System requirements & FAQ¶
System requirements¶
| Requirement | |
|---|---|
| Python | 3.10 or newer |
| Operating system | Linux and macOS are supported and tested. On Windows, use WSL2 — MRtrix3 and some dependencies are awkward to install natively. |
| Structural network mapping | MRtrix3 on your PATH |
| Native-space masks | Register to MNI first — see the spatial normalization guide (uses ANTsPy, installed separately) |
| Disk space | Depends on the normative connectome you fetch (see below) |
Disk space for normative connectomes¶
| Connectome | Used by | Approx. size |
|---|---|---|
| GSP1000 | Functional network mapping | ~200 GB |
| HCP1065 | Structural network mapping | ~1.5 GB |
| dTOR985 full | Structural network mapping | ~11 GB |
| dTOR985 10% | Structural network mapping | ~1 GB |
| dTOR985 25% | Structural network mapping | ~3 GB |
Frequently asked questions¶
My lesion masks aren't in MNI space. What do I do?¶
Lacuna requires masks in MNI152NLin6Asym or MNI152NLin2009cAsym. If yours are
in native space, register them first — see the
spatial normalization guide. Lacuna
handles alignment between supported MNI spaces automatically, but it will not
guess the space of a mask that carries no spatial information.
Do I need to download a connectome to try Lacuna?¶
No. Run focal damage first — it uses the bundled atlases and needs no
download. Use lacuna tutorial my_dataset to generate a small synthetic dataset
to experiment with.
Does fetching a connectome require an account or API key?¶
It depends on the source:
- GSP1000 (functional) — requires a free Harvard Dataverse API key.
- HCP1065 (structural) — no key required.
- dTOR985 full (structural) — requires a Figshare API key.
- dTOR985 10% (structural) — no key required.
- dTOR985 25% (structural) — no key required.
Provide keys via the --api-key flag, an environment variable
(DATAVERSE_API_KEY / FIGSHARE_API_KEY), or ~/.config/lacuna/config.yaml.
See the fetch CLI reference for details.
Should I use FNM or AFNM?¶
Both compute the functional network of a lesion. Use the standard functional network mapping when you need a voxel-resolution map; use the accelerated version when you are processing many subjects and a parcel-resolution result is sufficient.
How do I process a whole cohort?¶
Point lacuna run at a BIDS dataset root rather than a single mask, or use
batch_process from the Python API. See the
getting started tutorial.
How do I cite Lacuna?¶
See CITATION.cff
in the repository.