niftynet.io.misc_io module

infer_ndims_from_file(file_path, loader=None)

dtype_casting(original_dtype, interp_order, as_tf=False)

Making image dtype based on user specified interp order and best compatibility with Tensorflow.

(if interp_order > 1, all values are promoted to float32,

this avoids errors when the input images have different dtypes)

The image preprocessing steps such as normalising intensities to [-1, 1] will cast input into floats. We therefore cast almost everything to float32 in the reader. Potentially more complex casting rules are needed here.

Parameters:

• original_dtype – an input datatype
• interp_order – an integer of interpolation order
• as_tf – boolean

Returns:

normalised numpy dtype if not as_tf else tensorflow dtypes

create_affine_pixdim(affine, pixdim)

Given an existing affine transformation and the pixel dimension to apply, create a new affine matrix that satisfies the new pixel dimension.

Parameters:

• affine – original affine matrix
• pixdim – pixel dimensions to apply

Returns:

correct_image_if_necessary(img)

rectify_header_sform_qform(img_nii)

Look at the sform and qform of the nifti object and correct it if any incompatibilities with pixel dimensions

Parameters:img_nii – Returns:

do_reorientation(data_array, init_axcodes, final_axcodes)

Performs the reorientation (changing order of axes)

Parameters:

• data_array – 5D Array to reorient
• init_axcodes – Initial orientation
• final_axcodes – Target orientation

Return data_reoriented:  

New data array in its reoriented form

do_resampling(data_array, pixdim_init, pixdim_fin, interp_order)

Performs the resampling

Parameters:

• data_array – 5D Data array to resample
• pixdim_init – Initial pixel dimension
• pixdim_fin – Targeted pixel dimension
• interp_order – Interpolation order applied

Return data_resampled:  

Array containing the resampled data

save_data_array(filefolder, filename, array_to_save, image_object=None, interp_order=3, reshape=True)

write image data array to hard drive using image_object properties such as affine, pixdim and axcodes.

expand_to_5d(img_data)

Expands an array up to 5d if it is not the case yet; The first three spatial dims are rearranged so that 1-d is always [X, 1, 1] 2-d is always [X, y, 1] :param img_data: :return:

save_volume_5d(img_data, filename, save_path, affine=array([[1., 0., 0., 0.], [0., 1., 0., 0.], [0., 0., 1., 0.], [0., 0., 0., 1.]]))

Save the img_data to nifti image

Parameters:

• img_data – 5d img to save
• filename – filename under which to save the img_data
• save_path –
• affine – an affine matrix.

Returns:

split_filename(file_name)

squeeze_spatial_temporal_dim(tf_tensor)

Given a tensorflow tensor, ndims==6 means:

[batch, x, y, z, time, modality]

this function removes x, y, z, and time dims if the length along the dims is one.

Returns:squeezed tensor

touch_folder(model_dir)

This function returns the absolute path of model_dir if exists otherwise try to create the folder and returns the absolute path.

resolve_module_dir(module_dir_str, create_new=False)

to_absolute_path(input_path, model_root)

resolve_file_name(file_name, paths)

check if file_name exists, if not, go though the list of [path + file_name for path in paths]. raises IOError if all options don’t exist

Parameters:

• file_name –
• paths –

Returns:

resolve_checkpoint(checkpoint_name)

get_latest_subfolder(parent_folder, create_new=False)

image3(name, tensor, max_outputs=3, collections=('summaries', ), animation_axes=(1, ), image_axes=(2, 3), other_indices=None)

Summary for higher dimensional images

Parameters:

• name – string name for the summary
• tensor – tensor to summarize. Should be in the range 0..255. By default, assumes tensor is NDHWC, and animates (through D) HxW slices of the 1st channel.
• collections – list of strings collections to add the summary to
• animation_axes=[1],image_axes=[2,3] –

image3_sagittal(name, tensor, max_outputs=3, collections=('summaries', ))

image3_coronal(name, tensor, max_outputs=3, collections=('summaries', ))

image3_axial(name, tensor, max_outputs=3, collections=('summaries', ))

set_logger(file_name=None)