Repository

Troubleshooting

The CLI Rejects My Argument

Arguments must use --name=value. Space-separated forms such as --name value are not supported.

For PAM, see PAM CLI Parameters for supported options. Unknown options are accepted by the low-level parser, but they only matter if the runner reads them.

CT Data Is Missing

Skull-backed runs need private DICOM data. Use homogeneous water controls while testing setup:

julia --project=. scripts/run_pam.jl \
  --source-model=point \
  --sources-mm=30:0 \
  --aberrator=none

For skull runs, pass the data location explicitly:

--ct-path=/path/to/dicom-folder

CUDA Is Not Available

If CUDA.jl cannot see a GPU, set --recon-use-gpu=false for supported 2D runs. 3D reconstruction currently requires the CUDA path.

k-Wave simulation GPU usage is controlled separately with --kwave-use-gpu.

k-Wave Or Python Setup Fails

Python dependencies are managed through CondaPkg.toml. The first k-Wave run may take longer because Python packages and k-Wave resources are resolved.

If setup fails, try instantiating the Julia environment first:

julia --project=. -e 'using Pkg; Pkg.instantiate()'

Then run a homogeneous point-source example before attempting CT-backed skull simulations.

A Run Takes Too Long

The largest runtime controls are grid size, dimensionality, reconstruction bandwidth, window count, and GPU usage.

For faster iteration:

  • Start with --aberrator=none.
  • Use --source-model=point.
  • Reduce --axial-mm and --transverse-mm.
  • Use tighter --recon-bandwidth-khz.
  • Use --recon-progress=true to confirm reconstruction is moving.

Reconstruction-Only Reruns Fail

--from-run-dir reuses the previous simulation data. Remove options that would change the simulation, source geometry, medium, or grid. Keep only reconstruction and analysis options such as --recon-bandwidth-khz, --recon-step-um, --recon-use-gpu, and threshold settings.