DICOM SR and the structured report radiology needs

The free-text report problem

Most radiology reports are plain text: a page with findings, impression and recommendation. It works for a human eye but breaks at every integration point.

• No audit: a finding “lung nodule 8 mm” lives only as a string. No system can index, prioritize or alert on it.
• No integration: HL7 OBX can carry the text, but any downstream consumer (EHR, automated follow-up) needs structure.
• No triage: with thousands of studies a day, there is no way to prioritize critical ones without parsing text.

What DICOM SR brings

DICOM Structured Report (SR) changes the model: the report is a structured document with:

• code sequences that reference terminologies (SNOMED CT, RadLex, LOINC) for each finding and its clinical meaning.
• measurement groups that carry numeric measurements (diameter, volume, density) with units and method.
• relationships explicit: “this finding belongs to that observation”, “this measurement refines that finding”.

The result: a report both human-readable and machine-processable by any DICOM SR reader.

From dictation to SR with AI

The historical challenge was that asking radiologists to write SR manually was unworkable: it added friction to an already overloaded task.

The IA-native way out: the radiologist dictates normally, and the AI layer (Whisper + terminology models) produces SR automatically:

 dictation → Whisper transcription
         → finding extraction via medical NLP
         → SNOMED CT / RadLex mapping
         → DICOM SR with measurements and relationships
         → editable preview for the radiologist
         → final versioned, traceable commit

The radiologist approves and edits; they don’t write from scratch. The final report is structured SR, auditable and ready to integrate with any downstream system.

At Soft[in]Health we validate every SR against the DICOM SOP Class Basic Text SR and, where imaging allows, Enhanced SR for richer relationships. The parser lives in dicom_sr_processor, and generation in the NextRIS report workflow.

The value of DICOM SR isn’t the standard: it’s what it enables: automated prioritization, finding follow-up and audit without parsing text. ::