Guide

DICOM Format: The Medical Imaging Standard for CT, MRI, and Radiology

PC By Pablo Cirre • Updated May 7, 2026

What Is DICOM?

DICOM — Digital Imaging and Communications in Medicine — is the international standard for storing, transmitting, and displaying medical images and associated information. Developed jointly by the American College of Radiology (ACR) and the National Electrical Manufacturers Association (NEMA), DICOM 3.0 was released in 1993 and remains the dominant standard today, used by virtually every hospital, radiology department, and medical device manufacturer worldwide.

DICOM is two things in one: a file format for storing image data with rich patient and acquisition metadata, and a network protocol (built on TCP/IP) for transmitting images between PACS (Picture Archiving and Communication Systems), modalities (scanners), and viewing workstations. Every CT scanner, MRI machine, digital X-ray system, ultrasound unit, and PET scanner produces DICOM files.

The DICOM File Structure

A DICOM file has a fixed structure:

┌─────────────────────────────────────────────────┐
│ 128-byte File Preamble (usually all zeros)       │
├─────────────────────────────────────────────────┤
│ 4-byte DICM prefix ("DICM" in ASCII)             │
├─────────────────────────────────────────────────┤
│ File Meta Information Group (tags 0002,xxxx)     │
│   (0002,0000) MetaElementGroupLength             │
│   (0002,0001) FileMetaInformationVersion         │
│   (0002,0002) MediaStorageSOPClassUID            │
│   (0002,0003) MediaStorageSOPInstanceUID         │
│   (0002,0010) TransferSyntaxUID                  │
│   (0002,0012) ImplementationClassUID             │
├─────────────────────────────────────────────────┤
│ Data Set (all remaining data elements)           │
│   (0008,0016) SOPClassUID                        │
│   (0008,0018) SOPInstanceUID                     │
│   (0008,0060) Modality = "CT"                    │
│   (0008,103E) SeriesDescription                  │
│   (0010,0010) PatientName                        │
│   (0010,0020) PatientID                          │
│   (0010,0030) PatientBirthDate                   │
│   (0018,0050) SliceThickness                     │
│   (0020,0013) InstanceNumber                     │
│   (0028,0010) Rows                               │
│   (0028,0011) Columns                            │
│   (0028,0030) PixelSpacing                       │
│   (0028,1050) WindowCenter                       │
│   (0028,1051) WindowWidth                        │
│   (7FE0,0010) PixelData                          │
└─────────────────────────────────────────────────┘

Data Elements and Tags

Every piece of information in DICOM is a data element identified by a tag — a hexadecimal pair (group, element). Tags are the backbone of DICOM:

Tag	Keyword	Example Value
(0008,0016)	SOPClassUID	1.2.840.10008.5.1.4.1.1.2 (CT Image)
(0008,0060)	Modality	CT, MR, CR, DX, PT, US, NM
(0008,103E)	SeriesDescription	"HEAD CT W/O CONTRAST"
(0010,0010)	PatientName	Smith^John^A
(0010,0020)	PatientID	12345678
(0010,0030)	PatientBirthDate	19650312
(0018,0050)	SliceThickness	2.5 (mm)
(0018,0088)	SpacingBetweenSlices	2.5 (mm)
(0020,0013)	InstanceNumber	47 (slice number)
(0020,0032)	ImagePositionPatient	-245.7, -151.3, 112.0
(0020,0037)	ImageOrientationPatient	1,0,0,0,1,0 (axial)
(0028,0010)	Rows	512
(0028,0011)	Columns	512
(0028,0030)	PixelSpacing	0.625, 0.625 (mm)
(0028,0100)	BitsAllocated	16
(0028,1050)	WindowCenter	40 (Hounsfield units)
(0028,1051)	WindowWidth	400 (HU)
(0028,1052)	RescaleIntercept	-1024
(0028,1053)	RescaleSlope	1
(7FE0,0010)	PixelData	[binary pixel array]

