Towards a Global Imaging Procedure Code Mapping Resource
Share |

Towards a Global Imaging Procedure Code Mapping Resource

Scientific Session:

Enterprise Imaging


David A. Clunie, MBBS, PixelMed


Though considerable progress has been made towards standardizing coded information in orders, worklists, images and structured reports for such things as anatomy, standardization of procedures codes, other than those used purely for reimbursement purposes, has been slow.

Most sites throughout the world still use locally defined codes for ordering and acquisition and reporting workflow. Though several international and national standard coding schemes are available, even where their use is mandated, they are frequently used only at boundary interfaces, and not within organizations. This leads to images that contain essentially site-proprietary and often incorrectly or incompletely encoded procedure information.

As applications for sharing imaging procedure codes proliferate, including use in registries (such as for radiation dose surveillance), quality measures, decision-making (including appropriateness criteria used during ordering), image display (hanging protocols), retrieval of priors, and report template selection, so too does the number of nominally "standard" coding schemes. RadLex and ACR Common being cases in point.

Recognition of the inevitability of the continued use of site or application specific or other "not invented here" coding schemes, what appears to be needed is a "mapping resource" that comprehensively and reliably describes an equivalence between concepts in different schemes.

The UMLS is one such general resource, but unfortunately it has not developed a mapping between imaging procedure schemes. For example, LOINC and SNOMED concepts that are apparently identical are not mapped in UMLS, and are assigned different codes. Nor are coding schemes such as RadLex included in UMLS. The UMLS infrastructure, history and tools do provide rich resources though, for applying similar principles to imaging procedures, as does the extensive literature on mapping concepts in other domains (such as that related to mapping to and from SNOMED).

This project is a pilot to establish the feasibility of, and design principles for, a comprehensive Imaging Procedure Code Mapping Resource (IPCMR).


In this project, both semantic and lexical mapping techniques have been applied to imaging procedure codes, using various pragmatic techniques of identifying synonyms and their lexical variants, as well more formal modeling against a dictionary of atomic concepts extracted by comprehensive manual review of existing international, national and local coding schemes. The comparison spans the gamut of procedure complexity from nominally straightforward CT diagnostic examinations through nuclear medicine procedures to complex interventional procedures. Publicly available schemes examined include SNOMED (international and various national extensions, including those from UK and Canada), LOINC, RadLex, ICD-10-PCS, HCPCS (including CPT-4), UK NICIP, Japanese JJ1017, Australasian BSF, amongst others.

Significant limitations of lexical matching between terms within some schemes have been encountered, particularly with respect to the inappropriate use of synonyms that fail to convey the subtle distinctions intended, or that assume implied semantics that are nowhere explicitly mentioned. Some schemes contain more formally modeled definitions, based on child-parent relationships and/or atomic concepts and such models have been compared with the evolving model extracted for this mapping process. No existing model was found to be entirely satisfactory.


Most schemes suffer from inadequate or absent definitions. This is true even if they are formally modeled, in that the atomic concepts used may be inadequate to fully define the concept implied by the text of the term as provided, or the atomic concepts themselves may be inadequately defined or inconsistently used.

Often schemes expect out of band information to be implicit, such as whether or not a particular modality or technology is used, as distinct from other schemes (or concepts) for which such absence appears to be intended to imply a more general term (i.e., one without specification of modality).

The use of textual (prose) definitions for procedure terms, beyond the term itself, is relatively rare. Though prose definitions represent a challenge for automated lexical extraction, at least, when present, they serve to inform a human reviewer as to the intended meaning.

These are not novel observations with respect to the definition of meaning in general or as applied to mapping of concepts in particular, whether in ontologies or thesauri. As applied to the procedure code mapping exercise though, there is apparently a need to accept that for many concepts, an exact mapping will not be possible, and that in some cases more "fuzzy" relationships may need to be captured, whose applicability may then be judged depending on the use case for the mapping.


The pilot mapping exercise has served to define most of the concepts that will be needed for a unified model of the meaning of imaging procedure codes, as well as to elucidate both the precise and fuzzy relationships necessary to support such a model. Relatively simple and well known lexical techniques suffice to recognize many concepts from their terms alone, even in the absence of formal or prose definitions. Errors and weak design (or lack of discipline in application) of some existing formal models, as well as errors in existing lists of synonyms, limit their reliability for re-use in a robust universal mapping, increasing the burden for an accurate "re-definition" of many concepts in existing source schemes.


  1. SNOMED.
  2. LOINC.
  3. RadLex PlayBook.
  4. UK NICIP.
  5. JJ1017.
  6. ICD-10-PCS.


  • Procedure codes
  • UMLS
  • RadLex
Career Center more
Membership Software Powered by YourMembership  ::  Legal
Membership Software Powered by YourMembership  ::  Legal