name	tigerbx
description	Use TigerBx 0.2.3+ for research-use neuroimaging workflows: T1w brain extraction and segmentation, HLC parcellation, registration, VBM, EPI/DTI distortion correction, NERVE embeddings, and image metrics. Trigger when choosing or calling tigerbx Python APIs/CLI, interpreting outputs or labels, or building TigerBx batch pipelines.

TigerBx

Use this skill to select the smallest correct TigerBx API and avoid mixing pipeline contracts.

Route the task

Need	API	Read
BET, mask, ASEG, DeepGM, WMH, SynthSeg, CT, tissue PVE	`tigerbx.run()`	bx.md
HLC parcellation or HLC tissue outputs	`tigerbx.hlc()`	hlc.md
Rigid/affine/nonlinear registration; apply saved warp	`tigerbx.reg()`, `tigerbx.transform()`	reg.md
VBM	`tigerbx.vbm()` or `tigerbx.pipeline('vbm', ...)`	reg.md
EPI/DTI distortion correction	`tigerbx.gdm()`	gdm.md
Hippocampus/amygdala VAE latent/reconstruction	`tigerbx.nerve()`	nerve.md
Dice, IoU, distances, IQA, MI, classification metrics	`tigerbx.eval()`	eval.md
Numeric segmentation labels	—	labels.md

Read only the routed file. Read labels.md only when label meaning is needed.

Non-negotiable contracts

Treat all outputs as research-use only; do not make clinical or diagnostic claims.
reg plans are ordered uppercase steps: R/A/V/N/C/F. VBM is a separate API, never a reg step.
Single NIfTI input normally returns a dict of nibabel objects; batch input returns a list of dicts containing saved paths. NERVE returns named path dicts because its primary artifact is NPZ.
Access a nibabel image array with .get_fdata().
Models download automatically on first execution. Do not install packages or download models unless the user asks to run the workflow and the environment lacks them.
For GPU: use g inside tigerbx.run() flags; use the documented GPU=True or gpu=True spelling for other APIs.

Execution check

Only when executing code, verify once:

import tigerbx
print(tigerbx.__version__)  # expected >= 0.2.3

Prefer the Python API for composed workflows and the tiger CLI for one-off shell runs.

TigerBx

Use this skill to select the smallest correct TigerBx API and avoid mixing pipeline contracts.

Route the task

Need

API

Read

BET, mask, ASEG, DeepGM, WMH, SynthSeg, CT, tissue PVE

tigerbx.run()

bx.md

HLC parcellation or HLC tissue outputs

tigerbx.hlc()

hlc.md

Rigid/affine/nonlinear registration; apply saved warp

tigerbx.reg(), tigerbx.transform()

reg.md

VBM

tigerbx.vbm() or tigerbx.pipeline('vbm', ...)

reg.md

EPI/DTI distortion correction

tigerbx.gdm()

gdm.md

Hippocampus/amygdala VAE latent/reconstruction

tigerbx.nerve()

nerve.md

Dice, IoU, distances, IQA, MI, classification metrics

tigerbx.eval()

eval.md

Numeric segmentation labels

—

labels.md

Read only the routed file. Read labels.md only when label meaning is needed.

Non-negotiable contracts

Treat all outputs as research-use only; do not make clinical or diagnostic claims.

reg plans are ordered uppercase steps: R/A/V/N/C/F. VBM is a separate API, never a reg step.

Single NIfTI input normally returns a dict of nibabel objects; batch input returns a list of dicts containing saved paths. NERVE returns named path dicts because its primary artifact is NPZ.

Access a nibabel image array with .get_fdata().

Models download automatically on first execution. Do not install packages or download models unless the user asks to run the workflow and the environment lacks them.

For GPU: use g inside tigerbx.run() flags; use the documented GPU=True or gpu=True spelling for other APIs.

Execution check

Only when executing code, verify once:

import tigerbx print(tigerbx.__version__) # expected >= 0.2.3

Prefer the Python API for composed workflows and the tiger CLI for one-off shell runs.