tigerbx - SKILL.md Agent Skill

name: tigerbx description: "Use tigerbx for structural brain MRI (T1w NIfTI): brain extraction (BET), skull-stripping, tissue segmentation, ASEG, parcellation, cortical thickness, VBM, MNI registration, DTI/EPI distortion correction, hippocampus/amygdala embedding." argument-hint: "<bx|hlc|reg|gdm|nerve> <input.nii.gz> [output_dir]"

TigerBx skill

TigerBx is a Python package for deep-learning-based brain MRI analysis.

Models are downloaded automatically on first use and cached locally. No manual setup needed.

Environment check

Always verify the environment before using tigerbx. Run this first:

import importlib.metadata, packaging.version
try:
    v = importlib.metadata.version('tigerbx')
    assert packaging.version.Version(v) >= packaging.version.Version('0.2.1')
    print(f'tigerbx {v} ready')
except Exception:
    print('tigerbx >= 0.2.0 not found — install required')

If not installed or version is too old, install with:

# CPU
uv add "tigerbx[cpu] @ https://github.com/htylab/tigerbx/archive/release.zip"
# GPU (CUDA 12)
uv add "tigerbx[cu12] @ https://github.com/htylab/tigerbx/archive/release.zip"

Module dispatch

Task	Function	Detail
Brain extraction, skull-stripping	`tigerbx.run()`	bx.md
Brain mask, ASEG, deep GM, WMH, cortical thickness, CGW	`tigerbx.run()`	bx.md
Hierarchical parcellation (56 regions) or tissue probability maps	`tigerbx.hlc()`	hlc.md
Register T1 to MNI space, VBM, apply warp	`tigerbx.reg()` / `tigerbx.transform()`	reg.md
EPI / DTI geometric distortion correction	`tigerbx.gdm()`	gdm.md
Hippocampus/amygdala VAE embedding	`tigerbx.nerve()`	nerve.md
Quantitative metrics (Dice, HD95, PSNR, SSIM, …)	`tigerbx.eval()`	eval.md

Common conventions

input: single .nii/.nii.gz path, a directory path, a glob pattern, or a list of paths.
output: output directory string. When None, results are saved next to each input file.
GPU=True: enable GPU inference (requires onnxruntime-gpu).
Return value: dict of nibabel images (single input) or list of filename dicts (batch). Use .get_fdata() to access arrays.
QC score is always computed. A .log is auto-written if QC < 50.

Label definitions

See labels.md for full label tables: ASEG (43), DeepGM (12), HLC (56/171), SynthSeg.