Skip to content

Neuroimaging Informatics Technology Initiative

Personal tools
You are here: Home » NIfTI-1 Data Format » Documentation » NIfTI-1 FAQ


Document Actions
Brought to you by the members of the Data Format Working Group

Q1. What are NIfTI and the DFWG? (Mike Huerta)

The purpose of the Neuroimaging Informatics Technology Initiative (NIfTI) is to support service, training, and research to develop and enhance the utility of informatics tools used in neuroimaging, with a focus on functional magnetic resonance imaging (fMRI). NIfTI is jointly sponsored by the National Institute of Mental Health (NIMH) and the National Institute of Neurological Disorders and Stroke (NINDS), which are part of the National Institutes of Health (NIH) and the Department of Health and Human Services (DHHS). Further background can be found at this link. NIfTI was developed and has been implemented with input from the community of neuroscientists who use neuroimaging in their research, and from the community of developers of informatics tools used to make sense of neuroimaging data (with an emphasis on fMRI).

There are two major activities of NIfTI, thus far. The first is the support of a cluster of grants to characterize, compare, and validate informatics tools used in fMRI studies.

The second is the support of the Data Format Working Group. The DFWG is charged with arriving at a technical solution to the problem of multiple data formats used in fMRI research. This was identified by many investigators as a fundamental obstacle to fMRI fulfilling the promise it has to elucidate brain function. The solution arrived at by the DFWG will not be imposed on the research community, but several groups that develop fMRI-related software have already indicated their willingness to adopt the solution. Suggestions for membership on the DFWG were broadly solicited from the research community.

The DFWG members are:

  • Chair: Stephen C. Strother (Veterans Affairs Medical Center, Minneapolis, Minnesota)
  • John Ashburner (Functional Imaging Laboratory, London, United Kingdom)
  • Hester Breman (Brain Innovation B.V., Maastricht, The Netherlands)
  • Robert W. Cox (National Institute of Mental Health, NIH, DHHS, Bethesda, Maryland)
  • Kate Fissell (University of Pittsburgh, Pittsburgh, Pennsylvania)
  • Christian Haselgrove (Massachusetts General Hospital, Charlestown, Massachusetts)
  • Colin J. Holmes (Silicon Graphics, Inc., Mountain View, California)
  • Peter Kochunov (University of Texas Health Science Center, San Antonio, Texas)
  • David E. Rex (University of California, Los Angeles)
  • Stephen M. Smith (Oxford University Centre for Functional MRI of the Brain, United Kingdom)
  • Jeffrey B. Woodward (Dartmouth College, Hanover, New Hampshire).
  • ex officio: Michael F. Huerta (National Institute of Mental Health, NIH, DHHS)
  • ex officio: Yuan Liu (National Institute of Neurological Disorders and Stroke, NIH, DHHS)

It should be noted that in no way is the NIfTI-1 format an official format whose use is mandated by the NIH. This effort has been made to help developers and users of functional neuroimaging software. What use is made of the format is now up to the community.

Q2. Why didn't the committee choose the DICOM format? The MINC-2 format? (Steve Strother)

DICOM: The DICOM standard attempts to address the extremely general requirements of digital imaging in diagnostic and therapeutic healthcare environments (, while the DFWG has a much narrower focus. The DFWG committee's strategy has been to adopt a two-pronged approach to developing a technical solution to the problem of multiple fMRI data formats by initially focusing on a relatively simple format that has low storage overhead, resolves some immediate format problems in the fMRI community and is not difficult for developers to learn and use. The NIfTI-1 format specification represents the simple first prong of this approach with the goal of consolidating the variants of the ANALYZE 7.5 format currently being used by the major fMRI software packages while staying mostly ANALYZE 7.5 compatible and explicitly dealing with the specification and storage of a data coordinate system.

The committee felt that DICOM did not satisfy the general requirements of a simple format for the fMRI community with DICOM's large, clinically focused storage overhead and the relatively complex specifications for multiframe MRI and spatial registration. In addition, the DICOM specification is also heavily concerned with data communication, something the committee felt was well beyond the scope of NIfTI-1.

