No Lost Priors: The Imaging Migration Checklist Radiology Actually Needs
- Feb 15
- 5 min read
Radiology can adapt to new viewers, new workflows, and new vendors. What it can’t tolerate is opening a study and realizing the priors are missing, incomplete, or attached to the wrong patient. That’s when confidence drops, read times climb, and the help desk lights up.
This post turns “don’t lose priors” from a hope into a plan. It’s an audit-ready, workflow-first checklist for imaging data migration that treats DICOM + HL7 + patient identity + validation as one system because that’s how radiology experiences it.
Why “no lost priors” is the real success metric
A migration can be “technically successful” (data moved, new system live) and still fail clinically if:
priors are slow to retrieve,
studies are split across archives,
patient merges don’t carry over,
or completeness can’t be proven when someone asks, “Are we sure everything moved?”
“No lost priors” isn’t just a storage problem. It’s an availability + findability + correctness + proof problem.
The checklist
1) Define “priors” the way radiology uses them (not the way storage counts them)
Before you plan bandwidth or timelines, define what priors must be available for safe reading:
Day-one priors: last X months/years by modality and service line (ED/inpatient may need different rules than outpatient)
High-value comparisons: oncology follow-ups, neuro, trauma, breast, cardiac
Beyond basic images: key images, presentation states, structured reports, PDFs, and any non-DICOM attachments that clinicians expect to see
Tip: Write this down as a simple table: service line → modalities → “must-have” lookback window → exceptions.
2) Inventory every source that contains clinically relevant history
Most “missing priors” incidents are not caused by the application. They’re caused by missing a source system during discovery.
Include:
PACS, VNA, departmental archives (cardiology, ortho, GI), DR copies
cloud archives and long-term retention tiers
external ingestion pathways (CD imports, outside exams from image sharing, HIE sources)
HL7 feeds: ADT, orders (ORM), results (ORU), merges and identity change events
If your organization has grown through acquisitions or system swaps, assume there are “shadow” repositories until proven otherwise.
3) Lock down patient identity rules before moving a single byte
A perfect DICOM copy can still be “lost” if patient identity isn’t consistent.
Make identity explicit:
What is the mastering authority (MPI vs facility MRN rules)?
How are duplicates handled?
What is the policy for merges/unmerges and demographic corrections?
Then operationalize it:
Create an identity conflict quarantine lane (don’t let mismatches silently pass)
Decide what gets corrected upstream vs during migration vs post-cutover remediation
Reality check: If identity issues are unresolved pre-cutover, the migration becomes the moment they show up at clinical speed.
4) Choose a migration strategy that protects clinical operations
There are three common models. The right one depends on volume, time pressure, and clinical tolerance for “older history comes later.”
Phased migration
Move data by facility, modality, or time range
Pros: controlled risk, easier troubleshooting
Cons: longer project timeline, requires clear routing rules
Hybrid migration
Move high-value priors first; retrieve older studies on demand (or in the background)
Pros: reduces up-front load, speeds time-to-clinical usefulness
Cons: requires excellent monitoring and clear rules for retrieval
Big-bang migration
Move everything, cut over once
Pros: shorter calendar window
Cons: highest operational risk; “unknown unknowns” are more painful
Rule of thumb: The more uncertain your source data quality is, the more your plan should favor phased or hybrid approaches.
5) Normalize and reconcile DICOM intentionally
Vendor differences and historical quirks compound over years. During migration, they collide.
Plan for:
SOP class coverage (including SR, presentation states, secondary captures)
transfer syntaxes (compressed studies are common troublemakers)
tag consistency (patient/study identifiers, accession formats, referring physician, institution names)
“study split” prevention (routing rules that keep all series together)
Key idea: Don’t “normalize” by accident. Decide what is allowed to change—and document it—so clinicians aren’t surprised and auditors aren’t confused.
6) Treat HL7 as part of the migration (not a separate integration project)
Imaging isn’t just images. Orders, results, and patient movement events shape what appears in worklists and how priors are found.
Validate HL7 end-to-end:
ADT behavior (admit/discharge/transfer updates)
order messages (ORM) and results (ORU) behavior
ACK/error handling and retry logic
merges/unmerges patterns and traceability
Common pitfall: “Images migrated fine, but things don’t match the chart.” That’s usually identity + HL7 event handling, not storage.
7) Build monitoring like a production system, not a project spreadsheet
Migration isn’t a one-time copy—it’s a pipeline. Pipelines need observability.
Minimum operational telemetry:
throughput (studies/hour), queue depth, backlog age
error rate with categorization (source read, network, destination write, validation mismatch)
retry counts and “stuck” object detection
alerts with clear thresholds and escalation ownership
Then define an operating rhythm:
daily triage and remediation
re-run rules (what is safe to retry automatically vs requires review)
change control for routing/normalization updates midstream
8) Make validation auditable (this is where “no lost priors” becomes real)
If you can’t prove completeness, you can’t confidently say priors aren’t missing—especially months later when the source system is gone.
A practical validation package includes:
pre/post counts at patient → study → series → instance levels
exception reporting with dispositions (fixed, skipped with documented rationale, pending)
sampling for integrity (random + targeted high-risk modalities and known edge cases)
reconciliation for late arrivals and deltas during the migration window
Aim for evidence: When a clinician says, “I can’t find the priors,” you should have a workflow that answers: Was it migrated? Was it excluded by policy? Is it quarantined? Is it pending remediation?
9) Plan cutover like a clinical event
Cutover is not “flip the switch.” It’s a coordinated clinical change with patient safety implications.
Include:
freeze windows and “last-good” checkpoints
delta capture plan (what changes in the final days and how it’s applied)
downtime communications and contingency reading workflows
rollback triggers defined in advance (not “if it feels bad”)
Pro move: Define a small set of “sentinel workflows” (ED trauma, stroke, oncology follow-up) and test them as explicit acceptance criteria.
10) Don’t declare victory until stability is proven
The riskiest time isn’t always go-live day. It’s week two—when edge cases surface and teams are tired.
Post-cutover stabilization should include:
a defined reconciliation period
SLA for resolving missing priors and identity conflicts
final “proof package” (counts, exceptions, sampling outcomes, sign-offs)
lessons-learned updates to your runbook
The quick red-flag test
If you can’t answer yes to these, you’re at risk:
Can we prove completeness with multi-level counts and exceptions?
Do we have a repeatable remediation workflow (not one-off fixes)?
Are merges/identity changes traceable end-to-end?
Do we know exactly what will be available day one for each service line?
A radiology migration truly succeeds when clinicians no longer have to think about it. At UltraRAD Corporation, we get there by designing migrations that treat priors as a clinical dependency, validation as defensible evidence, and identity as the foundation so continuity is preserved, confidence is maintained, and go-live feels invisible to the people who matter most.
