SpinoSarc

An open-source desktop tool for quantitative paraspinal muscle and dural sac analysis on lumbar spine MRI.

⚠️ Research use only. SpinoSarc has not received FDA, CE, or any other regulatory approval. It must not be used for clinical decision-making. See DISCLAIMER.md.

---

Screenshots

Main application window: axial and sagittal viewers, patient demographics, automatic lumbar level detection, and per-muscle analysis results.

Overview

SpinoSarc is a native macOS application for automated quantitative analysis of paraspinal musculature and dural sac on lumbar spine MRI. It provides an end-to-end workflow from DICOM ingestion to PDF/Excel reporting, fully offline.

Key features:

• 🗂️ DICOM ingestion with multi-vendor support (Siemens, Philips, GE) including JPEG Lossless decompression
• 🧠 Paraspinal muscle segmentation via the MuscleMap U-Net backbone — 8 muscle classes (multifidus, erector spinae, psoas, quadratus lumborum; left and right)
• 🦴 Automatic lumbar level detection via TotalSpineSeg — labels intervertebral discs (L1-L2 … L5-S) and maps them onto the axial series
• 🩺 Automatic dural sac segmentation — cross-sectional area computed directly from the TotalSpineSeg canal mask (no manual ROI required)
• 🔬 Multi-level analysis — paraspinal muscle metrics + dural sac CSA at every detected level in a single run; levels falling in axial gaps use the nearest slice (flagged with distance)
• 📏 Canal narrowing flags — neutral measurement against published dural-sac thresholds (absolute <75, relative <100, early <130 mm²); a research aid for the radiologist, not a diagnosis
• 📐 Quantitative metrics: cross-sectional area (CSA), fat fraction, total psoas area (TPA), psoas muscle index (PMI)
• 📊 Reference values from published literature (Hamaguchi 2016, Englesbe 2010, Barz 2010) — for context, not classification
• 📄 PDF and Excel reports — single-slice and multi-level, for downstream analysis
• 🔒 Fully offline — no data leaves the device

Installation

macOS (Apple Silicon: M1, M2, M3, M4)

1. Download the latest release SpinoSarc-X.Y.Z.dmg from the Releases page.
2. Open the .dmg file.
3. Drag SpinoSarc.app to the Applications folder.
4. First launch only: Right-click SpinoSarc.app in Applications → "Open" → "Open" again (to bypass macOS Gatekeeper, as the app is not code-signed).

System requirements:

• macOS 11.0 (Big Sur) or later
• Apple Silicon Mac (M1 / M2 / M3 / M4)
• ~2 GB free disk space
• No internet connection required

From source (developers)

git clone https://github.com/neuromath/spinosarc.git
cd spinosarc
conda create -n spinosarc python=3.11
conda activate spinosarc
conda install -c conda-forge dcm2niix
pip install -r requirements.txt

# Clone MuscleMap separately (segmentation backbone)
git clone https://github.com/MuscleMap/MuscleMap.git ~/SpinoSarc/MuscleMap

# Run from source
python -m spinosarc_app.gui

Quick Start

1. Launch SpinoSarc
2. Drag a lumbar MR DICOM folder (or NIfTI files) onto the drop zone
3. Enter patient demographics (optional, used for PMI calculation)
4. Click "Detect Levels" — TotalSpineSeg labels the lumbar discs and segments the spinal canal (~60 seconds on Apple Silicon, first run)
5. Click a level in the list to jump to its axial slice
6. Click "Analyze" for the current slice, or "Analyze All Levels" to run every detected level (muscle + dural sac + canal flags) in one pass
7. Export results as PDF or Excel — the report is multi-level when a multi-level analysis has been run
8. Click "New Case" to clear and start over

What SpinoSarc Reports

For each analyzed slice:

Metric	Description
Per-muscle CSA (mm²)	Cross-sectional area for each of 8 muscles
Per-muscle fat fraction (%)	Intensity-based fat estimate
Total Psoas Area (cm²)	Sum of left + right psoas CSA
Psoas Muscle Index (cm²/m²)	TPA normalized by patient height (if provided)
Dural Sac CSA (mm²)	Automatic, from the TotalSpineSeg canal mask
Canal narrowing flag	Neutral comparison against literature thresholds (not a diagnosis)

When Analyze All Levels is used, these metrics are reported for every detected level (L1-L2 … L5-S) plus L3-based sarcopenia, with flags for levels measured from a nearby slice (axial gaps) and a caveat at L5-S where the thresholds are less reliable.

Reference cut-off values from the literature are displayed for context but no automated diagnosis is performed.

How It Differs From MuscleMap

SpinoSarc uses MuscleMap as its segmentation engine. The contribution of SpinoSarc is the clinical research infrastructure surrounding the segmentation:

• DICOM ingestion (MuscleMap requires NIfTI)
• Anatomical coordinate alignment between axial and sagittal series
• Automatic lumbar level detection and dural sac segmentation (TotalSpineSeg)
• Multi-level analysis across all lumbar levels in a single run
• Quantitative metric extraction with literature reference values
• Neutral canal-narrowing flags (research aid, not diagnosis)
• PDF / Excel reporting (single-slice and multi-level)
• Native macOS desktop application — no programming required

Citation

If you use SpinoSarc in your research, please cite:

@software{yilmaz2026spinosarc,
  author       = {Yılmaz, Berkay},
  title        = {SpinoSarc: An open-source tool for quantitative
                  paraspinal muscle and dural sac analysis on lumbar
                  spine MRI},
  year         = {2026},
  url          = {https://github.com/neuromath/SpinoSarc},
  version      = {0.1.0},
  doi          = {10.5281/zenodo.20760332},
  orcid        = {0009-0006-3108-8991}
}

Please also cite the underlying segmentation backbone:

@software{musclemap,
  author = {{MuscleMap contributors}},
  title  = {MuscleMap: Open-source skeletal muscle segmentation},
  url    = {https://github.com/MuscleMap/MuscleMap}
}

License

SpinoSarc is released under the MIT License.

Third-party components are distributed under their respective licenses; see LICENSE for the full list.

Acknowledgments

• MuscleMap team for the open-source segmentation backbone
• MONAI and PyTorch communities
• dcm2niix (Chris Rorden) for robust DICOM-to-NIfTI conversion
• Cerrahpaşa Faculty of Medicine for clinical context
• Reference literature:
  • Barz T et al. (2010) — Dural sac CSA thresholds
  • Hamaguchi Y et al. (2016) — PMI reference values
  • Englesbe MJ et al. (2010) — Psoas area in surgical risk

Author

Berkay Yılmaz Radiology Resident, Cerrahpaşa Faculty of Medicine Istanbul University-Cerrahpaşa, Istanbul, Turkey ORCID: 0009-0006-3108-8991 GitHub: @neuromath

Contributing

Issues and pull requests are welcome. Please open a GitHub issue for bug reports or feature requests.

For substantive contributions, please ensure:

• No patient data is committed (see .gitignore)
• Code is documented in English
• Changes are tested on at least one public dataset (e.g., RSNA 2024 Lumbar Spine)

Disclaimer

Research use only. See DISCLAIMER.md for full terms.