DICOM Structured Reports (SR) explained

  • A DICOM Structured Report stores clinical measurements and findings as a tree of linked Content Items, not flat text or pixels.
  • Each Content Item carries a value type, such as TEXT, CODE, NUM, or SCOORD, plus a relationship type to its parent.
  • AI vendors prefer DICOM SR over PDF because viewers can parse numeric values and link them to image coordinates directly.
  • Template TID 1500 is the standard structure for reporting quantitative imaging findings and AI inferences.
  • Comprehensive SR and Comprehensive 3D SR allow the broadest value types and relationships for AI-generated findings.

A DICOM Structured Report (SR) is a standardized document object that stores clinical measurements, observations, and diagnostic codes in a machine-readable tree structure. Instead of flattening findings into text or pixels, an SR instance connects numeric data directly to referenced images and standard medical vocabularies.

Medical imaging software relies on DICOM SR to exchange clinical evidence. When an algorithm detects a lung nodule or measures an organ volume, it must output clinical evidence. Storing findings as a Structured Report allows picture archiving and communication systems (PACS) and electronic health records (EHR) to parse, index, and display them interactively.

How the DICOM SR document content tree structures medical findings

The DICOM SR Document Content Tree (defined in PS3.3 Section C.17.3) organizes all observations into a directed tree of Content Items. Every SR document begins with a single root Content Item of value type CONTAINER, and child nodes branch outward to represent sections, findings, measurements, and references.

┌────────────────────────────────────────────────────────┐
│ CONTAINER: Document Title (e.g. Imaging Measurement)   │
└──────────────────────────┬─────────────────────────────┘
                           │ CONTAINS
          ┌────────────────┴────────────────┐
          ▼                                 ▼
┌───────────────────────────┐   ┌───────────────────────────┐
│ CODE: Finding (Nodule)    │   │ NUM: Volume (42.5 mm³)    │
└─────────────┬─────────────┘   └─────────────┬─────────────┘
              │ INFERRED FROM                 │ INFERRED FROM
              ▼                               ▼
┌───────────────────────────┐   ┌───────────────────────────┐
│ IMAGE: Referenced Slice   │   │ SCOORD: Coordinates (2D)  │
└───────────────────────────┘   └───────────────────────────┘

The root container carries the Document Title inside the Concept Name Code Sequence (0040,A043). For example, it might specify code 126000 from the DICOM Content Mapping Resource (DCM) for an “Imaging Measurement Report”.

Under the root container, the Content Sequence (0040,A730) holds child items. Each child node contains 3 core attributes:

  1. Relationship Type (0040,A010): Defines the semantic connection to its parent node.
  2. Value Type (0040,A040): Defines what kind of data the node stores.
  3. Concept Name Code Sequence (0040,A043): Defines what the node means using a standard clinical coding scheme.

This hierarchy means that a measurement never floats in isolation. A numeric node for a 14.2 mm diameter links to its anatomical finding. That finding links directly to the image series, slice instance, and exact pixel coordinates.

Understanding Content Item value types and relationship types

Content Item value types determine how data is stored inside each node of the tree. The DICOM standard defines several distinct value types in PS3.3 Section A.35:

  • CONTAINER: A grouping node that holds other content items without storing a raw value.
  • TEXT: Free-form clinical text up to 1024 characters.
  • CODE: A coded term pointing to an external ontology like SNOMED-CT, RadLex, or LOINC.
  • NUM: A numeric measurement paired with a mandatory unit code from the Unified Code for Units of Measure (UCUM).
  • IMAGE and WAVEFORM: References pointing to other SOP instances via Referenced SOP Sequence (0008,1199).
  • SCOORD and SCOORD3D: 2D or 3D spatial coordinates defining points, polylines, circles, or ellipses on referenced images.
  • UIDREF and PNAME: Explicit unique identifier references and person names.

Relationship types establish the exact medical logic between parent and child nodes:

  • CONTAINS: The parent container includes this item as part of its section or list.
  • HAS PROPERTIES: The child provides descriptive details about the parent, such as laterality or measurement method.
  • INFERRED FROM: The parent conclusion is derived from the child evidence, such as deriving a tumor volume from a contour coordinate list.
  • HAS OBS CONTEXT: Defines who or what made the observation, such as the software algorithm version or the observing physician.
  • HAS CONCEPT MOD: Modifies the meaning of the concept name itself, such as qualifying a finding as “suspected” or “preliminary”.

Here is how you inspect an SR node using pydicom:

import pydicom

ds = pydicom.dcmread("report.dcm")

# Verify the root container title
root_title = ds.ConceptNameCodeSequence[0].CodeMeaning
print(f"Document Title: {root_title}")

# Inspect top-level child items
for item in ds.ContentSequence:
    rel = item.RelationshipType
    vtype = item.ValueType
    concept = item.ConceptNameCodeSequence[0].CodeMeaning
    
    if vtype == "NUM":
        value = item.NumericValue
        unit = item.MeasurementUnitsCodeSequence[0].CodeValue
        print(f"[{rel}] {concept}: {value} {unit}")
    elif vtype == "CODE":
        meaning = item.ConceptCodeSequence[0].CodeMeaning
        print(f"[{rel}] {concept}: {meaning}")

If you construct these items manually, you must maintain correct relationship rules. Setting an invalid relationship type, like placing an INFERRED FROM link under a node that does not accept inferences, causes strict PACS validators to reject the file.

Why do AI vendors send DICOM SR instead of PDF reports?

AI vendors deliver findings as DICOM SR because clinical imaging systems can parse numeric values, track lesions over time, and link measurements directly to viewer tools. While an encapsulated PDF is easy to generate, it traps data in visual text that medical systems cannot query.