MINC-2: The DFWG committee is very aware of the latest MINC developments through Colin Holmes' able and enthusiastic advocacy of MINC to the committee. MINC is being actively and seriously considered as a candidate for a NIfTI-2 format specification (see below). In addition, we have been provided with MINC 1.0 namespace (7/5/2003) and MINC 2.0 file format (11/18/2003) documents by Robert Vincent in Montreal. Most of the DFWG committee meet with Alan Evans, John Sled, Robert Vincent and several others to discuss MINC and its proposed developments at the last OHBM meeting in New York.

NIfTI-2: The second prong of the committee's approach (NIfTI-2) will be a more complex format specification to address the problem of comprehensive data interchange within the fMRI research community. Development of NIfTI-2 and inclusion of specifications for such data as nonlinear registrations, surfaces and various types of metadata (e.g., data analysis design matrices) are under discussion by the committee.

Q3. Can you add my favorite variable X to the header? (Jeff Woodward)

[Temporary answer by RW Cox]
Not very likely. We stayed within the restriction of a 348 byte header to keep compatibility with ANALYZE 7.5. Most of the space is now taken by new fields, agreed upon by the DFWG committee after extensive wrangling.

Q4. What is the minimum set of fields required to create a NIfTI-1 header? (Christian Haselgrove)

While not all of the NIfTI-1 header fields are necessary to describe a volume, most fields require a specific value to indicate the field's non-use.

Required fields are:

    FIELD         NOTES
sizeof_hdr must be 348
dim dim[0] and dim[1] are always required;
dim[2] is required for 2-D volumes,
dim[3] for 3-D volumes, etc.
datatype needed to specify type of image data
bitpix should correspond correctly to datatype
pixdim with the exception of pixdim[0] (which
is required when qform_code != 0),
pixdim[n] is required when dim[n] is
vox_offset required for an "n+1" header
magic must be "ni1\0" or "n+1\0"
The descrip and aux_file fields are not required. The NIfTI-1 standard does not specify any use or interpretation of these fields.

The remaining fields are not required, but their non-use must be indicated as in the following table. The simplest way to ensure this when creating a 348-byte header structure is to fill it with zeros (e.g., with calloc()) before setting any desired fields to their appropriate values.

intent_code intent_code = NIFTI_INTENT_NONE = 0
intent_name and intent_name[0] = '\0'
scl_slope scl_slope = 0
xyzt_units xyzt_units = NIFTI_UNITS_UNKNOWN = 0
cal_max cal_max = 0 and cal_min = 0
toffset toffset = 0
slice_code slice_code = 0 and dim_info = 0
qform_code qform_code = 0 (see note below)
sform_code sform_code = 0
N.B.: Using qform_code=0 is legal, but not recommended. This case is intended for the case when you don't want to encode any orientation information in the header. It is mainly intended for compatibility with older ANALYZE 7.5 files; however, the ANALYZE 7.5 orient field is not contained in a NIfTI-1 header. If qform_code=0, a program may treat the orientation of a NIfTI-1 file in any way; this is unlikely to be portable or inter-operable.

Q5. What format converters are available? (David Rex)

Q6. What I/O libraries are available? (Kate Fissell)

A C/C++ I/O library for NIfTI-1 is available at

  • nifti1.h -- The header file for the NIfTI-1 format.
  • nifti1_io.c -- The I/O functions for the NIfTI-1 format.
  • nifti1_io.h -- The header file for the I/O functions for the NIfTI-1 format.
  • nifti1_test.c -- A short sample program to read and write NIfTI-1 files.
These files were contributed by Robert Cox, NIH, and contain functions to read and write NIfTI-1 headers and data blocks, print header fields, and, functions to manipulate some of the more complicated data items in the NIfTI-1 header (e.g., transform matrices).

We expect that I/O libraries in MATLAB and Java will be available in the not too distant future. Any and all contributed libraries are welcome.

