Guide

NIfTI: The Standard Format for Brain MRI and fMRI Data

PC By Pablo Cirre • Updated May 7, 2026

NIfTI: The Standard Format for Brain MRI and fMRI Data

What Is NIfTI?

NIfTI — Neuroimaging Informatics Technology Initiative — is an open file format designed specifically for storing brain imaging data from MRI, fMRI, PET, DTI, and other neuroimaging modalities. Developed in 2004 by a working group under the NIH, NIfTI was created to replace the aging ANALYZE 7.5 format and establish a single, unambiguous standard for sharing volumetric brain data across research labs worldwide.

Today, NIfTI is the lingua franca of neuroimaging: virtually every major scanner vendor, analysis pipeline, and open dataset uses it. Files come in two flavors:

.nii — a single file containing both the header and the image data concatenated together
.hdr / .img paired files — the legacy two-file layout inherited from ANALYZE (header separate from raw voxel data)
.nii.gz — gzip-compressed .nii, by far the most common form in practice (reduces file size 60–80% for typical brain volumes)

Internal Structure of a NIfTI File

A .nii file begins with a 348-byte header (identical in size to ANALYZE 7.5 for backward compatibility), followed by a 4-byte extension indicator, optional extensions, and then the raw voxel data stored in row-major order (fastest index = x).

Key header fields you'll encounter:

Field	Bytes	Purpose
`sizeof_hdr`	4	Must be 348 (NIfTI-1) or 540 (NIfTI-2)
`dim[8]`	16	Array dimensions: dim[0]=rank, dim[1–7]=sizes
`pixdim[8]`	32	Voxel sizes in mm (and TR in seconds for dim[4])
`datatype`	2	Voxel data type: INT16, FLOAT32, COMPLEX64, etc.
`scl_slope` / `scl_inter`	8	Linear rescaling: stored_value × slope + intercept
`qform_code` / `sform_code`	4	Coordinate system declarations
`quatern_` / `qoffset_`	28	Quaternion-based scanner coordinates (qform)
`srow_x/y/z`	48	Affine matrix rows mapping voxels to mm (sform)

The sform and qform fields are what make NIfTI powerful: they encode the precise orientation and position of the volume in a real-world coordinate space (scanner-native, MNI152 standard space, Talairach, etc.), eliminating the left-right ambiguity that plagued ANALYZE.

NIfTI-2 (2011) extends dim and pixdim from 16-bit integers / 32-bit floats to 64-bit, enabling volumes larger than 32,767 voxels per dimension — critical for high-resolution 7T MRI datasets.

Coordinate Systems: qform vs. sform

NIfTI uses two optional coordinate transforms, each tagged with a code:

qform (codes 1–4): Encodes the scanner's acquisition geometry as a quaternion rotation plus three offsets. Code 1 = scanner anatomical coordinates; code 2 = aligned to another scan; codes 3–4 = Talairach or MNI space.
sform (code 1–4): A full 4×3 affine matrix — more flexible, used for arbitrary warps or after preprocessing to standard space.

The convention: if sform_code > 0, use sform; if only qform_code > 0, use qform. Software like FSL, FreeSurfer, and SPM all respect this hierarchy.

Voxel Data Types

NIfTI supports a rich set of datatype values:

UINT8 (2), INT16 (4), INT32 (8) — integer anatomical MRI (typical T1 from scanner is INT16)
FLOAT32 (16) — statistical maps (t-maps, z-maps, beta images from GLM)
FLOAT64 (64) — double-precision for intermediate computations
COMPLEX64 (32) / COMPLEX128 (1792) — raw k-space or spectroscopy data
RGB24 (128) — color-coded FA (fractional anisotropy) maps from DTI

The scl_slope and scl_inter fields let scanners store data as INT16 while preserving the full physical range in FLOAT after rescaling — a common pattern for scanner output.

Working with NIfTI in Python: nibabel

nibabel is the go-to Python library:

import nibabel as nib
import numpy as np

# Load a .nii.gz
img = nib.load('sub-01_T1w.nii.gz')

# Image data as NumPy array
data = img.get_fdata()   # returns float64, applies scl_slope/inter automatically
print(data.shape)        # e.g. (256, 256, 176) for a 1mm isotropic T1
print(img.header['pixdim'][1:4])  # voxel sizes in mm

# Affine matrix (4×4, sform or qform depending on codes)
aff = img.affine
print(aff)

# Modify and save
new_img = nib.Nifti1Image(data * 2, affine=aff, header=img.header)
nib.save(new_img, 'doubled.nii.gz')

# Access specific header fields
print(img.header.get_data_dtype())   # e.g. float64
print(img.header.get_zooms())        # (1.0, 1.0, 1.0, 2.0) — mm + TR

For 4D fMRI data (x, y, z, time), data.shape will be (64, 64, 36, 240) for a typical 6-minute run at 2×2×3 mm with TR=1.5 s and 240 volumes.

