A multi-dimensional spatial image data structure for scientific Python.

To facilitate:

• Multi-scale processing and analysis
• Registration
• Resampling
• Subregion parallel processing
• Coupling with meshes, point sets, and annotations

with scientific images, which are typically multi-dimensional with anisotropic sampling, this package provides a spatial-image data structure. In addition to an N-dimensional array of pixel values, spatial metadata defines the location of the pixel sampling grid in space time. We also label the array dimensions. This metadata is easily utilized and elegantly carried through image processing pipelines.

This package defines spatial image metadata, provides a function, is_spatial_image, to verify the expected behavior of a spatial image instance, and provides a reference function, to_spatial_image to convert an array-like, e.g. a NumPy ndarray or a Dask array, to a spatial image.

The spatial-image data structure is implemented with Xarray, a library for N-D labeled arrays and datasets in Python. The Xarray library is well-tested, relatively mature, and integrates well with scientific Python ecosystem tooling. The Xarray library leverages NumPy and pandas for labeled array indexing, integrates well with machine-learning libraries utilizing the scikit-learn interface, integrates with Dask for distributed computing, and zarr for serialization.

In essence, a spatial image is an xarray.DataArray with a defined set of dims labels, {'c', 'x', 'y', 'z', 't'}, constraints on the coords, to enforce uniform spacing in a given direction, and defined set of additional metadata attrs.

pip install spatial-image

A spatial image's xarray dims belong to the set: {'c', 'x', 'y', 'z', 't'}. These dimensions are:

c

Component / channel dimension.

x

First spatial dimension.

y

Second spatial dimension.

z

Third spatial dimension.

t

Time dimension.

Each dim has an axis with additional attributes to describe the dimension.

long_name

A descriptive name for the axis, e.g. anterior-posterior or x-axis. Defaults to the dim name.

units

Units for the axis, e.g. millimeters. Defaults to the empty string.

A spatial image's Xarray coords specify the spatial location of pixels in the image for the 'x', 'y', and 'z' data dimensions. For the 'c' and 't' data dimensions, component identities and timestamps can optionally be provided.

Spatial coordinates define the position in the coordinate reference frame of the image. In general, the image's coordinate reference frame may be different from the world coordinate reference frame.

Pixels are sampled on a uniform, possibly anisotropic, spatial grid. Spatial coordinates have a 64-bit float type. The difference between adjacent coordinates, i.e. the pixel spacing, for a dimension must be uniform. The first coordinate value defines the origin or offset of an image.

The component or channel dimension coordinates defaults to a sequence of integer identifiers but can be strings describing the channels, e.g. ['r', 'g', 'b'].

The time coordinates can have integer, float, or datetime64 type.

• Image-axis-aligned Cartesian coordinate reference frames enable Pythonic subscripting in processing pipelines on xarray.DataArray's. When indexing with slices, the same slices are applied to the multi-dimensional pixel array as the 1-D coordinate arrays, and the result is valid.

• Regular coordinate spacing enables processing optimizations, both algorithmically and computationally.

Contributions are welcome and appreciated.

To run the test suite:

git clone https://github.com/spatial-image/spatial-image
cd spatial-image
pip install -e ".[test]"
pytest