TimelapsedHRpQCT v2

Quantifying remodelling activity from time-lapsed HR-pQCT images of the distal radius or tibia.

This repository is the v2 codebase. The original v1 repository is here: https://github.com/wallematthias/TimelapsedHRpQCTv1/tree/main

Changes from v1:

• Added functionality for multistack images
• Elastix Backend for registration
• More detailed remodelling outputs

Citation

If you use this tool in a publication, please cite:

• Walle M, Whittier DE, Schenk D, Atkins PR, Blauth M, Zysset P, Lippuner K, Muller R, Collins CJ. Precision of bone mechanoregulation assessment in humans using longitudinal high-resolution peripheral quantitative computed tomography in vivo. Bone. 2023;172:116780.

For related methodology, cite:

• Whittier DE, Walle M, Schenk D, Atkins PR, Collins CJ, Zysset P, Lippuner K, Muller R. A multi-stack registration technique to improve measurement accuracy and precision across longitudinal HR-pQCT scans. Bone. 2023;176:116893.
• Walle M, Duseja A, Whittier DE, Vilaca T, Paggiosi M, Eastell R, Muller R, Collins CJ. Bone remodeling and responsiveness to mechanical stimuli in individuals with type 1 diabetes mellitus. Journal of Bone and Mineral Research. 2024;39(2):85-94.
• Walle M, Gabel L, Whittier DE, Liphardt AM, Hulme PA, Heer M, Zwart SR, Smith SM, Sibonga JD, Boyd SK. Tracking of spaceflight-induced bone remodeling reveals a limited time frame for recovery of resorption sites in humans. Science Advances. 2024;10(51):eadq3632.

What The Pipeline Does

For each subject, the pipeline can:

1. Import raw AIM sessions into stack-level working artifacts.
2. Consume supplied masks/ROIs and segmentations.
3. Register each stack longitudinally across sessions.
4. When the selected profile enables multistack correction, estimate stack-to-stack correction transforms from per-stack superstacks.
5. Apply the canonical final transforms once to original grayscale, mask, and segmentation data.
6. Fill missing support regions in the fused transformed outputs.
7. Compute pairwise remodelling and trajectory metrics.

Profiles

Most routine workflow choices are made with --profile. The default --mode auto then follows the selected profile: single-stack profiles skip stack correction and filling, while multistack profiles run those stages.

Use --mode regular or --mode multistack only as an explicit override; profile selection is the preferred interface.

Install

Preferred installation:

pip install timelapsed-hrpqct

Python support: 3.11, 3.12, 3.13.

Minimal setup in a fresh conda environment:

conda create -n timelapsed-hrpqct python=3.13 -y
conda activate timelapsed-hrpqct
pip install timelapsed-hrpqct

Install into an existing environment:

pip install timelapsed-hrpqct

This package is pip-first and pulls runtime dependencies (including aimio-py and itk-elastix) automatically.

Development install:

pip install -e ".[test]"

Optional conda environment for local development:

conda env create -f environment.yml
conda activate timelapsed-hrpqct

The installable package name is timelapsed-hrpqct, and the import package is timelapsedhrpqct.

The CLI uses the bundled package default config (src/timelapsedhrpqct/configs/defaults.yml) automatically if you do not pass --config.

Slicer GUI / HR-pQCT Toolbox

The Slicer extension is now organized as an HR-pQCT toolbox with modules for longitudinal timelapsed analysis, motion scoring, Scanco AIM I/O, and contouring/segmentation helpers.

• Slicer extension repository: https://github.com/wallematthias/SlicerTimelapsedHRpQCT

When available through the Slicer Extensions Manager, install HR-pQCT Toolbox and restart Slicer. For developer mode:

1. Clone TimelapsedHRpQCTSlicer.
2. In Slicer: View -> Python Interactor.
3. Run:

   script = "<repo>/TimelapsedHRpQCTSlicer/scripts/link_local_toolbox_modules.py"
   exec(open(script).read(), {"__name__": "__main__", "SCRIPT_PATH": script})
   

4. Restart Slicer and open modules from the HR-pQCT category.
5. In Timelapsed HR-pQCT or Contours and Segmentation, click Install / Update timelapsed-hrpqct if the pipeline package is not installed in Slicer Python.

Manual developer-mode alternative: add all toolbox module paths in Edit -> Application Settings -> Modules:

• <repo>/TimelapsedHRpQCTSlicer/TimelapsedHRpQCT
• <repo>/TimelapsedHRpQCTSlicer/MotionScoreHRpQCT
• <repo>/TimelapsedHRpQCTSlicer/ScancoIO
• <repo>/TimelapsedHRpQCTSlicer/HRpQCTSegmentation

Quick Start

Preview discovery:

timelapse import /path/to/raw_data --dry-run

Import raw AIM files into a TimelapsedHRpQCT dataset without running the downstream stages:

timelapse import /path/to/raw_data --output-root /path/to/TimelapsedHRpQCT

By default raw files are kept in place (no sourcedata/hrpqct copy):

timelapse run /path/to/raw_data

Use a bundled profile for a study/workflow preset:

timelapse run /path/to/raw_data --profile eth-uofc

Combine a profile with a custom config when you want profile defaults plus local overrides:

timelapse run /path/to/raw_data --profile eth-uofc --config /path/to/study.yml

The configuration precedence is: bundled defaults, then --profile, then --config.