Image Data and Hounsfield Units (CT)

CT images use Hounsfield Units (HU) — a calibrated scale where:

−1000 HU = air
−100 HU = fat
0 HU = water
700–3000 HU = bone/cortical
4000+ HU = metallic implants

Stored pixel values are encoded as unsigned integers. The actual HU value is calculated:

HU = (stored_pixel_value × RescaleSlope) + RescaleIntercept

For most CT: RescaleSlope = 1, RescaleIntercept = -1024, so a stored value of 1000 → 1000 × 1 + (−1024) = −24 HU (approximately water/soft tissue).

Windowing controls what part of the HU range is displayed as visible grayscale:

Window Center (WC) and Window Width (WW) define the display range
Common presets:
- Lung window: WC = −600, WW = 1500
- Brain: WC = 40, WW = 80
- Bone: WC = 400, WW = 1800
- Abdomen/soft tissue: WC = 40, WW = 400

Reading DICOM with Python (pydicom)

pydicom is the standard Python library for DICOM:

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

# Load a DICOM file
ds = pydicom.dcmread("CT_slice_047.dcm")

# Access metadata
print(f"Patient: {ds.PatientName}")
print(f"Patient ID: {ds.PatientID}")
print(f"Modality: {ds.Modality}")
print(f"Series: {ds.SeriesDescription}")
print(f"Image size: {ds.Rows} × {ds.Columns}")
print(f"Pixel spacing: {ds.PixelSpacing} mm")
print(f"Slice thickness: {ds.SliceThickness} mm")
print(f"Window Center: {ds.WindowCenter}, Width: {ds.WindowWidth}")

# Get the pixel array
pixels = ds.pixel_array      # shape (512, 512), dtype uint16

# Convert to Hounsfield Units
hu = pixels * float(ds.RescaleSlope) + float(ds.RescaleIntercept)
print(f"HU range: {hu.min():.0f} to {hu.max():.0f}")

# Apply windowing for display
wc = float(ds.WindowCenter)
ww = float(ds.WindowWidth)
img_min = wc - ww / 2
img_max = wc + ww / 2
windowed = np.clip(hu, img_min, img_max)
windowed = ((windowed - img_min) / ww * 255).astype(np.uint8)

plt.imshow(windowed, cmap="gray")
plt.title(f"CT slice {ds.InstanceNumber}")
plt.axis("off")
plt.savefig("ct_slice.png", dpi=150, bbox_inches="tight")

# Load an entire series (multiple slices)
import os

dicom_files = sorted(
    [os.path.join("ct_series/", f) for f in os.listdir("ct_series/") if f.endswith(".dcm")],
    key=lambda f: pydicom.dcmread(f).InstanceNumber
)

slices = [pydicom.dcmread(f).pixel_array for f in dicom_files]
volume = np.stack(slices, axis=0)   # shape (N_slices, 512, 512)
print(f"3D volume shape: {volume.shape}")

DICOM Anonymization

DICOM files contain patient-identifying information that must be removed before sharing for research:

import pydicom

def anonymize_dicom(input_path, output_path):
    ds = pydicom.dcmread(input_path)

    # Remove or replace identifying tags
    tags_to_clear = [
        (0x0010, 0x0010),   # PatientName
        (0x0010, 0x0020),   # PatientID
        (0x0010, 0x0030),   # PatientBirthDate
        (0x0010, 0x0040),   # PatientSex
        (0x0010, 0x1010),   # PatientAge
        (0x0008, 0x0020),   # StudyDate
        (0x0008, 0x0030),   # StudyTime
        (0x0008, 0x0090),   # ReferringPhysicianName
        (0x0008, 0x1030),   # StudyDescription
        (0x0008, 0x103E),   # SeriesDescription  (optional — may carry diagnostic info)
    ]

    for tag in tags_to_clear:
        if tag in ds:
            ds[tag].value = "ANONYMOUS"

    # Remove burned-in annotation pixel data
    ds.BurnedInAnnotation = "NO"

    ds.save_as(output_path)
    print(f"Anonymized: {output_path}")

anonymize_dicom("original.dcm", "anonymized.dcm")

For robust DICOM de-identification, use the DICOM PS 3.15 Appendix E profile — implemented in tools like dicom-anonymizer, dciodvfy, and Orthanc's de-identification plugin.

