Repository navigation
DICOM Background
Digital Imaging and Communications in Medicine (DICOM) is the ubiquitously and globally adopted international standard for storing, transmitting, and handling medical imaging information and related data. It enables interoperability across devices and systems used in clinical imaging workflows.
In contrast, other image formats, such as jpeg, png, and NIfTI are not typically used in clinical practice, with the exception of visible light images, such as dermatology or endoscopy, where JPEG formats may be encapsulated within a DICOM wrapper. Modalities like radiology, cardiology, and pathology require higher bit-depth (e.g., 12-bit grayscale) and rich metadata, which these formats do not support. DICOM, on the other hand, embeds image acquisition parameters (e.g., slice thickness) and patient information directly within the file.
Have a question? Post questions or issues to forums.ohdsi.org with the tag #imaging and we will get right back to you!
DICOM has three primary components: 1.) Real-World Information Model 2.) Instance Object Definitions (IODs) 3.) Transport and Communication Services
Additionally, DICOM specifies numerous enumerated value sets (vocabularies) that support interoperability across systems and vendors.
The DICOM standard is comprised of "Parts" and is freely available at https://www.dicomstandard.org/current.
The real-world data model and the objects are defined in DICOM Part 3. Objects are defined for different modalities, e.g., "CT Image Object Definition (IOD)" in radiology and "VL Whole Slide Microscopy Image IOD" for pathology. The figure below is the basic DICOM Real-World Model. The real-world model describes how imaging data are structured conceptually and stored operationally. It introduces a hierarchy: patient-study-series-instance. Let's examine an example.
<INSERT Part 3: Figure 7-1a. DICOM Model of the Real World>
In its most basic form, this data model defines that a patient may have one to many imaging studies, a study is comprised of one to many series, and a series may have one to many instances (often images). Note that the Equipment module, which determines the modality, is defined at the series level, allowing a single study to include multiple modalities, such as MR and PET. Also note that the Frame of Reference, which defines spatial consistency by using a common coordinate system, is defined at the series level.
DICOM image objects are "composite" objects, meaning that for every image, both the "header" (attributes) and "body" (pixel data) are contained within each "instance". These IODs are defined in Part 3. A high-level example is given below:
The header of an image consists of Modules and Attributes. A Module is a grouped set of information, e.g., the Patient Identification Module. Attributes are individual pieces of information, e.g., "Patient Name". In turn, Attributes are identified by attribute numbers, also commonly referred to as "tags". Tags are represented as (GGGG,EEEE), where G is the Group Number and E is the element number. As an example, the tag for "Patient Name" is (0010,0010). Even numbers are used for public groups; odd numbers indicate privately defined groups, which may be unpublished and undefined.
The enumerated values for attribute contents are defined in both Part 3 and Part 16. While DICOM does define its own code system, enumerated values often consist of SNOMED CT and LOINC codes.
For clinical practice, the two primary forms of transport are the original "DIMSE Services," a unique TCP/IP protocol, and "DICOMweb," which is based on RESTful (HTTP) services, as defined in DICOM Part 4. DICOMweb does have a /metadata retrieve parameter to retrieve only the DICOM header information.
# Retrieve all instances within a study through DICOMweb
url = f'{base_url}/studies/{study_uid}'
headers = {'Accept':'multipart/related; type="application/dicom"; transfer-syntax=*', "Authorization":bearer_token}
response = client.get(url, headers=headers) #, verify=False)The DICOMweb cheat sheet is also useful.
The concept of a "unique identifier" (UID) is critical in the DICOM standard.
A UID is a globally unique, long numeric string that follows the ISO OID (Object Identifier) structure, e.g., "1.2.840.113754.5.23454656". In this example, "1.2.840.113754" is the UID root registered through ANSI for the Veterans Affairs (VA) Vista system. The DICOM UID root is "1.2.840.10008".
The institution that owns this root is responsible for maintaining the unique values associated with every UID generated. There are many algorithms that guarantee UID uniqueness by adding components, including equipment instance, date, and time (to ms) stamp, and other variables.
Every DICOM study created by any manufacturer has a unique Study Instance UID (0020,000D). Similarly, every Series Instance (0020,000E) and SOP Class Instance (0008,0018) has a unique identifier (UID). DICOM Part 6 contains a list of DICOM-registered UIDs.. Every DICOM object definition has an associated UID.
