analyze

Read / write access to the basic Mayo Analyze format

The Analyze header format

This is a binary header format and inherits from WrapStruct

Apart from the attributes and methods of WrapStruct:

Class attributes are:

.default_x_flip

with methods:

.get/set_data_shape
.get/set_data_dtype
.get/set_zooms
.get/set_data_offset
.get_base_affine()
.get_best_affine()
.data_to_fileobj
.data_from_fileobj

and class methods:

.from_header(hdr)

More sophisticated headers can add more methods and attributes.

Notes

This - basic - analyze header cannot encode full affines (only diagonal affines), and cannot do integer scaling.

The inability to store affines means that we have to guess what orientation the image has. Most Analyze images are stored on disk in (fastest-changing to slowest-changing) R->L, P->A and I->S order. That is, the first voxel is the rightmost, most posterior and most inferior voxel location in the image, and the next voxel is one voxel towards the left of the image.

Most people refer to this disk storage format as ‘radiological’, on the basis that, if you load up the data as an array img_arr where the first axis is the fastest changing, then take a slice in the I->S axis - img_arr[:,:,10] - then the right part of the brain will be on the left of your displayed slice. Radiologists like looking at images where the left of the brain is on the right side of the image.

Conversely, if the image has the voxels stored with the left voxels first - L->R, P->A, I->S, then this would be ‘neurological’ format. Neurologists like looking at images where the left side of the brain is on the left of the image.

When we are guessing at an affine for Analyze, this translates to the problem of whether the affine should consider proceeding within the data down an X line as being from left to right, or right to left.

By default we assume that the image is stored in R->L format. We encode this choice in the default_x_flip flag that can be True or False. True means assume radiological.

If the image is 3D, and the X, Y and Z zooms are x, y, and z, then:

if default_x_flip is True::
    affine = np.diag((-x,y,z,1))
else:
    affine = np.diag((x,y,z,1))

In our implementation, there is no way of saving this assumed flip into the header. One way of doing this, that we have not used, is to allow negative zooms, in particular, negative X zooms. We did not do this because the image can be loaded with and without a default flip, so the saved zoom will not constrain the affine.

AnalyzeHeader([binaryblock, endianness, check])	Class for basic analyze header
AnalyzeImage(dataobj, affine[, header, ...])	Class for basic Analyze format image

AnalyzeHeader

class nibabel.analyze.AnalyzeHeader(binaryblock=None, endianness=None, check=True)

Bases: LabeledWrapStruct, SpatialHeader

Class for basic analyze header

Implements zoom-only setting of affine transform, and no image scaling

Initialize header from binary data block

Parameters:

binaryblock{None, string} optional

binary block to set into header. By default, None, in which case we insert the default empty header block

endianness{None, ‘<’,’>’, other endian code} string, optional

endianness of the binaryblock. If None, guess endianness from the data.

checkbool, optional

Whether to check content of header in initialization. Default is True.

Examples

>>> hdr1 = AnalyzeHeader() # an empty header
>>> hdr1.endianness == native_code
True
>>> hdr1.get_data_shape()
(0,)
>>> hdr1.set_data_shape((1,2,3)) # now with some content
>>> hdr1.get_data_shape()
(1, 2, 3)

We can set the binary block directly via this initialization. Here we get it from the header we have just made

>>> binblock2 = hdr1.binaryblock
>>> hdr2 = AnalyzeHeader(binblock2)
>>> hdr2.get_data_shape()
(1, 2, 3)

Empty headers are native endian by default

>>> hdr2.endianness == native_code
True

You can pass valid opposite endian headers with the endianness parameter. Even empty headers can have endianness

>>> hdr3 = AnalyzeHeader(endianness=swapped_code)
>>> hdr3.endianness == swapped_code
True

If you do not pass an endianness, and you pass some data, we will try to guess from the passed data.

