One local language for every study description

Vocabulary mismatch

Outside systems name exams for their workflow—not yours

Procedure descriptions evolve from billing catalogs, scanner protocols, scheduling conventions, legacy character limits, and local preference. They are rarely designed for exchange. When the exact string becomes the key for downstream logic, each outside source creates new exceptions.

Radiologists absorb some of that variation cognitively, but it slows work and makes the viewer less predictable. Hanging protocols may not recognize an unfamiliar description. Routers may miss keyword rules. Analytics may split one procedure across a dozen labels. Normalization gives those systems a stable local term while preserving source identifiers where they are needed for audit.

PACS migration

Do not carry every historical description inconsistency into the new PACS

A migration combines years of naming history with a new destination catalog. Simple find-and-replace tables handle exact known terms, but historical data often contains spelling variations, punctuation changes, site prefixes, retired codes, and free-text descriptions. Mapping each distinct string manually can become a major workstream.

RadMatch can load the organization’s target code set, apply exact matches and persistent overrides, then use local machine-learning and similarity methods to propose the best local description for the remaining terms. Lower-confidence cases enter a review path. Confirmed mappings become durable knowledge that can be reused as the migration continues.

Normalization should be staged and auditable. Test representative samples from every source, preserve the original value where policy requires it, and validate the effect on hanging protocols, routing, priors, and reporting. A confident string match is still part of a clinical data migration and deserves the same change control as other tag transformations.

Patient media and exchange

Make outside imaging look familiar when it reaches the local PACS

A patient may arrive with a disc whose studies are perfectly valid DICOM but use unfamiliar descriptions. An exchange gateway may import exams from dozens of affiliates. The clinical content is useful; the naming inconsistency is not. Translating the description during import lets the radiologist see the exam labeled as it would have been acquired locally.

This improves more than aesthetics. The local name can help the viewer select a hanging protocol, make a prior easier to find, and keep downstream worklists or analytics from accumulating another source-specific category. Saved term replacements also let administrators correct recurring language—for example, a facility-specific abbreviation—without retraining a model or editing every object by hand.

Choose the right layer

Normalize at the HL7 level, the DICOM level, or both

For prospective workflows, RadMatch can apply a normalized procedure description to the HL7 order before other automation uses it. That gives worklist, routing, and prefetch logic a consistent term early in the lifecycle. It can also rewrite the DICOM Study Description as images pass through RadFuzion so the receiving PACS displays the local name.

Migration and patient-media imports may have no useful order message, making DICOM-level normalization the natural choice. A live multi-system workflow may benefit from both layers so the order and images agree. The implementation should define which identifiers remain authoritative and avoid changing codes that other systems use for reconciliation merely to improve display text.

Human control

Automate the clear matches and review the ambiguous ones

A normalization system should know when it is uncertain. RadMatch scores candidate matches and can route lower-confidence results to a review queue instead of silently choosing. Administrators can accept a suggestion, select another local term, or create a persistent override for a source phrase.

Local exact matching and saved decisions run before more flexible methods. ML-assisted similarity can handle the long tail, with an optional language-model fallback where configured. This layered approach reserves more expensive or less deterministic processing for cases that need it while keeping common, approved translations stable.

RadFuzion module

RadMatch turns many source vocabularies into one operational vocabulary

RadMatch uses uploaded local code sets, exact and persistent mappings, local ML, confidence thresholds, and review to normalize study descriptions. It can apply the result to HL7 orders, DICOM Study Description, or both, depending on whether the workflow is a live interface, teleradiology feed, migration, or outside imaging import.

The benefit is consistency where users actually feel it: a prior imported from another facility can be labeled the way the local radiologist expects, and a decade of migrated descriptions can support the same downstream automation as newly acquired exams.