Standard Neuroimaging Pipelines That Use NIfTI

BIDS (Brain Imaging Data Structure) — the community standard for organizing NIfTI files: sub-01/ses-01/anat/sub-01_ses-01_T1w.nii.gz, sub-01/ses-01/func/sub-01_ses-01_task-rest_bold.nii.gz
fMRIPrep — preprocessing pipeline for fMRI; inputs and outputs exclusively NIfTI
FSL — Oxford's neuroimaging suite; native NIfTI throughout (FEAT, MELODIC, BET, FLIRT)
SPM — Statistical Parametric Mapping (MATLAB); uses NIfTI for all analyses since SPM8
FreeSurfer — cortical surface reconstruction; imports/exports NIfTI alongside MGH format
MRtrix3 — diffusion tractography; reads NIfTI, stores streamlines in .tck
AFNI — Analysis of Functional NeuroImages; supports NIfTI natively alongside BRIK/HEAD

Converting to NIfTI: dcm2niix

Raw MRI output from scanners is in DICOM. The standard tool for DICOM→NIfTI conversion is dcm2niix (Chris Rorden):

dcm2niix -z y -f "%p_%s" -o ./nifti/ ./dicom_folder/

Options: -z y produces .nii.gz; -f sets the filename template using DICOM metadata tags (protocol name %p, series number %s). dcm2niix also writes a .json sidecar with scanner parameters (TR, TE, FlipAngle, SliceTiming) for BIDS compliance.

Practical Tips

Always check orientation before analysis: use fslorient, fslhd, or nibabel to verify the sform/qform before feeding into pipelines. Wrong orientation is the single most common source of subtle errors.
Prefer .nii.gz for storage and sharing; most tools handle compressed NIfTI transparently via zlib.
Use fslinfo / fslhd (FSL) for quick header inspection on the command line.
NIfTI-1 dim limit: if any dimension > 32,767, you need NIfTI-2. This matters for ultra-high-res atlases and large group-level data arrays.
4D volumes: temporal fMRI data is a single 4D NIfTI — do not split into per-volume files unless specifically required.
MNI space: standard group analysis aligns individual brains to MNI152 template; post-registration NIfTI files should carry sform_code=4 and the MNI affine.

NIfTI's combination of simplicity, rich coordinate metadata, and universal toolchain support makes it the cornerstone of modern human and animal neuroimaging research.

Related conversions

Frequent conversions across the catalogue:

Frequently Asked Questions

.nii is a single uncompressed file containing both the 348-byte header and raw voxel data. .nii.gz is the same file compressed with gzip, which typically reduces size by 60–80% for brain volumes. Almost all modern neuroimaging tools read .nii.gz transparently, making it the preferred format for storage and sharing.

.nii is a single não comprimido arquivo containing both the 348-byte header e raw voxel data. .nii.gz is the same arquivo comprimido com gzip, which tipicamente reduces size by 60–80% para brain volumes. Almost all moderno neuroimaging ferramentas read .nii.gz transparentely, making it the preferred formato para storage e sharing.

.nii is a single uncompressed Datei containing both the 348-byte header und raw voxel data. .nii.gz is the same Datei komprimiert mit gzip, which typically reduces size by 60–80% für brain volumes. Almost all modern neuroimaging Werkzeuge read .nii.gz transparently, making it the preferred Format für storage und sharing.

.nii is a single uncompressed archivo containing both the 348-byte header y raw voxel data. .nii.gz is the same archivo comprimido con gzip, which typically reduces size by 60–80% para brain volumes. Almost all moderno neuroimaging herramientas read .nii.gz transparentely, making it the preferred formato para storage y sharing.

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.

Both encode the mapping from voxel indices to real-world millimeter coordinates. qform uses a quaternion rotation plus three offsets (describing scanner geometry). sform uses a full 4×3 affine matrix (more flexible, used after registration to standard space like MNI152). When sform_code > 0, software should use the sform; otherwise use the qform.

Both encode the mapping de voxel indices to real-world millimeter coordinates. qform uses a quaternion rotation plus three offsets (describing scanner geometry). sform uses a full 4×3 affine matrix (more flexible, used depois registration to padrão space like MNI152). When sform_code > 0, software should usar the sform; otherwise usar the qform.

Both encode the mapping von voxel indices to real-world millimeter coordinates. qform uses a quaternion rotation plus three offsets (describing scanner geometry). sform uses a full 4×3 affine matrix (more flexible, used nach registration to Standard space like MNI152). When sform_code > 0, Software should verwenden the sform; otherwise verwenden the qform.

Both encode the mapping de voxel indices to real-world millimeter coordinates. qform uses a quaternion rotation plus three offsets (describing scanner geometry). sform uses a full 4×3 affine matrix (more flexible, used después registration to estándar space like MNI152). When sform_code > 0, software should usar the sform; otherwise usar the qform.

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 500 MB per file (1 GB on paid plans).

Use dcm2niix, the standard tool: run `dcm2niix -z y -o ./nifti/ ./dicom_folder/`. The -z y flag produces .nii.gz files. dcm2niix also writes a JSON sidecar with scanner parameters (TR, TE, flip angle) needed for BIDS-compliant datasets.

Use dcm2niix, the padrão tool: run `dcm2niix -z y -o ./nifti/ ./dicom_folder/`. The -z y flag produces .nii.gz files. dcm2niix also writes a JSON sidecar com scanner parameters (TR, TE, flip angle) needed para BIDS-compliant datasets.

Use dcm2niix, the Standard tool: run `dcm2niix -z y -o ./nifti/ ./dicom_folder/`. The -z y flag produces .nii.gz files. dcm2niix also writes a JSON sidecar mit scanner parameters (TR, TE, flip angle) needed für BIDS-compliant datasets.

Use dcm2niix, the estándar tool: run `dcm2niix -z y -o ./nifti/ ./dicom_folder/`. The -z y flag produces .nii.gz files. dcm2niix also writes a JSON sidecar con scanner parameters (TR, TE, flip angle) needed para BIDS-compliant datasets.

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.

nibabel is the standard library: `import nibabel as nib; img = nib.load("brain.nii.gz"); data = img.get_fdata()`. get_fdata() returns a float64 NumPy array with scl_slope/scl_inter rescaling applied automatically. The affine matrix is in img.affine and header fields are accessible via img.header.

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.