>>> binblock3 = hdr3.binaryblock
>>> hdr4 = AnalyzeHeader(binblock3)
>>> hdr4.endianness == swapped_code
True

__init__(binaryblock=None, endianness=None, check=True)

Initialize header from binary data block

Parameters:

binaryblock{None, string} optional

binary block to set into header. By default, None, in which case we insert the default empty header block

endianness{None, ‘<’,’>’, other endian code} string, optional

endianness of the binaryblock. If None, guess endianness from the data.

checkbool, optional

Whether to check content of header in initialization. Default is True.

Examples

>>> hdr1 = AnalyzeHeader() # an empty header
>>> hdr1.endianness == native_code
True
>>> hdr1.get_data_shape()
(0,)
>>> hdr1.set_data_shape((1,2,3)) # now with some content
>>> hdr1.get_data_shape()
(1, 2, 3)

We can set the binary block directly via this initialization. Here we get it from the header we have just made

>>> binblock2 = hdr1.binaryblock
>>> hdr2 = AnalyzeHeader(binblock2)
>>> hdr2.get_data_shape()
(1, 2, 3)

Empty headers are native endian by default

>>> hdr2.endianness == native_code
True

You can pass valid opposite endian headers with the endianness parameter. Even empty headers can have endianness

>>> hdr3 = AnalyzeHeader(endianness=swapped_code)
>>> hdr3.endianness == swapped_code
True

If you do not pass an endianness, and you pass some data, we will try to guess from the passed data.

>>> binblock3 = hdr3.binaryblock
>>> hdr4 = AnalyzeHeader(binblock3)
>>> hdr4.endianness == swapped_code
True

as_analyze_map()

Return header as mapping for conversion to Analyze types

Collect data from custom header type to fill in fields for Analyze and derived header types (such as Nifti1 and Nifti2).

When Analyze types convert another header type to their own type, they call this this method to check if there are other Analyze / Nifti fields that the source header would like to set.

Returns:

analyze_mapmapping

Object that can be used as a mapping thus:

for key in analyze_map:
    value = analyze_map[key]

where key is the name of a field that can be set in an Analyze header type, such as Nifti1, and value is a value for the field. For example, analyze_map might be a something like dict(regular='y', slice_duration=0.3) where regular is a field present in both Analyze and Nifti1, and slice_duration is a field restricted to Nifti1 and Nifti2. If a particular Analyze header type does not recognize the field name, it will throw away the value without error. See Analyze.from_header().

Notes

You can also return a Nifti header with the relevant fields set.

Your header still needs methods get_data_dtype, get_data_shape and get_zooms, for the conversion, and these get called after using the analyze map, so the methods will override values set in the map.

data_from_fileobj(fileobj)

Read scaled data array from fileobj

Use this routine to get the scaled image data from an image file fileobj, given a header self. “Scaled” means, with any header scaling factors applied to the raw data in the file. Use raw_data_from_fileobj to get the raw data.

Parameters:

fileobjfile-like

Must be open, and implement read and seek methods

Returns:

arrndarray

scaled data array

Notes

We use the header to get any scale or intercept values to apply to the data. Raw Analyze files don’t have scale factors or intercepts, but this routine also works with formats based on Analyze, that do have scaling, such as SPM analyze formats and NIfTI.

data_to_fileobj(data, fileobj, rescale=True)

Write data to fileobj, maybe rescaling data, modifying self

In writing the data, we match the header to the written data, by setting the header scaling factors, iff rescale is True. Thus we modify self in the process of writing the data.

Parameters:

dataarray-like

data to write; should match header defined shape

fileobjfile-like object

Object with file interface, implementing write and seek

rescale{True, False}, optional

Whether to try and rescale data to match output dtype specified by header. If True and scaling needed and header cannot scale, then raise HeaderTypeError.