Converting DICOM to NIfTI for Neuroimaging

NIfTI (Neuroimaging Informatics Technology Initiative) is the standard format used by neuroimaging analysis software (FSL, FreeSurfer, SPM, ANTs, nilearn):

# dcm2niix — the gold standard DICOM → NIfTI converter
dcm2niix -o ./output/ -f "%p_%t_%s" -z y ./dicom_series/

# This produces:
# output/HeadCT_20240315_001.nii.gz  (compressed NIfTI)
# output/HeadCT_20240315_001.json    (BIDS sidecar with acquisition parameters)

dcm2niix correctly handles multi-frame DICOM, mosaic encoding (Siemens fMRI), and extracts diffusion gradient tables for DTI (b-values, b-vectors).

DICOM Viewing and PACS

3D Slicer: free, open-source medical image analysis platform — loads DICOM series, renders 3D volumes, performs segmentation
OsiriX / Horos (macOS): PACS workstation with curved MPR and 3D rendering
OHIF Viewer: web-based DICOM viewer via DICOMweb REST API (WADO-RS/STOW-RS)
Orthanc: open-source DICOM server for building local PACS
dwv: browser-based viewer (JavaScript) using pure web APIs

DICOM Transfer Syntaxes (Compression)

The Transfer Syntax UID determines how pixel data is compressed:

UID	Name	Notes
1.2.840.10008.1.2	Implicit VR Little Endian	Original; no explicit type tags
1.2.840.10008.1.2.1	Explicit VR Little Endian	Most common uncompressed
1.2.840.10008.1.2.4.70	JPEG Lossless (process 14)	Lossless, widely supported
1.2.840.10008.1.2.4.90	JPEG 2000 Lossless	Better compression
1.2.840.10008.1.2.4.91	JPEG 2000 Lossy	Smaller files, may not be acceptable for diagnosis
1.2.840.10008.1.2.5	RLE Lossless	Run-length encoding

Related conversions

Frequent conversions across the catalogue:

Frequently Asked Questions

A DICOM file (.dcm) stores one medical image plus all associated metadata in a single structured binary file. The metadata includes: patient information (name, ID, birth date), study information (date, time, institution, referring physician), acquisition parameters (slice thickness, pixel spacing, kV, mA for CT), and image display parameters (windowing, rescale slope/intercept). The actual image pixels are in the (7FE0,0010) PixelData element — typically 512×512 unsigned 16-bit integers for CT, or variable dimensions for MRI, ultrasound, and X-ray. One CT exam typically produces 200–2,000 individual DICOM files (slices).

A DICOM arquivo (.dcm) stores one medical image plus all associated metadata em um single structured binary file. The metadata includes: patient information (name, ID, birth date), study information (date, time, institution, referring physician), acquisition parameters (slice thickness, pixel spacing, kV, mA para CT), e image display parameters (windowing, rescale slope/intercept). The actual image pixels are no (7FE0,0010) PixelData element — tipicamente 512×512 unsigned 16-bit integers para CT, ou variable dimensions para MRI, ultrasound, e X-ray. One CT exam tipicamente produces 200–2,000 individual DICOM arquivos (slices).

A DICOM Datei (.dcm) stores one medical image plus all associated metadata in einem single structured binary file. The metadata includes: patient information (name, ID, birth date), study information (date, time, institution, referring physician), acquisition parameters (slice thickness, pixel spacing, kV, mA für CT), und image display parameters (windowing, rescale slope/intercept). The actual image pixels are im (7FE0,0010) PixelData element — typically 512×512 unsigned 16-bit integers für CT, oder variable dimensions für MRI, ultrasound, und X-ray. One CT exam typically produces 200–2,000 individual DICOM Dateien (slices).

