MyoPath is a deep learning pipeline for objective morphometric assessment of skeletal muscle biopsies from routine H&E-stained whole slide images (WSI).
It implements a four-tissue segmentation pipeline — Cellpose-SAM for myofiber instance segmentation, a pixel classifier for fat infiltration, watershed detection for nuclei, and Boolean subtraction for connective tissue — to extract 37 morphometric features per sample. From these, seven clinically interpretable pathology indicators are derived, with Nuclear Centralization Index (NCI) and Fiber Size Variability Coefficient (Fiber CV) serving as primary biomarkers.
The pipeline has been validated on 478 H&E whole-slide images from two independent cohorts (HuashanMuscle, n = 79; GTEx, n = 399).
Data Download and Full documentation: https://myotoolkit.huashanmuscle.com/myopath/introduction.html
| Dataset | Content | Download |
|---|---|---|
| HuashanMuscle H&E WSI | 78 de-identified whole slide images (SVS, 72.1 GB); per-slide metadata and checksums in release huashanmuscle-wsi-v1.0 | Baidu Pan / OneDrive |
| HuashanMuscle ROI images | 79 exported ROI (TIFF/JPEG + GeoJSON) | Baidu Pan / OneDrive |
| GTEx H&E WSI | 399 whole slide images | GTEx Portal |
| GTEx ROI images | 399 exported ROI (TIFF/JPEG + GeoJSON) | Baidu Pan / OneDrive |
The HuashanMuscle whole slide images are released under CC BY-NC 4.0. Their full-resolution pixel grid matches the ROI GeoJSON annotations; format, de-identification and coordinate details are in the dataset README attached to the release.
- Four-tissue segmentation: Myofiber, fat, nucleus, and connective tissue from a single H&E section
- 37 morphometric features across five biological categories
- Seven pathology indicators covering nuclear positioning, fiber size, shape, tissue composition, and cellular reaction
- MyoPath Score: Composite severity measure (AUC = 0.873 on external validation)
- Primary biomarkers: NCI (p = 1.3 × 10⁻⁵) and Fiber CV (p = 2.9 × 10⁻⁴)
| Layer | Tissue | Method | Dice |
|---|---|---|---|
| 1 | Myofiber instances | Cellpose-SAM | 0.92 |
| 2 | Fat infiltration | Pixel classifier | 0.95 |
| 3 | Nuclei | Watershed detection | 0.87 |
| 4 | Connective tissue | Boolean subtraction | 0.88 |
Click the image above to watch the full QuPath segmentation walkthrough.
| Step | Script | Platform | What it does |
|---|---|---|---|
| 1 | MyoPath1_roi.groovy |
QuPath | Create a standardized 1500 µm × 1500 µm ROI |
| 2 | MyoPath2_segment.groovy |
QuPath | Segment muscle fibers (Cellpose-SAM), fat (pixel classifier), nuclei (watershed) |
| 3 | MyoPath3_export.groovy |
QuPath | Export ROI image (TIFF/JPEG) + annotations (GeoJSON) |
| 4 | MyoPath4_analysis.py |
Python | Batch tissue analysis, visualization, and dystrophy report |
- QuPath 0.6.0+
- Cellpose extension — configure the Cellpose and Cellpose SAM Python paths in
Edit → Preferences → Cellpose/Omnipose - GPU recommended for Cellpose segmentation
- Python 3.9+
- Conda environment recommended
# Clone the repository
git clone https://github.com/Hirriririir/MyoPath.git
cd MyoPath
# Set up Python environment
conda create -n myopath python=3.10 -y
conda activate myopath
pip install -r requirements.txt
# Verify installation
python -c "import tiatoolbox, numpy, scipy, matplotlib, shapely, pandas; print('All dependencies OK')"MyoPath/
├── MyoPath1_roi.groovy # Step 1: Create standardized ROI
├── MyoPath2_segment.groovy # Step 2: Three-layer tissue segmentation
├── MyoPath3_export.groovy # Step 3: Export ROI image + annotations
├── MyoPath4_analysis.py # Step 4: Batch analysis orchestrator
├── classifiers/
│ ├── classes.json # QuPath path class definitions
│ └── pixel_classifiers/
│ ├── Fat in muscle.json # Required pixel classifier
│ └── ... # Other optional classifiers
├── src/
│ ├── __init__.py
│ ├── data_loader.py # WSI and GeoJSON data loading
│ ├── coordinate_processor.py # Coordinate transformation
│ ├── tissue_analyzer.py # Multi-tissue area analysis
│ ├── visualizer.py # Segmentation visualization
│ └── dystrophy_analyzer.py # Pathology indicators & scoring
├── exports/ # Step 3 output (per-sample)
├── results/ # Step 4 aggregated output
├── requirements.txt
├── LICENSE
└── README.md
Open your slide in QuPath and run MyoPath1_roi.groovy to create a standardized 1500 µm × 1500 µm ROI.