Examples

>>> from nibabel.analyze import AnalyzeHeader
>>> hdr = AnalyzeHeader()
>>> hdr.set_data_shape((1, 2, 3))
>>> hdr.set_data_dtype(np.float64)
>>> from io import BytesIO
>>> str_io = BytesIO()
>>> data = np.arange(6).reshape(1,2,3)
>>> hdr.data_to_fileobj(data, str_io)
>>> data.astype(np.float64).tobytes('F') == str_io.getvalue()
True

classmethod default_structarr(endianness=None)

Return header data for empty header with given endianness

default_x_flip = True

classmethod from_header(header=None, check=True)

Class method to create header from another header

Parameters:

headerHeader instance or mapping

a header of this class, or another class of header for conversion to this type

check{True, False}

whether to check header for integrity

Returns:

hdrheader instance

fresh header instance of our own class

get_base_affine()

Get affine from basic (shared) header fields

Note that we get the translations from the center of the image.

Examples

>>> hdr = AnalyzeHeader()
>>> hdr.set_data_shape((3, 5, 7))
>>> hdr.set_zooms((3, 2, 1))
>>> hdr.default_x_flip
True
>>> hdr.get_base_affine() # from center of image
array([[-3.,  0.,  0.,  3.],
       [ 0.,  2.,  0., -4.],
       [ 0.,  0.,  1., -3.],
       [ 0.,  0.,  0.,  1.]])

get_best_affine()

Get affine from basic (shared) header fields

Note that we get the translations from the center of the image.

Examples

>>> hdr = AnalyzeHeader()
>>> hdr.set_data_shape((3, 5, 7))
>>> hdr.set_zooms((3, 2, 1))
>>> hdr.default_x_flip
True
>>> hdr.get_base_affine() # from center of image
array([[-3.,  0.,  0.,  3.],
       [ 0.,  2.,  0., -4.],
       [ 0.,  0.,  1., -3.],
       [ 0.,  0.,  0.,  1.]])

get_data_dtype()

Get numpy dtype for data

For examples see set_data_dtype

get_data_offset()

Return offset into data file to read data

Examples

>>> hdr = AnalyzeHeader()
>>> hdr.get_data_offset()
0
>>> hdr['vox_offset'] = 12
>>> hdr.get_data_offset()
12

get_data_shape()

Get shape of data

Examples

>>> hdr = AnalyzeHeader()
>>> hdr.get_data_shape()
(0,)
>>> hdr.set_data_shape((1,2,3))
>>> hdr.get_data_shape()
(1, 2, 3)

Expanding number of dimensions gets default zooms

>>> hdr.get_zooms()
(1.0, 1.0, 1.0)

get_slope_inter()

Get scalefactor and intercept

These are not implemented for basic Analyze

get_zooms()

Get zooms from header

Returns:

ztuple

tuple of header zoom values

Examples

>>> hdr = AnalyzeHeader()
>>> hdr.get_zooms()
(1.0,)
>>> hdr.set_data_shape((1,2))
>>> hdr.get_zooms()
(1.0, 1.0)
>>> hdr.set_zooms((3, 4))
>>> hdr.get_zooms()
(3.0, 4.0)

classmethod guessed_endian(hdr)

Guess intended endianness from mapping-like hdr

Parameters:

hdrmapping-like

hdr for which to guess endianness

Returns:

endianness{‘<’, ‘>’}

Guessed endianness of header

Examples

Zeros header, no information, guess native

>>> hdr = AnalyzeHeader()
>>> hdr_data = np.zeros((), dtype=header_dtype)
>>> AnalyzeHeader.guessed_endian(hdr_data) == native_code
True

A valid native header is guessed native

>>> hdr_data = hdr.structarr.copy()
>>> AnalyzeHeader.guessed_endian(hdr_data) == native_code
True

And, when swapped, is guessed as swapped

>>> sw_hdr_data = hdr_data.byteswap(swapped_code)
>>> AnalyzeHeader.guessed_endian(sw_hdr_data) == swapped_code
True

The algorithm is as follows:

First, look at the first value in the dim field; this should be between 0 and 7. If it is between 1 and 7, then this must be a native endian header.