Q7. How do I tell if an ANALYZE header is NIfTI-1 compatible? (Christian Haselgrove)

NIfTI-1 headers must contain a specific sequence of bytes in the last four bytes of the header (bytes 344-348):

  • When there is a separate file for the image data, these bytes must be "ni1\0" (hexadecimal 6e 69 31 00).
  • When the image data is in the same file as the header, these bytes must be "n+1\0" (hexadecimal 6e 2b 31 00).

If neither of these conditions is satisfied, then a program should not presume that the NIfTI-1 additions/changes to the ANALYZE format are present in the header.
Q8. Is NIfTI-1 compatible with the ANALYZE 7.5 file format? (Jeff Woodward)

[Temporary answer by RW Cox]
We hope so. The interpretation of many of the fields in the ANALYZE 7.5 header is not very clear. Most such fields aren't widely used by software other than ANALYZE itself. After considerable discussion, we kept the ANALYZE 7.5-defined fields that seemed widely used in the FMRI community. The goal was to have non-NIfTI-aware FMRI packages that use the ANALYZE 7.5 format still be able to read most NIfTI-1 files. Time and experience will tell if this goal has been met.

Q9. Do I need to convert all of my files to this format? (Colin Holmes)

No. NIfTI-1 was created to provide a common and widely acceptable version of the ANALYZE-7.5 format that could act as an image data interchange format between neuroimaging programs and analysis packages. As it provides little support for storage of required metadata (e.g., image acquisition parameters), it is not a complete solution for long-term archiving of neuroimaging data. For this function, please see the NIfTI-2 format (coming soon to a lab near you).

Q10. Is the NIfTI-1 header extensible? (Hester Breman)

The NIfTI-1 files can exist in two forms. There can be a single *.nii file or two separate files, a *.hdr and a *.img file (see the NIfTI-1 header file for more details).

When the information is separated into a *.hdr and *.img file, which resembles the original ANALYZE 7.5 format, any programs that read the ANALYZE format will expect a header of 348 bytes. Therefore the NIfTI-1 header is not extensible in this form, since any program is likely to fail when it detects that the header file is not the expected size.

While it is possible to store data between the header and the image part of a single *.nii file, this is non-standard and no other program will will be able to interpret these data. Therefore it is not recommended to produce such files.

In the current NIfTI-1 format, there are many possibilities to store different kinds of data in the image part of the file. Datasets, many types of statistical values, vectors, meshes, etc., all can be stored in NIfTI-1. Each NIfTI-1-aware program will be able to read these files. Interpretation of the files will also be possible, since the meaning of the fields is determined, when the intent of the file is known.

For any other types of data that cannot be stored in the NIfTI-1 format, eventually a NIfTI-2 format for complex data is to come.

Q11. What modalities does this format support? (Colin Holmes)

NIfTI-1 can store image data from any modality such as PET, MRI, CT, EEG that produces regularly sampled 1-5D rasters. No support is provided for modality-specific metadata such as dose, pulse sequence or indeed the majority of subject oriented descriptive items.

A limited amount of information specific to FMRI time series acquisition is available in the NIfTI-1 header:

  • Marking the frequency, phase, and slice encoding directions.
  • Storing the per-slice acquisition time.
  • Storing the slice acquisition order (e.g., interleaving).
This particular information is present since the primary thrust of the DFWG has been to make FMRI data analysis software inter-operable.

Q12. What types of data can I store in the format? (Colin Holmes)

NIfTI-1 provides support for regularly sampled raster-oriented data of up to five dimensions (e.g., image data). The data type of each value stored in the raster grid must be the same, and is indicated in the datatype header field. These data types are intended to encompass a wide range of commonly usages. Expressed in C language syntax, these basic data types are:

   unsigned char      = 8-bit byte     [     0..  255]
signed char = 8-bit byte [ -128.. 127]
signed short = 16-bit integer [-32768..32767]
unsigned short = 16-bit integer [ 0..65535]
signed int = 32-bit integer
unsigned int = 32-bit integer
signed long long = 64-bit integer
unsigned long long = 64-bit integer

float = 32-bit floating point value
double = 64-bit floating point value
long double = 128-bit floating point value

