Glossary

DICOM Segmentation Object

A specialized DICOM Information Object Definition (IOD) that stores pixel-level segmentation results as binary or fractional label maps alongside spatial registration metadata for interoperability across clinical systems.

Get in touch Learn more

Data scientist building training data pipeline on laptop, data preprocessing visible, technical workspace.

PIXEL-LEVEL ENCODING STANDARD

What is a DICOM Segmentation Object?

A specialized DICOM Information Object Definition (IOD) that encodes pixel-level classification results as binary or probabilistic label maps, complete with spatial registration metadata linking each segmented pixel to its source imaging study.

A DICOM Segmentation Object is a specialized Information Object Definition that stores semantic segmentation results—pixel-level classifications of anatomical structures or pathologies—as raster label maps within the DICOM standard. Unlike DICOM RT Structure Sets, which represent contours as geometric coordinates, segmentation objects encode classifications directly in the image domain, preserving the spatial resolution of the source modality and enabling voxel-level analysis.

Each segmentation object contains one or more segments identified by numeric labels, with each segment referencing a coded terminology entry (e.g., SNOMED or RadLex) to unambiguously define the anatomical or pathological entity. The object stores the segmentation as either binary or fractional frames, where fractional values represent probability maps or partial volume estimates. Critically, the object includes spatial registration metadata—referencing the source image's Frame of Reference UID—ensuring precise alignment between the segmentation and the original acquisition.

STRUCTURE & INTEROPERABILITY

Key Features of DICOM Segmentation Objects

The DICOM Segmentation Object (SEG) is a specialized Information Object Definition (IOD) that stores pixel-level classification results as binary or fractional label maps, ensuring spatial registration with the source imaging study.

Binary vs. Fractional Segmentation Types

DICOM SEG supports two distinct segmentation types to accommodate different clinical and research needs:

BINARY: Each segment is a separate bit-plane in a single image, where a pixel value of 1 indicates membership. This is efficient for storing multiple non-overlapping structures like organs-at-risk.
FRACTIONAL: Each segment occupies an entire image plane with continuous values between 0 and 1, representing the probability or partial volume of tissue membership. This is critical for storing model outputs like softmax probabilities before thresholding.

Spatial Registration Fidelity

The SEG object maintains pixel-perfect spatial alignment with the referenced source images through mandatory registration metadata:

Frame of Reference UID: Uniquely identifies the coordinate system shared between the segmentation and the source DICOM series.
Image Orientation (Patient) and Image Position (Patient): Define the exact 3D location and orientation of every segmentation frame.
Pixel Spacing and Slice Thickness: Ensure voxel dimensions match the source, preventing geometric distortion when overlaying masks on original scans.

Multi-Frame Encoding Efficiency

DICOM SEG leverages the Enhanced Multi-frame module to store an entire 3D volume or multiple segments in a single SOP instance, avoiding the fragmentation of legacy single-slice storage:

Shared Functional Groups: Common attributes like plane orientation and pixel measures are stored once and referenced by each frame, dramatically reducing storage overhead.
Per-Frame Functional Groups: Allow each frame to override specific attributes, such as segment identification, enabling a single object to encode dozens of anatomical structures simultaneously.

Segment Attribute Macros

Each segmented structure is described using a rich set of coded attributes stored in the Segment Sequence, enabling semantic interoperability with treatment planning and analysis systems:

Segmented Property Category: A coded term (e.g., 'Anatomical Structure' or 'Morphologically Abnormal Structure') classifying the type of entity.
Segmented Property Type: A specific code from SNOMED, FMA, or other ontologies (e.g., 'Liver' from SNOMED) identifying the exact anatomy.
Tracking ID and Tracking UID: Allow longitudinal tracking of the same anatomical structure across multiple segmentation instances over time.

Nested Content Identification

DICOM SEG supports hierarchical segmentation through the Segment Algorithm Name and Segment Algorithm Type attributes, which document the exact method used to generate each mask:

Algorithm Name: Records the specific software or model (e.g., 'TotalSegmentator v2', 'nnU-Net').
Algorithm Type: Categorizes the approach as AUTOMATIC, SEMIAUTOMATIC, or MANUAL, providing crucial provenance for regulatory audits and clinical review.
This metadata ensures that downstream users can assess the reliability of a segmentation based on its generation method.

Surface Mesh Extraction Support

While the SEG object stores voxel-based label maps, its design facilitates conversion to surface-based representations for visualization and 3D printing:

The Marching Cubes algorithm can be applied directly to the binary or thresholded fractional data to generate polygonal meshes.
The resulting meshes are often stored in the companion DICOM RT Structure Set for radiotherapy or exported as STL files for surgical planning.
The precise spatial registration metadata in the SEG ensures that extracted surfaces maintain geometric fidelity to the source anatomy.

DICOM SEG OBJECT FAQ

Frequently Asked Questions

Clear, technically precise answers to the most common questions about the DICOM Segmentation Object, its structure, and its role in medical imaging AI pipelines.

