Toolbox concepts

Geometry

Most of the operations supported by the ASTRA Toolbox require you to specify the geometry of the (real or virtual) experimental setup you have used or want to simulate. This involves two parts: a volume geometry, and a projection geometry.

The volume geometry specifies where in 2D or 3D space the sample or phantom is located. It takes the form of a 2D rectangle or 3D box, and is usually centered around the origin.

The projection geometry specifies the location and trajectory of the ray source and detector. The available types of projection geometries are 2D parallel beam, 2D fan beam, 3D parallel beam and 3D cone beam.

n_rows = 256
n_cols = 256
volume_geometry = astra.create_vol_geom(n_rows, n_cols)

angles = np.linspace(0, np.pi, 180)
detector_pixel_count = 256
detector_spacing = 1.0
projection_geometry = astra.create_proj_geom('parallel', detector_spacing, detector_pixel_count, angles)

n_rows = 256;
n_cols = 256;
volume_geometry = astra_create_vol_geom(n_rows, n_cols);

angles = linspace(0, pi, 180);
detector_pixel_count = 256;
detector_spacing = 1.0;
projection_geometry = astra_create_proj_geom('parallel', detector_spacing, detector_pixel_count, angles);

Data

Data in ASTRA Toolbox is represented by objects separate from Python/MATLAB, and come in four varieties: 2D/3D volume data, and 2D/3D projection data. They can be manipulated with astra.data2d/astra_mex_data2d and astra.data3d/astra_mex_data3d commands. Each data object has a unique ID that can be used to refer to it in algorithms or data manipulation commands.

volume_data = np.random.rand(256, 256)
projection_data = np.random.rand(180, 256)

# Create ASTRA data object from NumPy arrays
volume_data_id = astra.data2d.create('-vol', volume_geometry, data)
projection_data_id = astra.data2d.create('-sino', volume_geometry, data)

# Delete ASTRA data objects using their unique IDs
astra.data2d.delete(volume_data_id)
astra.data2d.delete(projection_data_id)

volume_data = rand(256, 256);
projection_data = rand(180, 256);

% Create ASTRA data object from MATLAB arrays
volume_data_id = astra_mex_data2d('create', '-vol', volume_geometry, volume_data);
projection_data_id = astra_mex_data2d('create', '-sino', projection_geometry, projection_data);

% Delete ASTRA data objects using their unique IDs
astra_mex_data2d('delete', volume_data_id);
astra_mex_data2d('delete', projection_data_id);

Algorithms

To operate on data objects, the ASTRA Toolbox defines a number of algorithms. These include the basic forward and backward projection operations, and also various reconstruction algorithms.

The most flexible way to apply an algorithm to data is to create an algorithm configuration, initialize an algorithm object, and call run command. There result can then be retrieved from the corresponding ASTRA data object using get command.

config = astra.astra_dict('FP_CUDA')
config['ProjectionDataId'] = projection_data_id
config['VolumeDataId'] = volume_data_id

algorithm_id = astra.algorithm.create(config)

astra.algorithm.run(algorithm_id)

sinogram = astra.data2d.get(projection_data_id)

config = astra_struct('FP_CUDA');
config.ProjectionDataId = projection_data_id;
config.VolumeDataId = volume_data_id;

algorithm_id = astra_mex_algorithm('create', config);

astra_mex_algorithm('run', algorithm_id);

sinogram = astra_mex_data2d('get', projection_data_id);

Each ASTRA algorithm object has a unique ID, which can be used to clean it up after use, freeing the resources. Note that data objects are stored in RAM, but algorithm objects may additionally reserve GPU memory.

astra.algorithm.delete(algorithm_id)

astra_mex_algorithm('delete', algorithm_id);