{ float r,i ; } = 64-bit complex = 2 32-bit floats
{ double r,i ; } = 128-bit complex = 2 64-bit floats
{long double r,i;} = 256-bit complex = 2 128-bit floats

{unsigned char r,g,b;} = RGB triple = 3 8-bit bytes
Not all data types will be supported by all programs; in fact, most programs will probably only operate on a limited set of these data types. A well-crafted NIfTI-1 compatible program should detect when it cannot deal with an input file's data type, and exit gracefully.

The first 3 dimensions are reserved for space; the fourth dimension for time (or frequency); the fifth dimension is intended to allow storing multiple values at each spatio-temporal location. In addition, the intent_code field can be used to describe if and how the fifth dimension of the image data is organized. Possible organizations include vector, general matrix, symmetric matrix, and quaternion.

NIfTI-1 does not provide for the storage of data sets defined over irregularly sampled domains, such as cortical surface models. Nor does NIfTI-1 provide for storage of the definitions of such domains.

Q13. Does NIfTI-1 solve the problem of data interchange in fMRI? (Colin Holmes)

Yes and no. NIfTI-1 can act as a conduit for image data (only) being exchanged between programs in an analysis sequence. It does not provide for the long-term and complete storage of image and metadata and associated computed results (e.g., surfaces) that is necessary to provide for the complete and robust preservation of data for historical, archival or full-fledged interlaboratory exchange of neuroimaging data.

Q14. What is the default orientation? (John Ashburner)

The NIfTI format allows storage on disk to be in either a left- or right-handed coordinate system. However, the format includes an implicit spatial transformation into a RIGHT-HANDED coordinate system. This transform maps from data coordinates (e.g. column i, row j, slice k), into some real world (x,y,z) positions in space. These positions could relate to Talairach-Tournoux (T&T) space, MNI space, or patient-based scanner coordinates.

For T&T, and MNI coordinates, x increases from left to right, y increases from posterior to anterior, and z increases in the inferior to superior direction. Directions in the scanner coordinate system are similar. MRI data is usually exported as DICOM format, which encodes the positions and orientations of the slices. When data are converted from DICOM to NIfTI-1 format, the relevant information can be determined from the Pixel Spacing, Image Orientation (Patient) and Image Position (Patient) fields of the DICOM files.

NIfTI-1 also allows the space of one image to be mapped to that of another (via a rigid or affine transform). This is to enable on-the-fly resampling of registered images. This would allow intra-subject images, collected with lots of different orientations or resolutions, to be treated as if they are all in register.

Q15. Is the default radiological or neurological orientation? (John Ashburner)

Neurological and radiological conventions only relate to visualization of axial images. They are unrelated to how the data are stored on disk, or even how the real-world coordinates are represented.

It is more appropriate to consider whether the real-world coordinate system is left- or right-handed (see below). Talairach and Tournoux use a right-handed system, whereas the storage convention of ANALYZE files is usually considered as left-handed. These coordinate systems are mirror images of each other (if you are a psychologist, try explaining why mirror images appear to be left-right flipped, rather than flipped up-down, or back-front). Transforming between left- and right-handed coordinate systems involves flipping, and can not be done by rotations alone.

Q16. What is a quaternion? (Bob Cox)

A quaternion is a mathematical object that contains 4 real numbers. Historically, quaternions were discovered* by William Rowan Hamilton in 1843. This antedates the discovery of vectors and matrices, which are more widely taught in modern times.

