BasisSimulator.jl

GPU-portable polychromatic CT simulator in Julia. Runs on Metal, CUDA, ROCm, oneAPI, or CPU via AcceleratedKernels.jl. Models energy-integrating and photon-counting detectors, single- and dual-kVp acquisitions, and reconstructs with FBP (FDK), OS-PWLS iterative reconstruction, and material-basis VMI.

Install

using Pkg
Pkg.add(url="https://github.com/MolloiLab/BasisSimulator.jl")

For portable device selection, add GPUSelect plus your backend:

Pkg.add("GPUSelect")
Pkg.add("Metal")     # Apple Silicon
Pkg.add("CUDA")      # NVIDIA
Pkg.add("AMDGPU")    # AMD
Pkg.add("oneAPI")    # Intel

Quick example

import BasisSimulator as BS
import GPUSelect

AT = GPUSelect.Storage()  # MtlArray / CuArray / ROCArray / oneArray / Array
to_gpu(x) = AT(x)

phantom_cpu = BS.create_gammex_472(n_voxels=256)
phantom = BS.Phantom(to_gpu(phantom_cpu.mask),
                     phantom_cpu.materials,
                     phantom_cpu.voxel_size)

scanner = BS.Scanner(
    source_to_isocenter = 626.0,   # mm
    source_to_detector  = 1097.0,
    detector_rows       = 64,
    detector_cols       = 832,
    detector_row_size   = 0.625,
    detector_col_size   = 1.053,
)
protocol = BS.CTProtocol(kVp=120.0, mA=200.0, views=984)
sim_opts = BS.SimOptions(fidelity=:eict)
rec_opts = BS.ReconOptions(matrix_size=(512, 512, 64), fov_cm=35.0)

ws = BS.create_eict_workspace(scanner, protocol, sim_opts, rec_opts, phantom)
BS.simulate!(ws, phantom, protocol, sim_opts)

bhc = BS.calibrate_bhc_water(sim_opts, protocol; scanner=scanner, geom=ws.geom)
sino_bhc = to_gpu(BS.apply_bhc_water(ws.sinogram, bhc))
ws_fdk = BS.create_fdk_recon_workspace(sino_bhc, ws.geom, rec_opts.matrix_size)
hu = BS.to_hounsfield(
    Array(BS.reconstruct!(ws_fdk, sino_bhc, ws.geom));
    μ_water = bhc.μ_water_ref,
)

Forward projector: dd_fast is the fast default

SimOptions(; projector=:dd_fast) uses the fastest general-purpose projector and is the default. Its single-pass distance-driven kernels are anti-aliased and robust in severe beam-hardened regions while outperforming both the legacy :dd path and :siddon for full polychromatic and spectral simulations. :siddon remains available for comparison and compatibility, but its point-sampled rays can alias in those regions. If you reconstruct iteratively, pass the same projector to create_hir_recon_workspace(; projector=…) so the IR system matrix matches the data — FDK reconstruction is unaffected.

Documentation

Full API reference, getting-started guide, and eleven worked-example notebooks: https://molloilab.github.io/BasisSimulator.jl/. Docstrings are also available via ?Function in the Julia REPL.

License

MIT. Core ray tracing ported from TIGRE; calibration workflow follows CatSim/XCIST.