Pet Skill

PET Skill (Modality Layer)

Overview

pet-skill is the NeuroClaw modality-layer interface skill responsible for all PET neuroimaging data processing tasks.

It strictly follows the NeuroClaw hierarchical design principles:

• This skill only describes WHAT needs to be done and which tool skill to delegate to.
• It contains no implementation code or concrete commands.
• All concrete execution is delegated to existing base/tool skills: fsl-tool, freesurfer-tool, nibabel-skill, and claw-shell.
• Companion scripts in scripts/ provide reference implementations for SUVR computation and reference region extraction.

Core workflow (never bypassed):

1. Identify input PET data and tracer type (PiB, FDG, tau, or other).
2. Ensure T1w structural data is available (via smri-skill if not yet processed).
3. Generate a numbered execution plan clearly stating WHAT needs to be done and which tool skill will handle each step.
4. Present the full plan, estimated runtime, resource requirements, and risks to the user and wait for explicit confirmation ("YES" / "execute" / "proceed").
5. On confirmation, delegate every step to the appropriate skill via claw-shell.
6. After execution, save all outputs in a clean directory structure (pet_output/).

Research use only.

Quick Reference (Common PET Tasks)

Tracer-Specific Reference Regions

Core Processing Pipeline

Stage 1: T1w Preprocessing (via smri-skill)

• Brain extraction, tissue segmentation, cortical parcellation (FreeSurfer)
• Required for reference region definition and PVC

Stage 2: PET-to-T1w Coregistration (via fsl-tool)

• Rigid-body registration of mean PET frame to T1w using FLIRT
• Apply transformation to full dynamic or static PET series

Stage 3: Reference Region Definition

• Use FreeSurfer parcellation to extract reference region mask in T1w space
• Common references: cerebellar cortex (Cerebellum_Cortex in Desikan-Killiany), pons
• Project mask to PET space or keep in T1w space with partial volume correction

Stage 4: SUVR Computation (via scripts/compute_suvr.py)

• Extract mean signal from target ROI and reference region
• SUVR = mean(target) / mean(reference)
• Output per-region SUVR values as CSV

Stage 5 (Optional): Spatial Normalization to MNI

• Warp PET (in T1w space) to MNI152 using T1w-to-MNI warp
• Enable group-level voxelwise analysis

Scripts

scripts/compute_suvr.py

Computes SUVR from a PET image and ROI/reference masks.

python skills/pet-skill/scripts/compute_suvr.py \
  --pet /path/to/pet_in_t1w_space.nii.gz \
  --target-mask /path/to/target_roi_mask.nii.gz \
  --ref-mask /path/to/reference_region_mask.nii.gz \
  --output /path/to/pet_output/suvr_values.csv

Standard Output Layout

pet_output/
├── coregistration/          # PET-to-T1w registration matrices and resampled PET
├── suvr/                    # SUVR maps and per-region CSV values
│   ├── suvr_values.csv
│   └── suvr_map.nii.gz
├── pvc/                     # Partial volume corrected values (if requested)
├── mni/                     # PET in MNI152 space (if normalization requested)
├── qc/                      # Coregistration quality, reference region coverage
└── logs/

Installation (Handled by dependency-planner)

No manual installation required at this layer. When first used, pet-skill automatically calls dependency-planner to ensure fsl-tool, freesurfer-tool, nibabel-skill, and claw-shell are ready.

Important Notes & Limitations

• PET images are typically low-resolution (~2–4 mm); coregistration to high-resolution T1w is essential.
• Reference region selection is tracer-dependent; using the wrong reference region invalidates SUVR.
• Partial volume correction is recommended for atrophy-prone populations (e.g., Alzheimer's disease).
• Dynamic PET requires frame timing information from DICOM headers or sidecar JSON.
• Static PET (single late frame) is sufficient for most clinical SUVR analyses.
• This skill is for research workflows; not for clinical decision-making.

When to Call This Skill

• After smri-skill when T1w structural preprocessing is complete and PET data needs processing.
• When the user needs SUVR computation from amyloid (PiB), metabolism (FDG), or tau PET data.
• When PET-to-T1w coregistration or normalization to MNI space is required.
• When partial volume correction is requested for ROI-based PET quantification.
• When dataset skills (e.g., aibl-skill, adni-skill) delegate PET processing.

Complementary / Related Skills

• smri-skill → T1w structural preprocessing (brain extraction, parcellation)
• fmri-skill → if PET is used alongside fMRI for multimodal analysis
• fsl-tool → FLIRT (coregistration), FNIRT (normalization), PETPVC (partial volume correction)
• freesurfer-tool → cortical/subcortical parcellation for ROI definition
• nibabel-skill → NIfTI I/O for mask manipulation
• brain-visualization → PET overlay visualization
• aibl-skill → AIBL dataset (PiB, FDG, tau PET)
• adni-skill → ADNI dataset (PET data available)

Reference

• Klunk et al. (2004): PiB amyloid imaging
• Landau et al. (2012): Amyloid imaging with PiB and florbetapir
• Baker et al. (2017): AV-1451 tau PET imaging
• BIDS PET extension: https://bids-specification.readthedocs.io/en/stable/04-modality-specific-files/09-positron-emission-tomography.html
• FSL: https://fsl.fmrib.ox.ac.uk/fsl/
• FreeSurfer: https://surfer.nmr.mgh.harvard.edu/

Created At: 2026-05-06 12:19 HKT Last Updated At: 2026-05-06 12:19 HKT Author: chengwang96