>>> hdr_data = np.zeros((), dtype=header_dtype) # blank binary data
>>> hdr_data['dim'][0] = 1
>>> AnalyzeHeader.guessed_endian(hdr_data) == native_code
True
>>> hdr_data['dim'][0] = 6
>>> AnalyzeHeader.guessed_endian(hdr_data) == native_code
True
>>> hdr_data['dim'][0] = -1
>>> AnalyzeHeader.guessed_endian(hdr_data) == swapped_code
True

If the first dim value is zeros, we need a tie breaker. In that case we check the sizeof_hdr field. This should be 348. If it looks like the byteswapped value of 348, assumed swapped. Otherwise assume native.

>>> hdr_data = np.zeros((), dtype=header_dtype) # blank binary data
>>> AnalyzeHeader.guessed_endian(hdr_data) == native_code
True
>>> hdr_data['sizeof_hdr'] = 1543569408
>>> AnalyzeHeader.guessed_endian(hdr_data) == swapped_code
True
>>> hdr_data['sizeof_hdr'] = -1
>>> AnalyzeHeader.guessed_endian(hdr_data) == native_code
True

This is overridden by the dim[0] value though:

>>> hdr_data['sizeof_hdr'] = 1543569408
>>> hdr_data['dim'][0] = 1
>>> AnalyzeHeader.guessed_endian(hdr_data) == native_code
True

has_data_intercept = False

has_data_slope = False

classmethod may_contain_header(binaryblock)

raw_data_from_fileobj(fileobj)

Read unscaled data array from fileobj

Parameters:

fileobjfile-like

Must be open, and implement read and seek methods

Returns:

arrndarray

unscaled data array

set_data_dtype(datatype)

Set numpy dtype for data from code or dtype or type

Examples

>>> hdr = AnalyzeHeader()
>>> hdr.set_data_dtype(np.uint8)
>>> hdr.get_data_dtype()
dtype('uint8')
>>> hdr.set_data_dtype(np.dtype(np.uint8))
>>> hdr.get_data_dtype()
dtype('uint8')
>>> hdr.set_data_dtype('implausible') 
Traceback (most recent call last):
   ...
HeaderDataError: data dtype "implausible" not recognized
>>> hdr.set_data_dtype('none') 
Traceback (most recent call last):
   ...
HeaderDataError: data dtype "none" known but not supported
>>> hdr.set_data_dtype(np.void) 
Traceback (most recent call last):
   ...
HeaderDataError: data dtype "<type 'numpy.void'>" known but not supported

set_data_offset(offset)

Set offset into data file to read data

set_data_shape(shape)

Set shape of data

If ndims == len(shape) then we set zooms for dimensions higher than ndims to 1.0

Parameters:

shapesequence

sequence of integers specifying data array shape

set_slope_inter(slope, inter=None)

Set slope and / or intercept into header

Set slope and intercept for image data, such that, if the image data is arr, then the scaled image data will be (arr * slope) + inter

In this case, for Analyze images, we can’t store the slope or the intercept, so this method only checks that slope is None or NaN or 1.0, and that inter is None or NaN or 0.

Parameters:

slopeNone or float

If float, value must be NaN or 1.0 or we raise a HeaderTypeError

interNone or float, optional

If float, value must be 0.0 or we raise a HeaderTypeError

set_zooms(zooms)

Set zooms into header fields

See docstring for get_zooms for examples

sizeof_hdr = 348

template_dtype = dtype([('sizeof_hdr', '<i4'), ('data_type', 'S10'), ('db_name', 'S18'), ('extents', '<i4'), ('session_error', '<i2'), ('regular', 'S1'), ('hkey_un0', 'S1'), ('dim', '<i2', (8,)), ('vox_units', 'S4'), ('cal_units', 'S8'), ('unused1', '<i2'), ('datatype', '<i2'), ('bitpix', '<i2'), ('dim_un0', '<i2'), ('pixdim', '<f4', (8,)), ('vox_offset', '<f4'), ('funused1', '<f4'), ('funused2', '<f4'), ('funused3', '<f4'), ('cal_max', '<f4'), ('cal_min', '<f4'), ('compressed', '<i4'), ('verified', '<i4'), ('glmax', '<i4'), ('glmin', '<i4'), ('descrip', 'S80'), ('aux_file', 'S24'), ('orient', 'S1'), ('originator', 'S10'), ('generated', 'S10'), ('scannum', 'S10'), ('patient_id', 'S10'), ('exp_date', 'S10'), ('exp_time', 'S10'), ('hist_un0', 'S3'), ('views', '<i4'), ('vols_added', '<i4'), ('start_field', '<i4'), ('field_skip', '<i4'), ('omax', '<i4'), ('omin', '<i4'), ('smax', '<i4'), ('smin', '<i4')])

