Prism
Algorithm management platform for radiology
An on-premises platform that lets hospitals run radiology AI from any vendor on their own servers. The hospital picks the algorithms. Prism handles routing, pseudonymization, monitoring, and compliance. Patient data never leaves the hospital.
What is Prism
Prism is a program that talks to PACS and controls different radiology AI algorithms. The user decides which algorithms run in the system, on which studies, and how to measure whether they actually help. Prism is really just the plumbing needed to make it happen.
Use any algorithm that meets a few simple guidelines
If an algorithm runs in a container, it can run on Prism. A newly acquired algorithm can be registered in the system in a few minutes — no integration project needed.
See if an algorithm actually works
Access algorithm evaluation in a uniform, systematic way. Sensitivity, specificity, and predictive values are calculated automatically per algorithm. Evaluations can be broken down by hospital, patient group, and/or clinician.
Compare AI to reality
Prism is able to read radiological reports and conclusions and extract what the radiologist actually found. That becomes the truth the algorithm is measured against. All of this is done locally and without network access.
Define which algorithms evaluate which studies
Configured rules determine which studies are routed to which algorithm. It is possible to filter by modality, body part, or hospital, and run a randomized split for objective comparison.
EU AI Act and GDPR tracking
Each algorithm has a compliance page with CE marking, risk class, intended use, data processing details, and sub-processor info. Everything can be exported when reporting is needed.
Run a controlled trial before full rollout
Prism can randomize which studies receive AI assistance and which do not. Afterwards the two groups are compared — making it possible to document whether the algorithm actually improves clinical decisions before rolling out broadly.
Use cases
Evaluate on historical data
A new algorithm can be run on past studies from PACS without touching clinical workflow. Performance is measured against what the radiologist actually reported. So it’s clear what’s being deployed before it reaches patients.
AI in daily radiology
Chest X-ray triage, fracture detection, mammography screening — studies flow from PACS, get processed, and results come back as DICOM Structured Reports and HL7 messages to RIS.
Keep tabs on AI over time
Algorithms can degrade over time. Prism watches for drift, measures automation bias, and collects clinician feedback continuously. Notifications go out when something changes.
How it works
A study arrives from PACS. Prism strips patient identifiers, sends it to the right algorithm in an isolated container, and delivers results back to PACS and RIS. The algorithm never sees real patient data and has no network access.
PACS
AGFA Enterprise Imaging
Prism
Pseudonymize & route
Algorithm
Isolated container
Results
DICOM SR + HL7 to RIS
For algorithm partners
If an algorithm runs in a Docker container, it works with Prism. The vendor ships the container. Prism handles PACS integration, pseudonymization, monitoring, and compliance tracking.
Minimal interface
If the algorithm runs in an OCI container, most of the interface is already satisfied. Beyond that, the algorithm reads DICOM files from /input, writes results to /output, and exits with code 0. That’s all that’s required.
Compliance is handled
EU AI Act metadata, GDPR data categories, sub-processor agreements — all tracked in the compliance dashboard. As an algorithm developer, details only need to be filled in once.
Built for regulated healthcare
Clinical NLP, LDAP/AD authentication, GDPR Article 28 DPA handling, EU AI Act tracking. Designed for healthcare environments where compliance is non-negotiable.
The choice is yours
Reseller, pass-through, or direct licensing. Your choice, Prism is flexible.
Data handling
On-premises, air-gapped
Everything runs on the hospital’s own VMware infrastructure. Installation works offline from container bundles. Algorithms run with no network access. No cloud, no callbacks, no telemetry.
EU AI Act
Compliance dashboard per algorithm. CE marking, risk class, intended purpose, human oversight requirements. Export reports for Article 72 post-market surveillance.
GDPR
Patient data is pseudonymized before algorithms see it. Studies are auto-deleted after 30 days. Full audit trail. DSAR export built in.
No lock-in
Standard DICOM and HL7 protocols. Works with AGFA EI and Philips WIM. Algorithms are plain Docker containers — they work the same way on any platform.
Technical details
Talk to us
See Prism in action or ask questions about setup. Send a message.