Like real and complex numbers, quaternions can be added and multiplied. Unlike real and complex numbers, the order in which quaternions are multiplied makes a difference. In Hamilton's notation, which is still used, a quaternion is represented as the sum of 4 real numbers times the basis symbols 1, i, j, and k; for example:

   q = a 1 + b i + c j + d k
p = u 1 + v i + w j + x k
Quaterions are added component by component. Multiplication is done via the following rules:
  1 * X =  X   [for any X]
i * j = k
j * i = -k [anti-commutative]
j * k = i
k * j = -i [anti-commutative]
k * i = j
i * k = -j [anti-commutative]

i * i = -1 [each of i, j, and k ]
j * j = -1 [is a square root of -1]
k * k = -1
So the example quaternions above multiply out to
  q*p = (a*u-b*v-c*w-d*x) 1 + (a*v+b*u+c*x-d*w) i
+(a*w-b*x+c*u+d*v) j + (a*x+b*w-c*v+d*u) k
A principal application of quaternions is to represent 3D rotations. How this is done is described in great detail in the NIfTI-1 C header, and also here. The reason this representation is used in NIfTI-1 is to save space in the 348-byte header; a quaternion only requires 4 floats (16 bytes) to specify, but a 3x3 matrix requires 9 floats (36 bytes). The nifti1_io.c I/O library contains functions to convert quaternions to rotation matrices, and vice-versa.
*I am a strict Platonic Idealist in these matters.
Q17. What alignment information is stored in the NIfTI-1 header? (Christian Haselgrove)

Volume orientation is given by a transformation that maps voxel indices (i,j,k) to spatial coordinates (x,y,z), typically anatomical coordinates assigned by the scanner. This transformation ("Method 2" in the nifti1.h documentation) is generated using the voxel dimensions, a quaternion encoding a rotation matrix, and a 3D shift, all stored in the NIfTI-1 header; details can be found in the nifti1.h comments.

The NIfTI-1 header also provides for a general affine transformation, separate from that described by Method 2. This transformation ("Method 3") also maps voxel indices (i,j,k) to (x,y,z), which in this case are typically coordinates in a standard space such as the Talairach space. The elements of this transformation matrix are stored in the NIfTI-1 header. For example, the Method 2 transformation can be constructed from the attributes from a set of DICOM files; the Method 3 transform can be computed offline and inserted into the header later.

The exact "meaning" of the coordinates given by the Method 2 and Method 3 transformations is recorded in header fields qform_code and sform_code, respectively. Code values can indicate if the (x,y,z) axes are

  • Anatomical coordinaes from the scanner (e.g., the DICOM header)
  • Aligned to some anatomical "truth" or standard
  • Aligned and warped to Talairach-Tournoux coordinates
  • Aligned and warped to MNI-152 coordinates

It is possible that neither transformation is specified (i.e., qform_code=sform_code=0), in which case we are left with the voxel size in pixdim[], and no orientation is given or assumed. This use ("Method 1") is discouraged.

Q18. Does NIfTI-1 support a single file format for header and data? (Bob Cox)

Yes. In this format, the 348 byte header comes first in the file. The image data follows. The image data offset in the file is specified by the vox_offset field in the header. Note that the magic field for a single file is "n+1", whereas that for a double file is "ni1". This means that it is the contents of the header that determines if the data is stored in the same file as the header or a different file. The filename itself (*.hdr or *.nii) does not necessarily specify the double or single storage format.

In principle, one could put arbitrary information between the header and the image data, by making vox_offset be larger than 352. This is not recommended. Such a file would not be compatible or portable, which is the entire point of the NIfTI-1 effort.
However: also see Q21!

We recommend the use of the suffix .nii for a single header+data NIfTI-1 format file. This suffix does not have any previous meaning attached, unlike the more obvious .nif (Navy Image File). For more information on connotations of various suffixes that start with "n", see

Q19. Why does NIfTI-1 allow for two coordinate systems (the qform and sform)? (Mark Jenkinson)

The basic idea behind having two coordinate systems is to allow the image to store information about (1) the scanner coordinate system used in the acquisition of the volume (in the qform) and (2) the relationship to a standard coordinate system - e.g. MNI coordinates (in the sform).

The qform allows orientation information to be kept for alignment purposes without losing volumetric information, since the qform only stores a rigid-body transformation which preserves volume. On the other hand, the sform stores a general affine transformation which can map the image coordinates into a standard coordinate system, like Talairach or MNI, without the need to resample the image.