AnalyzeImage

class nibabel.analyze.AnalyzeImage(dataobj, affine, header=None, extra=None, file_map=None, dtype=None)

Bases: SpatialImage

Class for basic Analyze format image

Initialize image

The image is a combination of (array-like, affine matrix, header), with optional metadata in extra, and filename / file-like objects contained in the file_map mapping.

Parameters:

dataobjobject

Object containing image data. It should be some object that returns an array from np.asanyarray. It should have a shape attribute or property

affineNone or (4,4) array-like

homogeneous affine giving relationship between voxel coordinates and world coordinates. Affine can also be None. In this case, obj.affine also returns None, and the affine as written to disk will depend on the file format.

headerNone or mapping or header instance, optional

metadata for this image format

extraNone or mapping, optional

metadata to associate with image that cannot be stored in the metadata of this image type

file_mapmapping, optional

mapping giving file information for this image format

__init__(dataobj, affine, header=None, extra=None, file_map=None, dtype=None)

Initialize image

The image is a combination of (array-like, affine matrix, header), with optional metadata in extra, and filename / file-like objects contained in the file_map mapping.

Parameters:

dataobjobject

Object containing image data. It should be some object that returns an array from np.asanyarray. It should have a shape attribute or property

affineNone or (4,4) array-like

homogeneous affine giving relationship between voxel coordinates and world coordinates. Affine can also be None. In this case, obj.affine also returns None, and the affine as written to disk will depend on the file format.

headerNone or mapping or header instance, optional

metadata for this image format

extraNone or mapping, optional

metadata to associate with image that cannot be stored in the metadata of this image type

file_mapmapping, optional

mapping giving file information for this image format

ImageArrayProxy

alias of ArrayProxy

files_types: tuple[tuple[str, str], ...] = (('image', '.img'), ('header', '.hdr'))

classmethod from_file_map(file_map, *, mmap=True, keep_file_open=None)

Class method to create image from mapping in file_map

Parameters:

file_mapdict

Mapping with (kay, value) pairs of (file_type, FileHolder instance giving file-likes for each file needed for this image type.

mmap{True, False, ‘c’, ‘r’}, optional, keyword only

mmap controls the use of numpy memory mapping for reading image array data. If False, do not try numpy memmap for data array. If one of {‘c’, ‘r’}, try numpy memmap with mode=mmap. A mmap value of True gives the same behavior as mmap='c'. If image data file cannot be memory-mapped, ignore mmap value and read array from file.

keep_file_open{ None, True, False }, optional, keyword only

keep_file_open controls whether a new file handle is created every time the image is accessed, or a single file handle is created and used for the lifetime of this ArrayProxy. If True, a single file handle is created and used. If False, a new file handle is created every time the image is accessed. If file_map refers to an open file handle, this setting has no effect. The default value (None) will result in the value of nibabel.arrayproxy.KEEP_FILE_OPEN_DEFAULT being used.

Returns:

imgAnalyzeImage instance

get_data_dtype()

header_class

alias of AnalyzeHeader

makeable: bool = True

rw: bool = True

set_data_dtype(dtype)

to_file_map(file_map=None, dtype=None)

Write image to file_map or contained self.file_map

Parameters:

file_mapNone or mapping, optional

files mapping. If None (default) use object’s file_map attribute instead

dtypedtype-like, optional

The on-disk data type to coerce the data array.

valid_exts: tuple[str, ...] = ('.img', '.hdr')