Inferensys

Glossary

DICOM

DICOM (Digital Imaging and Communications in Medicine) is the international standard for transmitting, storing, and sharing medical images and related data, ensuring interoperability between radiology devices, picture archiving systems, and workstations.
Data scientist building training data pipeline on laptop, data preprocessing visible, technical workspace.
MEDICAL IMAGING STANDARD

What is DICOM?

DICOM is the international standard for transmitting, storing, and sharing medical imaging and related data, ensuring interoperability between radiology devices, picture archiving systems, and workstations.

Digital Imaging and Communications in Medicine (DICOM) is the international standard for medical imaging informatics, defining both a file format and a network communication protocol. It ensures that imaging modalities like CT scanners, MRI machines, and ultrasound devices from different vendors can produce, store, and transmit images that any compliant system can interpret.

A DICOM file encapsulates pixel data alongside a rich header of metadata, including patient demographics, acquisition parameters, and spatial coordinates. The protocol operates over TCP/IP using a service-based model where devices assume roles as Service Class Users (SCUs) or Service Class Providers (SCPs) to perform operations like store, query, and retrieve across a hospital network.

MEDICAL IMAGING STANDARD

Core Characteristics of DICOM

DICOM is far more than a file format; it is a comprehensive network protocol and semantic framework that governs the entire medical imaging workflow. These core characteristics define its role as the universal language of radiology, cardiology, and beyond.

01

Object-Oriented Information Model

DICOM structures data using Information Object Definitions (IODs) , which are abstract data models for real-world medical entities like patients, studies, and images. An IOD specifies the mandatory and optional attributes (data elements) that describe an object. For example, a CT Image IOD includes attributes for pixel data, slice thickness, and KVP. This object-oriented design ensures that data is self-describing and contextually complete, allowing any compliant system to interpret the content without external metadata.

02

Service-Oriented Operations (DIMSE)

Interoperability is achieved through DICOM Message Service Elements (DIMSE) , which define the commands and notifications that applications use to interact. These services are paired with IODs to form Service-Object Pair (SOP) Classes, the fundamental unit of DICOM conformance. Key operations include:

  • C-STORE: Transfers images to a PACS archive.
  • C-FIND: Queries a database for studies by patient name or ID.
  • C-MOVE: Instructs an archive to send images to a third-party workstation.
  • C-ECHO: A simple "ping" to verify network connectivity.
03

Unique Identifier (UID) System

Every entity in the DICOM universe is assigned a globally unique identifier. UIDs are registered via the ISO 8824 standard to prevent conflicts between vendors and institutions. Critical UID types include:

  • Study Instance UID: Uniquely identifies a single imaging exam.
  • Series Instance UID: Identifies a specific acquisition sequence within a study.
  • SOP Instance UID: Uniquely tags every single image or structured report object. This hierarchical identification system is the backbone of reliable data linking and retrieval across an enterprise.
04

Multi-Part Media Storage Model

DICOM defines a physical interchange model for offline media like CDs, DVDs, and USB drives. A DICOMDIR file acts as a table of contents, indexing all images on the media without requiring a network query. The media structure mandates a specific directory hierarchy and a DICOM Media Storage SOP Class to ensure that a disc created by one vendor's system can be read by any other. This standard was critical for patient image portability before ubiquitous network access.

05

Comprehensive Conformance Statement

Unlike a simple file specification, DICOM requires every compliant device to publish a DICOM Conformance Statement. This document is a detailed technical contract that explicitly lists which SOP Classes the device supports as a Service Class User (SCU) or Service Class Provider (SCP). An integration engineer uses these statements to determine if two systems can interoperate, checking for matching SOP Class UIDs and supported transfer syntaxes before any connection is attempted.

06

Pixel Data Encoding and Transfer Syntax

DICOM separates the semantic meaning of an image from its binary encoding. The Transfer Syntax defines how pixel data is compressed and byte-ordered for transmission. A single SOP Class can support multiple transfer syntaxes, negotiated during association establishment. Common examples include:

  • Implicit VR Little Endian: The default, uncompressed format.
  • JPEG Lossless: For mathematically reversible compression.
  • JPEG 2000: For high-efficiency, scalable compression. This negotiation allows a PACS to store images losslessly while sending a compressed preview to a workstation.
DICOM ESSENTIALS

Frequently Asked Questions

Clear, technical answers to the most common questions about the DICOM standard, its role in medical imaging interoperability, and its relationship to other healthcare data formats.

DICOM (Digital Imaging and Communications in Medicine) is the international standard for transmitting, storing, retrieving, and sharing medical images and associated data. It works by encapsulating both pixel data and a rich set of metadata—such as patient demographics, study identifiers, and acquisition parameters—into a single structured file. A DICOM file consists of a header containing Data Elements organized by numerical tags (e.g., (0010,0010) for Patient Name) and a body holding the image pixel data. The standard also defines network protocols built on TCP/IP, including DICOM Message Service Elements (DIMSE) for operations like C-STORE (store images), C-FIND (query), and C-MOVE (retrieve), enabling seamless communication between modalities (CT, MRI, ultrasound), Picture Archiving and Communication Systems (PACS), and workstations from different vendors.

Prasad Kumkar

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.