By having both coordinate systems, it is possible to keep the original data (without resampling), along with information on how it was acquired (qform) and how it relates to other images via a standard space (sform). This ability is advantageous for many analysis pipelines, and has previously required storing additional files along with the image files. By using NIfTI-1 this extra information can be kept in the image files themselves.

Note: the qform and sform also store information on whether the coordinate system is left-handed or right-handed (see Q15) and so when both are set they must be consistent, otherwise the handedness of the coordinate system (often used to distinguish left-right order) is unknown and the results of applying operations to such an image are unspecified.

Q20. How is the timing of acquisition encoded in the NIfTI-1 header? (Bob Cox)

For a few purposes, it is important to know the order and timing of the MRI slice acquisition. The NIfTI-1.1 header contains several fields that can be used to specify this type of information:

  • freq_dim, phase_dim, and slice_dim
    These fields encode which spatial dimension (1,2, or 3) corresponds to which acquisition dimension for MRI data. Two examples:
    • Rectangular scan multi-slice EPI:
       freq_dim = 1  phase_dim = 2  slice_dim = 3
      (or some permutation of these)
    • Spiral scan multi-slice EPI:
      freq_dim = phase_dim = 0  slice_dim = 3
      since the concepts of frequency- and phase-encoding directions don't apply to spiral scan
    • In many cases, freq_dim=1, phase_dim=2, and slice_dim=3 will make sense; however, the provision of these codes means that it isn't necessary to store the slices in the NIfTI-1.1 in the same physical layout in which they were acquired.
    • The fields freq_dim, phase_dim, slice_dim are all squished into the single byte field dim_info (2 bits each, since the values for each field are limited to the range 0..3). This unpleasantness is due to lack of space in the 348 byte allowance.
    • The C macros DIM_INFO_TO_FREQ_DIM, DIM_INFO_TO_PHASE_DIM, and DIM_INFO_TO_SLICE_DIM can be used to extract these values from the dim_info byte.
    • The C macro FPS_INTO_DIM_INFO can be used to put these 3 values into the dim_info byte.
  • slice_duration
    If this value is positive, and if slice_dim is nonzero, then slice_duration indicates the amount of time used to acquire one slice.
    • slice_duration * dim[slice_dim] can be less than pixdim[4] with a clustered acquisition method, for example.
  • slice_code
    If this value is nonzero, and if slice_dim is nonzero, and if slice_duration is positive, then slice_code indicates the timing pattern of the slice acquisition. The following codes are defined:
    • NIFTI_SLICE_SEQ_INC == sequential increasing
    • NIFTI_SLICE_SEQ_DEC == sequential decreasing
    • NIFTI_SLICE_ALT_INC == alternating increasing
    • NIFTI_SLICE_ALT_DEC == alternating decreasing
    • NIFTI_SLICE_ALT_INC2 == alternating increasing #2
    • NIFTI_SLICE_ALT_DEC2 == alternating decreasing #2
  • slice_start and slice_end
    These values indicate the start and end of the slice acquisition pattern, when slice_code is nonzero.
    • These values are present to allow for the possible addition of "padded" slices at either end of the volume, which don't fit into the slice timing pattern.
    • If there are no padding slices, then slice_start=0 and slice_end=dim[slice_dim]-1 are the correct values.
    • For these values to be meaningful, slice_start must be non-negative and slice_end must be greater than slice_start and less than dim[slice_code].
    • The following table indicates the slice timing pattern, relative to time=0 for the first slice acquired, for some sample cases. Here, dim[slice_dim]=7 (there are seven slices, labeled 0..6), slice_duration=0.1, and slice_start=1, slice_end=5 (one padded slice on each end).
      6 n/a n/a n/a n/a n/a n/a
      5 0.4 0.0 0.2 0.0 0.4 0.2
      4 0.3 0.1 0.4 0.3 0.1 0.0
      3 0.2 0.2 0.1 0.1 0.3 0.3
      2 0.1 0.3 0.3 0.4 0.0 0.1
      1 0.0 0.4 0.0 0.2 0.2 0.4
      0 n/a n/a n/a n/a n/a n/a
      'n/a' = 'not applicable'
    • The SEQ values for slice_code are sequential ordering (uncommon but not unknown), either increasing in slice number or decreasing (INC or DEC), as illustrated above.
    • The ALT values for slice_code are alternating ordering.
      • The 'standard' way for these to operate (without the '2' on the end) is for the slice timing to start at the edge of the slice_start .. slice_end group (at slice_start for INC and at slice_end for DEC).
      • With the ALT_*2 values for slice_code, the slice timing instead starts at the first slice in from the edge (at slice_start+1 for INC2 and at slice_end-1 for DEC2). This latter acquisition scheme is found on some Siemens scanners.
No doubt this is all a little confusing. There are threee concepts to grasp:
  1. ordering of the spatial dimensions in the image array relative to acquisition,
  2. time it takes to acquire a single slice,
  3. and time-ordering of the slice acquisition within the slice_dim direction (including the possibility of 'padded' slices that don't have an acquisition time order).