Departments that routinely generate DICOM-formatted images include Radiology, Cardiology, and increasingly Pathology. More recently, other specialties such as Dermatology and Gastroenterology have also begun to adopt DICOM by encapsulating visible light images (e.g., photographs or videos) within DICOM headers. These headers play a critical role in storing, indexing, and managing images by including key metadata such as patient identifiers, acquisition context, and modality type.
A modality refers to the type of imaging equipment or technique used to generate a medical image. Each modality is defined by the physical principles it uses (e.g., X-ray, sound waves, magnetic fields) and the clinical application it serves. In radiology, these include X-ray (XR), computed tomography (CT), magnetic resonance (MR), ultrasound (US), and more. In cardiology, common modalities include echocardiography (US), angiography (XA), and electrocardiography (ECG). In pathology, general microscopy (GM) or whole-slide microscopy (SM) is used. In other clinical areas, such as colonoscopy, video imaging (XC) may be used. The two-letter modality specified value is stored in the DICOM attribute Modality (0008,0060).
Radiologists and Cardiologists use a system called a Picture Archiving and Communication System (PACS) to retrieve a worklist (list of studies to be read), view images, create measurements and other analysis, interpret, and store medical images. These systems typically consist of multiple very high-resolution monitors, an EHR monitor, and a report generation monitor, as shown in the figure below. Typically, a PACS system is specialized to a particular department, such as Radiology or Pathology. PACS storage can be on-prem or cloud-based.
Radiology reports, which are typically narrative text with five predefined sections , are created using voice-to-text dictation in a separate, but integrated, application. As seen in the figure above, the radiologist is reviewing the images while dictating the interpretation using the reporting application on the RIGHT/LEFT monitor. After the radiologist has completed the interpretation, the report is reviewed, signed, and sent to the EHR where it is permanently stored.
A Vendor Neutral Archive (VNA) is a very large storage system for medical images from clinical departments across the entire healthcare enterprise. The intent of a VNA is to reduce costs and increase efficiency by centrally managing millions of images. VNAs can store petabytes, or even exabytes, of data. VNAs can be on-prem or cloud-based.
The metadata—or “header”—is an integral part of every DICOM file. It includes detailed information about the imaging acquisition process as well as patient-related data. This metadata is distinct from image-derived features, which refer to findings extracted from the image content itself, such as the presence of air, fluid, or solid masses. Image-derived features are often described in radiology reports, but in this context, our focus is on features extracted directly from the images through computational algorithms or manual annotations.
DICOM metadata captures essential acquisition parameters that provide context for how an image was generated. These include equipment settings, contrast administration, manufacturer and model information, and the names of the performing personnel when available. In addition to structured fields, DICOM metadata also contains free-text fields such as Study Description and Series Description, which can be helpful for identifying different sequences within the same imaging modality. For instance, a Magnetic Resonance Imaging (MRI) study may include several sequences like T1-weighted, T2-weighted, and FLAIR—each produced using distinct imaging techniques and generating different visual representations of anatomy or pathology. Recognizing these distinctions is not only important for clinical interpretation but also critical for ensuring that image analysis algorithms operate on the appropriate input data.
Image-derived features refer to quantifiable findings extracted directly from medical images, either through manual annotation or computational algorithms such as segmentation, classification, or radiomics. These features can describe anatomical structures, lesions, textures, shapes, or changes in signal intensity. For example, a segmentation algorithm may delineate the boundaries of a tumor, while a classification model may label it as benign or malignant based on image characteristics. These outputs can be encoded using standardized clinical terminologies such as the Radiology Lexicon (RadLex) or SNOMED CT.
There are several critical pieces of information that must be accurately mapped from the EHR to DICOM images and reports to ensure proper linkage of clinical information.
