images from PACS with relevant metadata from electronic medical records (EMR) and radiology reporting platforms.
Despite the difficulties that working with medical imaging presents, Flywheel empowers research teams to gather data from disparate locations and systems and make it actionable — all while maintaining data governance, security and compliance.
Imaging Modernization Is a Must-Have for Today’s AMCs
For modern AMCs, outdated imaging infrastructure is what stands in the way of achieving greater research goals, faster. Fragmented PACS environments and manual workflows can’t meet the demands of AI‑enabled research, large‑scale collaboration or heightened regulatory scrutiny. But a solution like Flywheel can work with existing PACS and cloud storage to unify and activate imaging data enterprise-wide while maintaining data governance.
Flywheel also offers:
- Automated de-identification, conversion, classification and QA at scale through Flywheel Gears, containerized algorithms that can be easily added to the platform
- Built-in security and governance features like access-controlled projects and role-based permission
- Audit trails, digital signatures and approval workflows ensure tracking and authorization
Automated de-identification, conversion, classification and QA at scale through Flywheel Gears, containerized algorithms that can be easily added to the platform
Built-in security and governance features like access-controlled projects and role-based permission
Audit trails, digital signatures and approval workflows ensure tracking and authorization
For CIOs and CTOs navigating imaging modernization, a purpose-built solution like Flywheel offers a tangible and cost-effective way to enable scale, reuse and trust as imaging data continues to spur innovative, life-changing research at AMCs.
Schedule a demo to get started with Flywheel.