A DICOM archivo (.dcm) stores one medical image plus all associated metadata en un single structured binary file. The metadata includes: patient information (name, ID, birth date), study information (date, time, institution, referring physician), acquisition parameters (slice thickness, pixel spacing, kV, mA para CT), y image display parameters (windowing, rescale slope/intercept). The actual image pixels are en el (7FE0,0010) PixelData element — typically 512×512 unsigned 16-bit integers para CT, o variable dimensions para MRI, ultrasound, y X-ray. One CT exam typically produces 200–2,000 individual DICOM archivos (slices).

On KaijuConverter every file is processed inside an isolated container, encrypted in transit (TLS 1.3) and at rest, and automatically deleted after 60 minutes with multi-pass overwrite. We never train on, share, or analyze user content. For maximum privacy on extremely sensitive material, prefer offline tools (ImageMagick, FFmpeg, LibreOffice) that you control end-to-end.

Hounsfield Units (HU) are the calibrated measurement scale for CT image data, named after Sir Godfrey Hounsfield who invented CT scanning. The scale anchors at: air = −1000 HU, water = 0 HU. Clinical ranges: fat (−120 to −80 HU), soft tissue (20–80 HU), blood (40–70 HU, higher when clotted), cortical bone (700–3000 HU), metallic implants (3000–4000+ HU). Raw DICOM pixel values are unsigned integers; the actual HU is: HU = pixel_value × RescaleSlope + RescaleIntercept. Windowing adjusts which part of the HU range maps to visible grayscale — different presets highlight different tissue types.

For 95% of use cases, yes — server-side ImageMagick, FFmpeg and LibreOffice produce identical output to the same tools on your laptop. Desktop software wins for: extremely large files (multi-GB), batch jobs of thousands of files, scripted pipelines, or content too sensitive to upload. KaijuConverter caps at 25 MB per file on the free tier (up to 2 GB on paid plans).

pydicom makes this scriptable in Python: read the file, set identifying tags (PatientName 0010,0010; PatientID 0010,0020; PatientBirthDate 0010,0030; StudyDate 0008,0020; ReferringPhysicianName 0008,0090) to "ANONYMOUS" or empty strings, and save. For thorough de-identification following DICOM PS 3.15 Appendix E, use the dicom-anonymizer Python package or the Orthanc DICOM server's anonymization API. Always check for burned-in annotations (text overlay in the pixel data) — these require pixel-level redaction. Generate new UIDs for anonymized files to avoid patient re-identification from UID patterns.

Pydicom makes this scriptable in Python: read o arquivo, set identifying tags (PatientName 0010,0010; PatientID 0010,0020; PatientBirthDate 0010,0030; StudyDate 0008,0020; ReferringPhysicianName 0008,0090) to "ANONYMOUS" ou empty strings, e save. para thorough de-identification following DICOM PS 3.15 Appendix E, usar the dicom-anonymizer Python package ou the Orthanc DICOM server's anonymization API. Always check para burned-in annotations (text overlay no pixel data) — these require pixel-level redaction. Generate new UIDs para anonymized arquivos to avoid patient re-identification de UID patterns.

Pydicom makes this scriptable in Python: read die Datei, set identifying tags (PatientName 0010,0010; PatientID 0010,0020; PatientBirthDate 0010,0030; StudyDate 0008,0020; ReferringPhysicianName 0008,0090) to "ANONYMOUS" oder empty strings, und save. für thorough de-identification following DICOM PS 3.15 Appendix E, verwenden the dicom-anonymizer Python package oder the Orthanc DICOM server's anonymization API. Always check für burned-in annotations (text overlay im pixel data) — these require pixel-level redaction. Generate new UIDs für anonymized Dateien to avoid patient re-identification von UID patterns.