Q21. How can I add customized (extension) data to the NIfTI-1 header? (Bob Cox)

There are two phases to this answer: clarifying where the image data is stored, and then discussing the NIfTI-1.1 approved way of storing information after the 348 byte header (and before the image data in a .nii file).

Clarifying vox_offset:
Of primary importance is the issue of where extended header data will be stored, and how a program is to know what is header data and what is image data. In NIfTI-1.1, the location of the image data is specified by the vox_offset header field. There are two cases:
  • In a .nii file, the vox_offset field value is the start location of the image data bytes in that file.
  • In a .hdr/.img file pair, the vox_offset field value is the start location of the image data bytes in the .img file.
All data at and after the vox_offset location is taken to be image data. Furthermore, the following limitations apply to the value of vox_offset:
  • If vox_offset is less than 352 in a .nii file, it is equivalent to 352 (i.e., image data never starts before byte #352 in a .nii file).
  • The default value for vox_offset in a .nii file is 352; in a .hdr file it is 0. The default value applies if the header vox_offset value is illegal (e.g., non-positive).
  • vox_offset should be an integer multiple of 16; otherwise, some programs may not work properly (e.g., SPM). This constraint is to allow memory-mapped input to be properly byte-aligned. However, this multiple-of-16 constraint is a suggestion, not a strict requirement.
Note that since vox_offset is an IEEE-754 32-bit float (for compatibility with the ANALYZE-7.5 format), it effectively has a 24-bit mantissa. All integers from 0 to 224 can be represented exactly in this format, but not all larger integers are exactly storable as IEEE-754 32-bit floats. However, unless you plan to have vox_offset be potentially larger than 16 MB, this limitation should not be an issue. (Actually, any integral multiple of 16 up to 227 can be represented exactly in the IEEE-754 format, which allows for up to 128 MB of random information before the image data. If this limit is not enough, perhaps the NIfTI-1.1 format is not right for your application.)

In a .img file (i.e., image data stored separately from the NIfTI-1.1 header), data bytes between #0 and #vox_offset-1 (inclusive) are completely undefined and unregulated by the NIfTI-1.1 standard. One potential use of having vox_offset > 0 in the .hdr/.img file pair storage method is to make the .img file be a copy of (or link to) a pre-existing image file in some other format, such as DICOM; then vox_offset would be set to the offset of the image data in this file. (It may not be possible to follow the multiple-of-16 rule with an arbitrary external file; using the NIfTI-1.1 format in such a case may lead to a file that is incompatible with software that relies on vox_offset being a multiple of 16.)

In a .nii file, data bytes between #348 and #vox_offset-1 (inclusive) may be used to store user-defined extra information; similarly, in a .hdr file, any data bytes after byte #347 are available for user-defined extra information. The (very weak) regulation of this extra header data is described next.

