Change log and model updates
Segmentation models in MuscleMap are periodically retrained on expanded and more diverse datasets to optimise performance, robustness, and cross-site generalisability.
All notable changes to MuscleMap are listed below.
Source: GitHub Releases
v2.0 Latest
- Refactors MuscleMap to load model parameters and template NIfTI files directly from Zenodo instead of bundling them in the repository
- Added Zenodo download logic in
mm_util.pywith support for multiple Zenodo versions - Updated
mm_gui.pyandmm_segment.pyto use the new Zenodo-based model loading - Added
mm_qc_gui.pyas a new QC interface - Updated
setup.pyandrequirements.txtfor v2.0 dependencies - Updated
.gitignoreto exclude large model files - Updated
README.mdwith instructions for Zenodo usage and template file setup
v1.3
- Included auto chunking to estimate a safe chunk size from currently available CPU or GPU memory and keeps extra headroom to reduce OOM failures
- Included new model parameters
- Included extra spatial padding to preserve higher accuracy at muscle boundaries in cropped images
- Note: Model weights (
.pth) and configuration files (.json) formm_segmenthave been removed from the git history to reduce repository size. To use this release, download the model parameters from Zenodo and place them in the corresponding subfolder, e.g.scripts/models/wholebody/orscripts/models/abdomen/ - Note: Template images (
.nii.gz) formm_register_to_templatehave been removed from the git history to reduce repository size. To use this release, download the template images from Zenodo and place them in the corresponding subfolder, e.g.scripts/templates/abdomen/
v1.2
- Included 10 more muscles in wholebody model (left and right pectineus, obturator externus, obturator internus, piriformis, gemelli and quadratus femoris)
- Included more training data for wholebody model
v1.1
- Trained whole-body model on a more diverse dataset to improve generalizability
- Now using foreground cropping to improve inference efficiency
v1.0
- Added the whole-body segmentation model (v1.0)
- Improved handling of large images in
mm_segment - Added new options to
mm_segment - Cleaned and improved
mm_extract_metrics
v0.3 Pre-release
- Introduced version numbers for trained models
- Preparation for the v1.0 release
v0.2 Pre-release
- Added
mm_register_to_templatefor anatomical registration
v0.1 Pre-release
- Initial public pre-release including:
mm_segmentmm_extract_metricsmm_gui
v0.0 Pre-release
- Initial repository version with early segmentation tools