Pydicom makes this scriptable in Python: read el archivo, set identifying tags (PatientName 0010,0010; PatientID 0010,0020; PatientBirthDate 0010,0030; StudyDate 0008,0020; ReferringPhysicianName 0008,0090) to "ANONYMOUS" o empty strings, y save. para thorough de-identification following DICOM PS 3.15 Appendix E, usar the dicom-anonymizer Python package o the Orthanc DICOM server's anonymization API. Always check para burned-in annotations (text overlay en el pixel data) — these require pixel-level redaction. Generate new UIDs para anonymized archivos to avoid patient re-identification de UID patterns.

Most format conversions are lossy by design — JPG, MP3, MP4, WebP all discard perceptual data to save bytes. Going through a lossy intermediate compounds the loss. To minimize visible/audible drift: convert from the original master, choose a higher quality setting, and avoid converting back and forth between lossy formats.

dcm2niix is the gold-standard tool: dcm2niix -o output/ -z y dicom_folder/ — it sorts slices by InstanceNumber, assembles them into a 3D volume, and exports compressed NIfTI (.nii.gz) with a JSON BIDS sidecar. For Python without converting to NIfTI: load all slices with pydicom, sort by InstanceNumber or ImagePositionPatient Z-coordinate, stack pixel arrays with numpy.stack(slices, axis=0) to get a (Z, Y, X) volume. For visualization and segmentation, 3D Slicer is the standard open-source tool — it loads entire DICOM series, renders 3D volume, and supports organ segmentation with AI-powered tools.

Dcm2niix is the gold-standard tool: dcm2niix -o output/ -z y dicom_folder/ — it sorts slices by InstanceNumber, assembles them em a 3D volume, e exports comprimido NIfTI (.nii.gz) com a JSON BIDS sidecar. para Python sem convertendo to NIfTI: load all slices com pydicom, sort by InstanceNumber ou ImagePositionPatient Z-coordinate, stack pixel arrays com numpy.stack(slices, axis=0) to get a (Z, Y, X) volume. para visualization e segmentation, 3D Slicer is the padrão open-source tool — it loads entire DICOM series, renders 3D volume, e suporta organ segmentation com AI-powered ferramentas.

Dcm2niix is the gold-standard tool: dcm2niix -o output/ -z y dicom_folder/ — it sorts slices by InstanceNumber, assembles them in einem 3D volume, und exports komprimiert NIfTI (.nii.gz) mit a JSON BIDS sidecar. für Python ohne umwandelnd to NIfTI: load all slices mit pydicom, sort by InstanceNumber oder ImagePositionPatient Z-coordinate, stack pixel arrays mit numpy.stack(slices, axis=0) to get a (Z, Y, X) volume. für visualization und segmentation, 3D Slicer is the Standard open-source tool — it loads entire DICOM series, renders 3D volume, und unterstützt organ segmentation mit AI-powered Werkzeuge.

Dcm2niix is the gold-standard tool: dcm2niix -o output/ -z y dicom_folder/ — it sorts slices by InstanceNumber, assembles them en a 3D volume, y exports comprimido NIfTI (.nii.gz) con a JSON BIDS sidecar. para Python sin convirtiendo to NIfTI: load all slices con pydicom, sort by InstanceNumber o ImagePositionPatient Z-coordinate, stack pixel arrays con numpy.stack(slices, axis=0) to get a (Z, Y, X) volume. para visualization y segmentation, 3D Slicer is the estándar open-source tool — it loads entire DICOM series, renders 3D volume, y soporta organ segmentation con AI-powered herramientas.

Yes — KaijuConverter accepts multiple files in a single drop and returns a ZIP. For very large batches (thousands of files) consider command-line tools or our API: <code>find . -name "*.heic" -exec magick {} {.}.jpg \;</code> or similar one-liners scale to millions of files when run locally.