Extended Header Data Section(s):
The NIfTI-1.1 header struct is 348 bytes long. In a .nii file, the image data can start no earlier than byte #352. The four bytes immediately following the NIfTI-1.1 header struct (bytes #348-351 inclusive) should be considered to be declared as follows:

 unsigned char extension[4] ;
After the end of the 348 byte header (e.g., after the magic field), the next 4 bytes are an byte array field named extension. By default, all 4 bytes of this array should be set to zero. In a .nii file, these 4 bytes will always be present, since the earliest start point for the image data is byte #352. In a separate .hdr file, these bytes may or may not be present (i.e., a .hdr file may only be 348 bytes long). If not present, then a NIfTI-1.1 compliant program should use the default value of extension={0,0,0,0}. The first byte (extension[0]) is the only value of this array that is specified at present. The other 3 bytes are reserved for future use.

If extension[0] is nonzero, it indicates that extended header information is present in the bytes following the extension array. In a .nii file, this extended header data is before the image data (and vox_offset must be set correctly to allow for this). In a .hdr file, this extended data follows extension and proceeds (potentially) to the end of the file.

The format of extended header data is only weakly specified. Each extension must be an integer multiple of 16 bytes long. The first 8 bytes of each extension comprise two 4-byte integers:

  int esize , ecode ;
These values may need to be byte-swapped, as indicated by dim[0] for the rest of the NIfTI-1.1 header.
  • esize is the number of bytes that form the extended header data
    • esize must be a positive integral multiple of 16
    • this length includes the 8 bytes of esize and ecode themselves
  • ecode is a non-negative integer that indicates the format of the extended header data that follows
In the interests of interoperability (a primary rationale for NIfTI), groups developing software that uses this extension mechanism are encouraged to document and publicize the format of their extensions. To this end, the NIfTI DFWG will assign even numbered codes upon request to groups submitting at least rudimentary documentation for the format of their extension; at present, the contact is The assigned codes and documentation will be posted on the NIfTI website. All odd values of ecode (and 0) will remain unassigned (at least, until the even ones are used up, when we get to 2,147,483,646).

Also in the interests of interoperability, any extension data should be checked for integrity. For example, an XML-formatted extension should be checked that it begins with the string "<?xml "; if not, the software should skip it. In this way, databases of NIfTI-1.1 files that wish to anonymize files can simply fill an extension field with zero bytes to wipe out any information that might identify a subject. In such a case, the anonymizing software should change the ecode value to NIFTI_ECODE_IGNORE, but programs should be prepared to deal with zero-ed out extensions that are still marked with a nonzero ecode.

Note that the other contents (past ecode and esize) of the extended header data section are totally unspecified by the NIfTI-1.1 standard. In particular, if binary data is stored in such a section, its byte order is not necessarily the same as that given by examining dim[0]; it is incumbent on the programs dealing with such data to determine the byte order of binary extended header data.

Multiple extended header sections are allowed, each starting with an esize,ecode value pair. The first esize value, as described above, is at bytes #352-355 in the .hdr or .nii file (files start at byte #0, just to be clear). If this esize1 value is positive, then the second (esize2) will be found starting at byte #352+esize1, the third (esize3) at byte #352+esize1+esize2, et cetera. Of course, in a .nii file, the value of vox_offset must be compatible with these extensions. If a malformed file indicates that an extended header data section would run past vox_offset, then that entire extended header section should be ignored. In a .hdr file, if an extended header data section would run past the end-of-file, that extended header data should also be ignored.

With the above scheme, a program can successively examine the esize and ecode values, and skip over each extended header section if the program doesnt know how to interpret the data within. Of course, any program can simply ignore all extended header sections simply by jumping straight to the image data using vox_offset.

The C reference implementation for NIfTI-1.1 I/O contains several functions to read and write header extensions. The nifti_tool command line program for manipulating NIfTI-1.1 files uses these functions to let you examine, remove, and/or add extension data to a file.

Created by Data Format Working Group
Last modified 2007-08-21 15:22

Powered by Plone

This site conforms to the following standards: