On-demand priors when PACS is not the permanent archive
Retrieve the comparisons a radiologist needs at reading time—and use clinical context to reduce the burden of maintaining endless prefetch rules.
Illustrated workflow. Sources, rules, and destinations are configured for each organization.
Not every PACS keeps every prior indefinitely. Older studies may move to a vendor-neutral archive, a legacy PACS, lower-cost storage, or an enterprise repository after a retention window. The images still exist, but they are no longer in the system where the radiologist reads today. On-demand prefetch closes that distance before it becomes a delay at the workstation.
The traditional answer is a large rule table: if the new study is this procedure, retrieve these modalities over this lookback period, subject to these site and count limits. Rules are useful guardrails. But selecting the most clinically relevant comparisons is a reasoning task, and the number of exceptions grows quickly as protocols and naming conventions change.
Tiered retention
The reading PACS and the permanent archive may be different systems
Keeping a complete longitudinal record in high-performance PACS storage can be expensive or inconsistent with an organization’s architecture. A migration may leave historical exams in the old platform. A cloud transition may tier studies after a fixed period. An acquisition may create several archives whose records are queryable but not immediately available to the current viewer.
The result is subtle: the current exam reaches the worklist, but useful comparisons do not. A radiologist searches manually, waits for a retrieval, or interprets without context. Prefetch shifts that work earlier, using the new order as a signal that a patient is likely to be read soon.
DICOM mechanics
Use HL7 intent to start DICOM query/retrieve
RadFetch listens for the HL7 order, extracts the current exam context, and searches configured archives with DICOM C-FIND. It builds a bounded candidate portfolio using patient identity, lookback limits, and configured source rules. Selected studies are then requested with DICOM C-MOVE to the reading destination, where normal PACS ingestion and routing continue.
Starting at order time creates a useful head start. The archive can retrieve and transmit a large prior while the patient is being scheduled, scanned, or waiting for images to complete. Job history and retry controls make the background work visible instead of leaving the radiologist to discover a silent failure later.
Rule maintenance
Procedure matrices grow faster than clinical relevance
Keyword and modality rules can handle straightforward cases. A new CT chest should probably retrieve prior chest CT examinations; a mammogram needs mammography comparisons. Complexity appears when local descriptions change, outside studies use different labels, or a current clinical question makes a less obvious prior important.
A rigid matrix also has to encode exclusion decisions. The archive may contain hundreds of studies for a complex patient. Retrieving all of them wastes time and storage, but narrow rules can miss the one comparison the radiologist would have chosen. Every new procedure, site, and protocol adds another branch for someone to maintain.
Clinical context
Let AI rank priors by how a radiologist would use them
AI-assisted selection can evaluate the current order against the available prior portfolio and choose the most relevant studies within configured limits. For a CT chest, a prior CT chest and CTA chest may be useful while unrelated knee radiographs are not. For follow-up oncology imaging, anatomy, indication, modality, and temporal context can matter together.
This does not mean removing all constraints. It means moving brittle procedure-by-procedure choice out of a sprawling rule matrix. Administrators still define sources, time windows, maximum selections, and confidence thresholds. The AI handles the semantic ranking inside that safe operating envelope, reducing the routine rule-management burden as exam names and clinical scenarios vary.
The distinction is important: “thinking like a radiologist” describes the selection objective, not autonomous diagnosis. RadFetch is choosing which existing studies to retrieve for comparison. The radiologist remains responsible for interpreting the images.
Validation and control
Observe recommendations before allowing automatic retrieval
RadFetch supports a shadow workflow for AI prior selection. Recommendations can be recorded and reviewed without initiating a move, letting the organization compare selections with radiologist expectations and conventional rules. After validation, automatic application can be enabled with minimum-score and maximum-selection controls.
Keep an audit trail of the current exam context, available candidates, chosen priors, reason or score, move status, and retry history. Monitor false negatives as carefully as extra retrievals: an AI system that is impressively selective but repeatedly misses a critical comparison has not met the operational goal.
RadFuzion module
RadFetch brings the archive into the reading workflow on demand
RadFetch connects HL7 order intent to DICOM C-FIND and C-MOVE, with facility and global rules, bounded lookback periods, job history, retries, and optional AI-assisted prior selection. It is designed for organizations whose useful history outlives the online retention of the reading PACS.
The practical outcome is simple: relevant priors arrive where radiologists already work, without treating the high-performance PACS as the only permanent copy and without asking administrators to predict every comparison through rules.
See it in your workflow
Bring relevant priors back before the radiologist opens the case
RadFetch combines HL7-triggered query/retrieve, configurable limits, job tracking, and AI-assisted prior selection for hybrid and tiered archives.