Select the ROI and run MyoPath2_segment.groovy. Three-layer segmentation completes automatically.
Select the ROI and run MyoPath3_export.groovy to export ROI image and GeoJSON annotations.
# Test with a single sample
python MyoPath4_analysis.py --test HE_M3669
# Process all samples
python MyoPath4_analysis.py --all --cores 16
# Process specific samples
python MyoPath4_analysis.py --samples HE_M1600,HE_M2806
# Custom directories
python MyoPath4_analysis.py --all --input /data/exports --output /data/results
# Debug mode
python MyoPath4_analysis.py --test HE_M3669 --sequential --verbose| Flag | Default | Description |
|---|---|---|
--test NAME |
— | Test a single sample |
--samples A,B |
— | Process specific samples (comma-separated) |
--all |
— | Process all samples in exports/ |
--input DIR |
../exports |
Exports directory path |
--output DIR |
../results |
Results directory path |
--cores N |
10 | CPU cores for parallel processing |
--sequential |
off | Single-process mode (for debugging) |
--verbose |
off | Detailed logging |
exports/{sample_name}/
├── {sample}_statistics.json # Tissue area statistics
├── {sample}_dystrophy_report.json # Full pathology report
├── original_image.png # H&E image
├── standard_segmentation.png # 4-tissue overlay
├── muscle_fiber_colormap.png # Individual fiber colors
└── three_view_comparison.png # Side-by-side comparison
results/
└── batch_analysis_summary.json # Batch processing summary
The JSON report includes 7 pathology indicators and an overall MyoPath Score (0–100):
| Indicator | Normal Range | What it measures |
|---|---|---|
| Nuclear Centralization Index (NCI) | < 0.03 | Nuclei migrating to fiber center |
| Fiber Size Variability (CV) | < 0.25 | Size heterogeneity |
| Fiber Shape Regularity | > 0.7 | Shape abnormality |
| Fat Infiltration | < 5% | Fat tissue percentage |
| Fibrosis | < 10% | Connective tissue percentage |
| Nuclear/Muscle Ratio | 1–3 | Nuclei inside muscle fibers |
| Inflammatory Infiltration | < 2,000/mm² | Nuclear density in connective tissue |
| Score Range | Severity |
|---|---|
| 0–20 | Normal / Minimal |
| 20–40 | Mild |
| 40–60 | Moderate |
| 60–100 | Severe |
| Parameter | Default | Description |
|---|---|---|
expectedDiameterMicrons |
80 | Expected myofiber diameter (µm) |
downsample |
10.0 | Segmentation resolution control |
splitThreshold |
0.1 | Fat classifier threshold |
threshold |
— | Hematoxylin channel detection threshold for nuclei |
Cellpose TileFile null error: Increase downsample (e.g., 12.0 or 15.0).
No muscle fibers detected: Ensure ROI covers muscle tissue; try lowering cellprobThreshold (e.g., from 0 to −2).
Python ModuleNotFoundError: Run from the MyoPath/ directory, or use --input to specify the exports path.
Conda An unexpected error has occurred: Known conda 25.7.0 bug on Windows with GBK encoding. Does not affect results. Suppress with: CONDA_NO_PLUGINS=true conda run -n myopath python MyoPath4_analysis.py --all
Out of memory: Reduce --cores (each sample uses ~2–4 GB RAM).
Zhong H*, Gao M*, Ma S, Zhang W, Cheng N, Jiao K, Zhu B, Song J, Yan C, Yue D, Xi J, Du Z, Zhu W, Zhao C#, Luo S#. MyoPath: A deep learning pipeline for objective morphometric assessment of skeletal muscle biopsies. Journal of Pathology Informatics. 2026;100747. doi:10.1016/j.jpi.2026.100747
@article{zhong2026myopath,
title = {MyoPath: A deep learning pipeline for objective morphometric assessment of skeletal muscle biopsies},
author = {Zhong, Huahua and Gao, Mingshi and Ma, Shang and Zhang, Wenhan and Cheng, Nachuan and Jiao, Kexin and Zhu, Bochen and Song, Jie and Yan, Chong and Yue, Dongyue and Xi, Jianying and Du, Zunguo and Zhu, Wenhua and Zhao, Chongbo and Luo, Sushan},
journal = {Journal of Pathology Informatics},
pages = {100747},
year = {2026},
doi = {10.1016/j.jpi.2026.100747}
}MIT License.