In radiology, as well as many other departments, the ordering physician from primary care or the emergency department creates an order in the EHR, which is a request for an imaging study to be performed. At this time, an order number is assigned automatically by the EHR. The type of procedure is specified at a high level. An example is "CT Head". The procedure is scheduled at a specific date, time, and location. For advanced imaging (CT, MR, PET), a radiologist will often "protocol" the exam by specifying details such as kVP, slice thickness, pulse sequence, etc., based on the person and the anatomy being imaged. At the time of the exam, the technologist will query the Radiology Information System (RIS) using a DICOM Modality Worklist Query (MWL). The RIS can be an application within the EHR or a separate system. The modality will often query for "the next patient," usually based on modality, date, and time, to accurately and electronically obtain the patient and procedure information. It was not that long ago that all the information was manually typed into the modality from a paper form by the technician.
The order number is at the level of an EHR "orderable unit of work". However, for some types of complex procedures, a single EHR order can map into multiple pieces of imaging work, based on either separation of time (e.g., two nuclear medicine (NM) scans required 8 hours apart, but part of the same "study") or by equipment (e.g., a PET and CT scans completed on two different pieces of equipment). An "Accession Number" (0008,0050) is a unit of work, which can be performed at a modality. Typically, the order number directly matches the Accession Number in a one-to-one mapping. For example, a "CT Head" is a single unit of work at a specific modality. However, a single EHR order number can map to multiple Accession numbers.
The Medical Record Number (MRN), also known as the Patient ID (PID), is assigned by the EHR and links all of a patient's information. In HL7 v2, the PID element can be found in the PID segment of most messages. In HL7 FHIR, the MRN is in the "identifier" element of the Patient Resource. The DICOM attribute is called Patient ID (0010,0020). Using the contents of (0010,0020), all of a patient's studies can be located in a PACS or VNA. The procedure names can be coded in LOINC, but often are site-specific orderables as text fields.
When a radiology report is created, there are several pieces of information that are critical to link the report back to the patient record in the EHR. These include the MRN, order number, and accession number. There can be a single report per order number, or, for complex imaging procedures, a single report can be generated per accession number, or a merged report can be created.
For a more detailed mapping of all required attributes, refer to the Integrating the Healthcare Enterprise (IHE) Radiology Technical Framework Scheduled Workflow (SWF.b) profile. IHE is a standards organization that maps multiple standards together to solve the complete workflow for specific clinical use cases.
What are the challenges of working with imaging data for research -- This section is still under construction!
Body part imaged is part of the image acquisition attributes and is captured in the image_occurrence table anatomic_site_concept_id.
Body part imaged can either be "coarse" (e.g., abdomen) or "fine grained" (e.g., liver). In DICOM Part 16, Table CID 4 contains a shorter "coarse" list of anatomical locations coded in SNOMED CT. Institutions which implement the MI-CDM extensions must use the OHDSI conventions for the body part imaged field to be useful.
It should be noted that there are some hierarchical relationships built into the SNOMED CT anatomical reference model, but this is not extensive.
There are several attributes to identify the body part imaged in the DICOM standard. Unfortunately, this is not a uniformly coded attribute across modality vendors and often requires mapping for consistency and to be able to query for known values.
The most commonly used attribute is "Body Part Examined" (0018,0015) which is defined at the Series level. The value representation is type "code string" or "CS". The defined terms for "Body Part Examined" are found in DICOM Part 16 Annex L "Correspondence of Anatomic Region Codes and Body Part Examined Defined Terms". These codes are primarily SNOMED CT codes. Sometimes, just the "Code Meaning" (text string) is used as the value for (0018,0015). However, the DICOM standard also allows for the technologist to enter "site specific text" in this field and, unfortunately, this is often the case in the real-world. As part of DICOM image header ETL, there typically needs to be logic to map to a standard concept.
There is also an "Anatomic Region Sequence" (0008,2218) attribute at the Series level intended to supercede "Body Part Examined". "Anatomic Region Sequence" has a value representation of "sequence" or "SQ". As a code sequence, this attributed is expected to use the coded value from Annex L above, not just the "Code Meaning" or text string described above.
Finally, for some body parts, laterality must be considered. DICOM contains "Laterality" (0020,0060) at the series level with the enumerated value options of "L" and "R". There is a coded sequence attribute "Anatomic Region Modifier Sequence" (0008, 2220) defined at the Series level which can be used for laterality using SNOMED CT codes for "Left", "Right", "Unilateral", and "Bilateral".
For additional information, consider the following resources:
Have a question? Post questions or issues to forums.ohdsi.org with the tag #imaging and we will get right back to you!