There are 3 main reasons to pick DICOM SR over a static PDF:

First, structured reports provide interactive viewer integration. When a radiologist opens a study in a modern PACS or web viewer (such as OHIF), the viewer loads the SR alongside the CT or MRI series. Clicking a measurement in the viewer’s measurement table jumps directly to the referenced slice UID and highlights the spatial coordinates (SCOORD). A PDF cannot navigate a viewer viewport.

Second, structured reports enable automated indexing. PACS archives and clinical registries can extract tumor volumes, calcium scores, or ejection fractions straight into queryable data columns. Extracting those numbers from a PDF requires natural language processing or error-prone OCR.

Third, structured reports support longitudinal tracking. When an oncology patient returns 6 months later, tracking UIDs link the new scan. Systems match the new measurement to the baseline lesion automatically.

Generating an SR has one real limitation: implementation complexity. Building nested sequences of coded items in raw DICOM is far more tedious than writing an HTML page and printing to PDF. But the clinical utility in hospital environments makes SR the required standard for serious AI integrations.

Standardizing measurements with Template TID 1500

Template TID 1500 (Measurement Report) is the standard template defined in DICOM Part 16 for reporting quantitative imaging findings and AI inferences. Most clinical research platforms and commercial AI applications standardize on TID 1500.

TID 1500 defines a rigid, repeatable structure:

  1. Observation Context (TID 1001): Records whether an algorithm or a human generated the report, including the device manufacturer, model name, and software version.
  2. Procedure Reported (TID 1002): Specifies the imaging procedure that the report evaluates.
  3. Measurement Groups (TID 1501): Groups measurements by finding site or tracking identifier. Each group contains individual measurements (TID 300), qualitative evaluations, and references to source images or segmentation objects.

Inside each measurement group, every tracked lesion receives a persistent Tracking Identifier string and a globally unique Tracking Unique Identifier (UID). If your software monitors 3 lung nodules, each nodule has its own measurement group with its own tracking UID.

Standardizing on TID 1500 guarantees that external tools know how to parse your findings. Rather than writing custom parsing code for your company’s proprietary schema, any viewer or research registry that supports TID 1500 can render your measurements immediately.

Choosing an SR Storage SOP Class: Basic, Enhanced, or Comprehensive

DICOM Part 4 Annex O specifies several Structured Reporting Storage SOP Classes that limit which value types and relationships you can use. Choosing the right SOP class depends on the complexity of your findings.

  • Basic Text SR (1.2.840.10008.5.1.4.1.1.88.11): The simplest class. It allows only TEXT, CODE, and reference items. It does not allow numeric measurements (NUM) or spatial coordinates (SCOORD). Use this only for simple narrative notes.
  • Enhanced SR (1.2.840.10008.5.1.4.1.1.88.22): Adds numeric measurements (NUM) and 2D spatial coordinates (SCOORD), but restricts complex relationship nesting.
  • Comprehensive SR (1.2.840.10008.5.1.4.1.1.88.33): The standard workhorse for AI findings. It allows all value types, deep nesting, and broad relationship types.
  • Comprehensive 3D SR (1.2.840.10008.5.1.4.1.1.88.34): Extends Comprehensive SR with 3D spatial coordinates (SCOORD3D). Coordinates use patient anatomical space, such as millimeter offsets from the patient origin.

If you generate imaging AI findings, choose Comprehensive SR or Comprehensive 3D SR. They allow your product to link measurements to segmented regions and image slices without hitting standard constraints.

Ingesting hospital studies before generating structured findings

Before your pipeline can generate a Structured Report, it must reliably ingest the original imaging study from the sending hospital. Hospital scanners and local PACS push raw CT and MRI studies over local networks using standard DICOM C-STORE associations.

Connecting a cloud analysis pipeline directly to hospital modalities introduces heavy networking overhead. You have to configure mutual TLS, manage client certificates for each sending facility, navigate hospital firewall rules, and maintain an internet-facing DICOM listener that runs 24/7.

A managed gateway such as Dicomly handles this inbound connectivity layer. It receives C-STORE transmissions from hospital senders over mutual TLS and forwards each instance to your HTTPS application as a standard DICOMweb STOW-RS request. You trade running low-level network listeners for a managed webhook endpoint. This leaves your team free to train models, run inference, and generate compliant DICOM SR instances. You can learn how to handle incoming data in our guide on receiving deliveries.

Next steps for creating your first Structured Report

Start by inspecting an existing Structured Report to see how child nodes nest in practice.

Download an open dataset with SR instances from The Cancer Imaging Archive. Then, write a small pydicom script to walk the ContentSequence. Print the value types, relationship types, and concept meanings to visualize the tree structure.

Once you understand the tree, read our guide on DICOM pixel data for AI and ML engineers to calibrate pixel matrices for your models. You can also review DICOM SOP classes and transfer syntaxes to configure association negotiation.

Dicomly builds infrastructure for healthcare software teams that receive DICOM studies from hospitals. Why we built Dicomly →

Sources

  1. DICOM Standard Part 3 (PS3.3, Section A.35 — Structured Report Document Information Object Definitions) (dicom.nema.org)
  2. DICOM Standard Part 3 (PS3.3, Section C.17.3 — SR Document Content Module) (dicom.nema.org)
  3. DICOM Standard Part 16 (PS3.16, Annex A — Template TID 1500 Measurement Report) (dicom.nema.org)
  4. DICOM Standard Part 4 (PS3.4, Annex O — Structured Reporting Storage SOP Classes) (dicom.nema.org)

Start receiving DICOM today.

Early access. No credit card. First endpoint free.