DICOM Format Complete Guide: Medical Imaging, File Structure & Converting DICOM

What Is DICOM?

DICOM (Digital Imaging and Communications in Medicine) is the international standard for medical images and related information. Developed by the American College of Radiology (ACR) and the National Electrical Manufacturers Association (NEMA) since 1985 (initially as ACR-NEMA), DICOM PS3.x defines everything about how medical images are stored, transmitted, and displayed — from CT scans and MRIs to ultrasounds, X-rays, mammograms, and endoscopy images.

A DICOM file is not just an image — it is an information object combining pixel data with rich metadata: patient identity, acquisition parameters, institution information, equipment calibration data, and clinical context. This comprehensive metadata is what distinguishes DICOM from consumer image formats like JPEG or PNG, and what makes it medically and legally essential.

DICOM is ubiquitous in healthcare: every hospital PACS (Picture Archiving and Communication System), every radiology workstation, every radiotherapy planning system, and every modality (CT, MRI, PET, ultrasound) produces and consumes DICOM. The standard is maintained by the DICOM Standards Committee and updated continuously.

DICOM File Structure

File Meta Information (Preamble + DICOM Header)

Every DICOM file starts with:

Bytes 0–127:  128-byte preamble (zeros by convention, sometimes ASCII text)
Bytes 128–131: "DICM" magic (0x44 0x49 0x43 0x4D)
Bytes 132+:   File Meta Information dataset (Group 0002 attributes)

The File Meta Information contains:

(0002,0001) File Meta Information Version: 2 bytes, always 0x00 0x01
(0002,0002) Media Storage SOP Class UID: What type of image (CT Image, MR Image, etc.)
(0002,0003) Media Storage SOP Instance UID: Globally unique ID for this specific image
(0002,0010) Transfer Syntax UID: Encoding rules (byte order, compression)

Data Elements

After the File Meta Information, the DICOM dataset consists of data elements in Tag-Length-Value (TLV) format:

|-- 4 bytes --|-- 2 bytes --|-- 2 or 4 bytes --|-- N bytes --|
|  Tag (GG,EE)|  VR        |  Length          |  Value      |

Tag: Two 16-bit unsigned integers: Group (GG) and Element (EE). Tags are written as (GGGG,EEEE).
VR (Value Representation): 2-byte ASCII code describing data type
Length: Size of value in bytes
Value: The actual data

Example tags:

(0008,0060) CS  Modality                     = "CT"
(0008,0020) DA  Study Date                   = "20240315"
(0010,0010) PN  Patient Name                 = "Doe^John^A"
(0010,0020) LO  Patient ID                   = "PAT-123456"
(0010,0030) DA  Patient Birth Date           = "19750822"
(0010,0040) CS  Patient Sex                  = "M"
(0020,000D) UI  Study Instance UID           = "1.2.840.10008.5.1.4.1.1..."
(0028,0010) US  Rows                         = 512
(0028,0011) US  Columns                      = 512
(0028,0030) DS  Pixel Spacing                = "0.703125\0.703125"
(0028,0100) US  Bits Allocated               = 16
(0028,0103) US  Pixel Representation         = 0  (unsigned)
(7FE0,0010) OW  Pixel Data                   = [binary pixel data]

Value Representations (VR)

VR	Name	Description
`CS`	Code String	Short uppercase string (max 16 chars): "CT", "MR", "M", "F"
`DA`	Date	8-char date: "YYYYMMDD"
`DS`	Decimal String	Floating-point as string: "0.703125"
`DT`	DateTime	14-26 char datetime string
`IS`	Integer String	Integer as string: "256"
`LO`	Long String	64-char string (patient ID, series description)
`PN`	Person Name	"Family^Given^Middle^Prefix^Suffix"
`SQ`	Sequence	Nested dataset (for complex structures)
`UI`	Unique Identifier	UID string: "1.2.840.10008.1.2.1"
`US`	Unsigned Short	2-byte unsigned int (0–65535)
`SS`	Signed Short	2-byte signed int (−32768–32767)
`UL`	Unsigned Long	4-byte unsigned int
`OW`	Other Word	Binary pixel data
`OB`	Other Byte	Binary data

Transfer Syntaxes

The Transfer Syntax UID (0002,0010) determines byte order and compression:

UID	Name	Byte Order	Compression
1.2.840.10008.1.2	Implicit VR Little Endian	Little	None (legacy)
1.2.840.10008.1.2.1	Explicit VR Little Endian	Little	None (default)
1.2.840.10008.1.2.2	Explicit VR Big Endian	Big	None (retired)
1.2.840.10008.1.2.4.50	JPEG Baseline (8-bit)	Little	JPEG lossy
1.2.840.10008.1.2.4.70	JPEG Lossless	Little	JPEG lossless
1.2.840.10008.1.2.4.90	JPEG 2000 Lossless	Little	JPEG 2000
1.2.840.10008.1.2.4.91	JPEG 2000 Lossy	Little	JPEG 2000
1.2.840.10008.1.2.5	RLE Lossless	Little	RLE

Explicit VR Little Endian (1.2.840.10008.1.2.1) is the default for modern DICOM files. Implicit VR (where VR must be inferred from the data dictionary) is a legacy format still found in older archives.

CT and MRI Pixel Data

CT Hounsfield Units (HU)

CT images store pixel values in Hounsfield Units, a standardized radiodensity scale:

Tissue	HU Range
Air	−1000
Lung	−600 to −400
Fat	−120 to −90
Water	0
Soft tissue	+20 to +80
Bone	+400 to +1000
Dense bone/metal	+1000 to +3000

Pixel values are stored as 12 or 16-bit integers. The DICOM tags (0028,1052) Rescale Intercept and (0028,1053) Rescale Slope convert stored values to HU:

HU = stored_pixel_value × RescaleSlope + RescaleIntercept

Typical: slope=1, intercept=−1024 → stored value 1024 = 0 HU (water).

