by Michael Bohl
Bookmark

Streamlining cmbX Mapping in the ACR Dose Index Registry

Background

On February 17, 2026 the American College of Radiology (ACR) announced an important enhancement to the Dose Index Registry (DIR): the introduction of a “combination study” designation, identified by the cmbX prefix in Study Descriptions.

This change represents a meaningful improvement in how the DIR handles multi-part or combined CT exams.

  • cmbX studies can now be mapped to RadLex combination RPIDs when appropriate
  • If no suitable combination RPID exists, the DIR recommends assigning RPID88
    • Assigning RPID88 removes the study from “Unmapped” status but, importantly, excludes them from your dose benchmarking metrics

The Challenge

While this update is beneficial, it introduces a significant operational burden:

  • Facilities may have dozens to 600+ cmbX studies requiring review
  • Each study must be:
    • Parsed into its component exams
    • Interpreted clinically
    • Matched to a valid RadLex combination RPID, if one exists

This is a manual, time-intensive process, particularly for organizations with busy CT departments.

A Practical Solution Using AI

To address this, I developed a repeatable, AI-assisted workflow that dramatically reduces the time required to map cmbX studies.

Approach Overview

  1. Input Data
    • Full RadLex Playbook
    • Facility-specific list of cmbX Study Descriptions
  2. Custom Instruction Set
    • Defines how to:
      • Parse cmbX descriptions (split into components)
      • Normalize anatomical regions
      • Interpret contrast usage
      • Compare against RadLex RPIDs
      • Identify valid combination matches
  3. AI Output
    • Suggested RPID mapping for each cmbX study
    • Confidence level assigned to each match:
      • High
      • Medium
      • Low
      • Not Matched

Real-World Performance

Based on repeated use across multiple facilities:

  • High Confidence Matches
    • ~90%+ accuracy
    • Typically safe to accept with minimal review
  • Medium Confidence Matches
    • Variable accuracy (~40–60%)
    • Require targeted review
  • Low Confidence Matches
    • Rarely correct
    • Best treated as “unmatched”
  • Not Matched
    • No viable RPID identified
    • Assign RPID88

Workflow Integration

Once AI results are generated:

  1. Export DIR Bulk Mapping file
  2. Use Excel formulas (e.g., VLOOKUP/XLOOKUP) to align AI suggestions with DIR records
  3. Apply mappings:
    • Accept High confidence matches w/ just a quick review
    • Review Medium matches
    • Assign RPID88 to remaining studies
  4. Upload via the DIR Bulk Mapping Tool

This approach allows facilities to move from hundreds of manual decisions to a manageable review process.

Why This Matters

  • Reduces time spent on cmbX mapping from days to hours
  • Improves mapping consistency
  • Minimizes unmapped studies
  • Supports cleaner, more reliable dose analytics

Availability of Instructions

I am willing to share the AI instruction set used to drive this process.

These instructions are:

  • Structured
  • Deterministic
  • Designed to work in a standalone environment (no prior context required)

If you are an imaging facility working through cmbX mapping and would like to try the AI instruction set I developed feel free to reach out.

Closing Thoughts

The cmbX enhancement is a net positive for the DIR and for dose transparency.
However, like many good ideas, it comes with implementation challenges.

With the right workflow—and the right tools—those challenges are very manageable.

Michael Bohl
Dose Registry Support Services, LLC
Questions? Contact us.

Leave a comment