Enable copying raw files into sourcedata/hrpqct only when desired:

timelapse run /path/to/raw_data --copy-raw-inputs

Enable moving raw files into the normalized dataset layout only when desired:

timelapse run /path/to/raw_data --restructure-raw

Undo restructure moves (preview first):

timelapse undo-restructure /path/to/raw_data/imported_dataset --dry-run
timelapse undo-restructure /path/to/raw_data/imported_dataset

Run the default workflow (auto mode; follows the selected profile):

timelapse run /path/to/raw_data

Run with supplied masks/ROIs (the default):

timelapse run /path/to/raw_data

Use this when your input already includes valid masks/ROIs (for example FULL_MASK, REGMASK, TRAB_MASK, CORT_MASK, or ROI*) and any segmentation required by the selected remodelling method. Prepare missing masks with Bone Contouring before running Timelapsed.

Input discovery is recursive, so your source folder can be either flat/unstructured or organized in a BIDS/MIDS-style nested layout. When filename parsing is ambiguous, discovery can fall back to AIM header metadata (Index Patient, Index Measurement, Site). Left/right site aliases are supported (RL/RR/TL/TR/KL/KR) while generic radius/tibia/knee remains fully supported.

Run a standard multistack workflow:

timelapse run /path/to/raw_data --profile multistack

Run a pediatric fracture multistack workflow:

timelapse run /path/to/raw_data --profile ped-fx

Run stages manually:

timelapse import /path/to/raw_data
timelapse register /path/to/raw_data/imported_dataset
timelapse stackcorrect /path/to/raw_data/imported_dataset
timelapse transform /path/to/raw_data/imported_dataset
timelapse fill /path/to/raw_data/imported_dataset
timelapse analyse /path/to/raw_data/imported_dataset

Pass --config /path/to/other.yml only when you want to override the built-in default.

The default analysis space is baseline_common, which is also the fastest option. pairwise_fixed_t0 is available for single-stack datasets, but it is slower because each timepoint pair is resampled during analysis.

Incremental Reruns

The run command is incremental:

• already imported sessions are skipped
• existing supplied masks/ROIs and segmentations are reused
• existing baseline transforms are reused
• existing final transforms are reused
• existing fused transformed sessions are reused
• existing filled sessions are reused
• existing analysis is reused unless you pass analysis overrides like --thr, --clusters, or --visualize

This makes it practical to rerun the pipeline after fixing one stage or adding new sessions without recomputing everything else.

Mask Roles And Naming

Discovery now supports both canonical and generic mask roles from filenames.

Examples:

# Distal radius (DR), standard trab/cort masks across sessions
SUBJ001_DR_T1.AIM
SUBJ001_DR_T1_TRAB_MASK.AIM
SUBJ001_DR_T1_CORT_MASK.AIM
SUBJ001_DR_T2.AIM
SUBJ001_DR_T2_TRAB_MASK.AIM
SUBJ001_DR_T2_CORT_MASK.AIM
SUBJ001_DR_T3.AIM
SUBJ001_DR_T3_TRAB_MASK.AIM
SUBJ001_DR_T3_CORT_MASK.AIM

# Distal tibia (DT)
SUBJ002_DT_T1.AIM
SUBJ002_DT_T1_TRAB_MASK.AIM
SUBJ002_DT_T1_CORT_MASK.AIM

# Knee (KN)
SUBJ003_KN_T1.AIM
SUBJ003_KN_T1_TRAB_MASK.AIM
SUBJ003_KN_T1_CORT_MASK.AIM

# Optional generic masks
SUBJ001_DR_T1_REGMASK.AIM
SUBJ001_DR_T1_ROI1.AIM
SUBJ001_DR_T1_ROI2.AIM
SUBJ001_DR_T1_MASK1.AIM

Behavior:

• REGMASK is preferred for registration when present.
• If no REGMASK exists, registration falls back to trab+cort union, then full, then generic MASK* unions.
• For analysis compartments, ROI* masks are preferred when present across sessions.
• If no ROI* masks are present, regmask is used as analysis ROI.
• Otherwise analysis uses configured compartments filtered by masks available in the current derivative layout.

Multistack Filename Parsing Notes

If your raw files are already split into physical stacks, include a stack token in the filename:

SUBJ001_DT_STACK01_T1.AIM
SUBJ001_DT_STACK01_T1_TRAB_MASK.AIM
SUBJ001_DT_STACK01_T1_CORT_MASK.AIM
SUBJ001_DT_STACK02_T1.AIM
SUBJ001_DT_STACK02_T1_TRAB_MASK.AIM
SUBJ001_DT_STACK02_T1_CORT_MASK.AIM

Accepted stack token styles include STACK01, STACK_01, and STACK-01.

Notes:

• If STACK... is present, files are grouped by that stack index during discovery.
• If STACK... is missing, the image is treated as a single acquisition and import splits by import.stack_depth (default 168).
• If site token is missing, discovery uses discovery.default_site (default tibia).
• REGMASK is optional and overrides registration mask selection when present.
• ROI* masks are optional and override analysis compartments when consistently present across sessions.

Repository Layout

• src/timelapsedhrpqct/workflows/: orchestration for each pipeline stage
• src/timelapsedhrpqct/processing/: reusable algorithmic and I/O helpers
• src/timelapsedhrpqct/dataset/: discovery, layout, artifact records, derivative paths
• src/timelapsedhrpqct/analysis/: remodelling analysis logic
• src/timelapsedhrpqct/configs/: bundled default YAML configuration
• tests/: unit, characterization, and end-to-end workflow tests

Documentation

Detailed documentation lives in docs/:

• Documentation Index
• Installation
• Usage
• Usage Examples
• Annotated Defaults
• Multistack Algorithm
• Timelapsed Analysis
• Settings Reference
• API Reference

Testing

Run the full test suite:

pytest -q

License

This repository is licensed under the MIT License. See LICENSE.

Packaging

The repository includes:

• environment.yml for local conda environments
• .github/workflows/ci.yml for tests and pip install smoke checks
• .github/workflows/publish-pypi.yml for trusted-publisher PyPI releases
• conda-recipe/ for conda packaging