AI orchestration
In developmentAI in the read, with a physician deciding: where the portfolio is headed
Diagnostic AI is multiplying, and each tool wants its own integration, login, and result format. The LynxMediq direction is one layer between your imaging and those models. The first step, AI-assisted reading, is in development in PixelMediq now. Routing studies across many models is planned after it.
What exists in PixelMediq today
A user with the right permission can run an AI job against a study by hand, and the result is displayed for the reader to accept, ignore, or contradict. Nothing runs on its own, no accuracy figure is published, and no regulatory clearance is claimed.
What is still planned
A dedicated orchestration service that routes studies across multiple models, automatic trigger rules, and adapters for models hosted outside the deployment. PixelMediq lists all three as roadmap, and they are not available today.
Why orchestration
Stop managing fifty AI vendors one integration at a time
Point-to-point AI integrations do not scale. The goal is an orchestration layer where adopting a new model is a setting and not a project, and where clinical operations never bend around the tooling.
Design goals
- One interface for in-house and commercial models
- Routing by modality, body part, and indication
- Results normalized to DICOM SR, SEG, and presentation states
- A single review workflow regardless of vendor
- A consistent audit trail across every model
- Adding or removing a model without touching clinical workflow
Planned scope
From inference to a confirmed finding
Orchestration is intended to cover the whole path: routing the study, running the model, scoring the output, and getting it in front of the right clinician.
Model routing
Sending each study to the right model based on modality, body part, and indication, with rules you control.
Third-party integration
Connecting external AI vendors through one standard interface instead of a bespoke integration each time.
In-house model support
Running your own models next to commercial ones, on the same routing and review workflow.
Job monitoring
Tracking every inference job, its status, and its latency, with alerts when something stalls.
Confidence scores
A confidence score on each result, so the cases that need a closer look rise to the top.
Clinical review
Findings landing in the worklist for a physician to confirm, edit, or dismiss before they reach a report.
Auditability
A record of the model, version, inputs, and the reviewer who acted on every finding.
Model governance
Versioning, monitoring, and retiring models, and validating them on your own data before they go live.
A physician in the loop, by design
AI anywhere in the portfolio is decision support. It flags, measures, and prioritizes, and a licensed clinician confirms or overrides every finding before it reaches a report. The software does not diagnose on its own, and it keeps a clear record of who decided what.
Make AI part of the read, not a side trip
See what AI-assisted reading looks like in PixelMediq today, or talk to us about the orchestration roadmap.