DICOM SR and the Structured Report Modern Radiology Needs

The Problem with Free-Text Reports

Most radiology reporting systems still produce reports as plain text: a document containing findings, an impression, and recommendations. While this format works well for human readers, it breaks down when integrated with other healthcare systems.

• No auditability: A finding such as “8 mm pulmonary nodule” exists only as a text string. It cannot be reliably indexed, prioritized, or used to trigger automated clinical alerts.
• Limited interoperability: HL7 OBX segments can transport the report text, but downstream systems—including Electronic Health Records (EHRs), follow-up workflows, and analytics platforms—require structured information rather than unstructured narrative.
• No automated triage: In radiology departments processing thousands of studies every day, critical findings cannot be prioritized efficiently without attempting to parse free text.

What DICOM SR Brings

DICOM Structured Report (SR) fundamentally changes the reporting model by representing the radiology report as a structured document composed of standardized data elements.

It includes:

• Code Sequences that reference standardized clinical terminologies such as SNOMED CT, RadLex, and LOINC, allowing every finding to have a precise semantic meaning.
• Measurement Groups that store quantitative information—including diameter, volume, density, and other measurements—along with units, acquisition methods, and contextual metadata.
• Explicit Relationships that define how observations relate to one another, such as associating a measurement with a specific lesion or linking multiple findings within the same examination.

The result is a report that remains fully readable by clinicians while simultaneously becoming machine-readable for any system capable of interpreting DICOM SR.

From Dictation to Structured Reports with AI

Historically, structured reporting adoption has been limited because asking radiologists to manually author structured reports significantly increases reporting time and disrupts established workflows.

The AI-native approach removes this barrier.

Radiologists continue dictating naturally, while an AI pipeline—combining Whisper speech recognition with medical NLP and terminology models—automatically generates the structured report.

Radiologist Dictation
        ↓
Whisper Speech Transcription
        ↓
Medical NLP Extracts Findings
        ↓
Mapping to SNOMED CT / RadLex
        ↓
DICOM SR Generation
(Measurements + Relationships)
        ↓
Editable Preview
        ↓
Versioned, Traceable Final Report

The radiologist reviews, edits if necessary, and approves the report instead of creating it from scratch. The final output is a fully structured, auditable DICOM SR document that integrates seamlessly with downstream clinical systems.

At Soft[in]Health, every generated SR is validated against the Basic Text SR DICOM SOP Class and, whenever imaging data supports richer semantic relationships, against Enhanced SR. The SR parser is implemented in the dicom_sr_processor module, while report generation is fully integrated into the NextRIS reporting workflow.

The true value of DICOM SR is not the standard itself—it’s what the standard enables: automated triage, structured follow-up of imaging findings, advanced analytics, and complete clinical auditability without relying on free-text parsing. ::