A DICOM Segmentation Object (SEG) is a specialized Information Object Definition (IOD) that stores pixel-level classification results as binary or fractional label maps, rather than raw pixel intensity data. Unlike a standard DICOM image, which contains the original acquired signal (e.g., Hounsfield Units for CT), a SEG object encodes the output of a segmentation algorithm—each voxel is assigned a label indicating tissue type, organ, or pathology. Critically, it stores this data alongside spatial registration metadata, ensuring the segmentation map is perfectly aligned with the source imaging study. This makes it the interoperable standard for exchanging AI-generated contours between systems, such as from a segmentation workstation to a treatment planning system or a clinical trial imaging core lab.

Enabling Efficiency, Speed & Accuracy

Intelligent Analysis, Decision & Execution

We build AI systems for teams that need search across company data, workflow automation across tools, or AI features inside products and internal software.

Talk to Us

Search across company data

Give teams answers from docs, tickets, runbooks, and product data with sources and permissions.

Useful when people spend too long searching or get different answers from different systems.

Enterprise searchRAGPermissions

Automate internal workflows

Use AI to route work, draft outputs, trigger actions, and keep approvals and logs in place.

Useful when repetitive work moves across multiple tools and teams.

AI agentsWorkflow automationGovernance

Add AI to products and internal tools

Build assistants, guided actions, or decision support into the software your team or customers already use.

Useful when AI needs to be part of the product, not a separate tool.

AI integrationDecision supportModel routing

SEGMENTATION ECOSYSTEM

Related Terms

Understanding the DICOM Segmentation Object requires familiarity with the evaluation metrics, architectures, and post-processing techniques that generate and validate its binary label maps.

Dice Score (F1 Score)

The primary metric for validating the spatial overlap stored in a DICOM Segmentation Object. It calculates twice the intersection of the predicted binary mask and the ground truth, divided by the total sum of pixels in both masks.

Formula: 2 * |A ∩ B| / (|A| + |B|)
Range: 0 (no overlap) to 1 (perfect agreement)
Clinical Relevance: A Dice score > 0.7 is generally considered excellent agreement for volumetric analysis.

0.0 - 1.0

Value Range

Hausdorff Distance

A boundary-based metric that quantifies the maximum surface-to-surface distance between the predicted segmentation and the ground truth. It identifies the worst-case segmentation error.

95th Percentile HD: Often used instead of maximum to exclude outliers.
Unit: Measured in millimeters (mm).
Importance: Critical for radiotherapy, where a single large boundary deviation can cause catastrophic radiation overdose to healthy tissue.

Unit of Measurement

U-Net Architecture

The foundational convolutional neural network that generates the pixel-level probability maps later encoded into a DICOM Segmentation Object. Its symmetric encoder-decoder structure with skip connections preserves fine spatial detail.

Encoder: Captures contextual information via downsampling.
Decoder: Enables precise localization via upsampling.
Skip Connections: Concatenate encoder feature maps to decoder layers to recover spatial resolution lost during pooling.

EXPLORE

MONAI Framework

An open-source, PyTorch-based ecosystem purpose-built for deep learning in medical imaging. It provides domain-optimized data loaders, transforms, and network architectures that output the data stored in DICOM Segmentation Objects.

MONAI Label: An intelligent labeling tool for rapid annotation.
MONAI Deploy: Bridges the gap from research models to clinical production.
DICOM Compliance: Native support for parsing and writing DICOM SEG objects.

EXPLORE

Connected Component Analysis

A post-processing algorithm applied to the binary mask before it is serialized into a DICOM Segmentation Object. It identifies and labels isolated contiguous regions of the same class.

Noise Removal: Eliminates small, spurious activations (e.g., removing a 3-pixel island predicted as a tumor).
Instance Separation: Distinguishes between two touching objects of the same class.
3D Extension: Operates on volumetric voxel data, not just 2D slices.

Marching Cubes

A computer graphics algorithm that extracts a polygonal mesh of an isosurface from the 3D scalar field stored in a DICOM Segmentation Object. It converts the binary voxel grid into a surface rendering.

Input: 3D binary segmentation mask.
Output: Triangle mesh (STL/OBJ) for 3D visualization or 3D printing.
Surgical Planning: Enables surgeons to visualize the spatial relationship between a tumor and surrounding vasculature.

About the author

Prasad Kumkar

CEO & MD, Inference Systems

Prasad Kumkar is the CEO & MD of Inference Systems and writes about AI systems architecture, LLM infrastructure, model serving, evaluation, and production deployment. Over 5+ years, he has worked across computer vision models, L5 autonomous vehicle systems, and LLM research, with a focus on taking complex AI ideas into real-world engineering systems.

His work and writing cover AI systems, large language models, AI agents, multimodal systems, autonomous systems, inference optimization, RAG, evaluation, and production AI engineering.

Limited slotsGet a Free AI Consultation

How We Work

Custom AI workflows for your Business

One-fit-all AI don't work for modern businesses. At Inferensys, we aim to understand your business & custom requirements; which we use to define most efficient agentic workflows, the data, and the tools for your business.