Window Center and Width

Displaying HU-range data on 8-bit screens requires windowing:

(0028,1050) Window Center: The HU value mapped to 50% brightness
(0028,1051) Window Width: The HU range mapped to 0–100% brightness

Standard window presets:

Abdomen: W=350, C=40
Lung: W=1500, C=−600
Brain: W=80, C=40
Bone: W=2000, C=300

DICOM Networking (DIMSE Services)

DICOM is also a network protocol. The DIMSE (DICOM Message Service Element) defines services:

C-STORE: Send an image from one system to another
C-FIND: Query for studies/series/instances (like a database query)
C-MOVE: Request a remote system to send images to a third system
C-GET: Request images directly (pull from server)
C-ECHO: Ping (verify connection)

Each DICOM network entity is an AE (Application Entity) identified by its AE Title (up to 16 characters). DICOM runs over TCP/IP, typically port 104 (or 11112 for non-root).

WADO-RS (RESTful DICOM)

The modern API for DICOM retrieval is WADO-RS (Web Access to DICOM Objects, RESTful):

# Retrieve a study
GET /wado/rs/studies/{StudyInstanceUID}

# Retrieve a series
GET /wado/rs/studies/{StudyUID}/series/{SeriesUID}

# Retrieve a specific instance
GET /wado/rs/studies/{StudyUID}/series/{SeriesUID}/instances/{SOPInstanceUID}

# Retrieve rendered JPEG (thumbnail)
GET /wado/rs/studies/{StudyUID}/series/{SeriesUID}/instances/{SOPInstanceUID}/rendered?accept=image/jpeg

Working with DICOM Files

pydicom (Python)

import pydicom
import numpy as np
import matplotlib.pyplot as plt

# Load a DICOM file
ds = pydicom.dcmread('CT_slice.dcm')

# Access metadata
print(ds.PatientName)        # "Doe^John"
print(ds.Modality)           # "CT"
print(ds.StudyDate)          # "20240315"
print(ds.Rows, ds.Columns)   # 512, 512
print(ds.PixelSpacing)       # [0.703125, 0.703125]

# Get pixel array (applies RescaleSlope/Intercept)
pixel_array = ds.pixel_array  # raw stored values
hu_array = pixel_array * ds.RescaleSlope + ds.RescaleIntercept  # HU

# Apply windowing and display
window_center = 40   # abdomen
window_width  = 350
img_min = window_center - window_width / 2
img_max = window_center + window_width / 2
windowed = np.clip(hu_array, img_min, img_max)
windowed = (windowed - img_min) / (img_max - img_min) * 255

plt.imshow(windowed, cmap='gray')
plt.axis('off')
plt.savefig('ct_slice.png', bbox_inches='tight', pad_inches=0)

Converting DICOM to PNG/JPEG

import pydicom
from PIL import Image
import numpy as np

def dicom_to_png(dcm_path, png_path, window_center=40, window_width=400):
    ds = pydicom.dcmread(dcm_path)
    arr = ds.pixel_array.astype(float)

    # Apply rescale
    slope = getattr(ds, 'RescaleSlope', 1)
    intercept = getattr(ds, 'RescaleIntercept', 0)
    arr = arr * slope + intercept

    # Window
    img_min = window_center - window_width / 2
    img_max = window_center + window_width / 2
    arr = np.clip(arr, img_min, img_max)
    arr = ((arr - img_min) / (img_max - img_min) * 255).astype(np.uint8)

    Image.fromarray(arr).save(png_path)

dicom_to_png('CT_slice.dcm', 'output.png')

DCMTK Command-Line Tools

# Display DICOM tags
dcmdump CT_slice.dcm | head -50

# Convert DICOM to PNG (with auto windowing)
dcm2pnm --write-png CT_slice.dcm output.png

# Anonymize DICOM (remove patient data)
dcmodify -ie "(0010,0010)" -ie "(0010,0020)" -ie "(0010,0030)" CT_slice.dcm

# Change transfer syntax (compress with JPEG lossless)
dcmcjpeg CT_slice.dcm CT_compressed.dcm

Converting to NIfTI (Neuroimaging)

# dcm2niix — converts DICOM series to NIfTI for neuroimaging analysis
dcm2niix -z y -f "scan_%p_%t" -o ./output ./dicom_series/

Privacy and Anonymization

DICOM files contain extensive PHI (Protected Health Information). Before sharing DICOM files for research, education, or development, always anonymize:

Required removals (HIPAA Safe Harbor):

Patient name, ID, birth date, address
Study date (replace with days-since-epoch or offset)
Institution name, physician name
Device serial numbers
UIDs (replace with new generated UIDs)

Tools: DICOM Confidentiality Option (PS3.15), pydicom deidentify(), DCMTK dcmodify, Horos/OsiriX DICOM anonymizer, CTP (Clinical Trials Processor).

DICOM Viewers

Tool	Platform	Use Case
Horos	macOS	Clinical review, 3D reconstruction
OsiriX	macOS	Same (Horos fork)
3D Slicer	Cross-platform	Research, segmentation, 3D visualization
Weasis	Cross-platform (Java)	Web/desktop PACS viewer
dwv	Browser	JavaScript DICOM viewer
MicroDicom	Windows	Simple viewer
RadiAnt	Windows	Fast clinical viewer

DICOM's four-decade legacy as the universal medical imaging language — combined with active standardization — makes it essential knowledge for anyone working in health informatics, medical AI, or imaging system integration.

DICOM Medical Imaging Format: The Complete Guide