BACKEND.md

LEAP: Differentiable forward and backward projectors for AI/ML-driven computed tomography applications

Python API

Projector class is provided to enable PyTorch integeration. It is a wrapper class that uses Python LEAP API. This section describes the Python API functions.

project_cpu

forward project on CPU

backproject_cpu

back project on CPU

project_gpu

forward project on GPU

backproject_gpu

back project on GPU

print_param

print all current projection and image parameters

save_param

save all current parameters to a file

set_gpu

set GPU device id for forward and back projection operations. begin with 0.

set_projector

set_symmetry_axis

set_cone_beam

set up cone beam projection parameters

set_parallel_beam

set up parallel beam projection parameters

set_modular_beam

set up modular beam projection parameters

set_volume

set up image volume dimension to be reconstructed

get_volume_dim

get image volume dimension

get_projection_dim

get projection data (sinogram) dimension

create_param

create a new parameter set, in addition to the default parameter set

clear_param_all

remove all parameter sets in memory, the main default parameter set will not be removed.

project_cone_cpu

forward project on CPU by cone beam geometry parameters

project_parallel_cpu

forward project on CPU by parallel beam geometry parameters

backproject_cone_cpu

back project on CPU by cone beam geometry parameters

backproject_parallel_cpu

back project on CPU by parallel beam geometry parameters

project_cone_gpu

forward project on GPU by cone beam geometry parameters

project_parallel_gpu

forward project on GPU by parallel beam geometry parameters

backproject_cone_gpu

back project on GPU by cone beam geometry parameters

backproject_parallel_gpu

back project on GPU by parallel beam geometry parameters

Backend C++ API

The core backend of the LEAP library is written in C++ and CUDA. It is dynamically linked in the Python environment and used with the Python binding mechanism.