A thigh muscle segmentation tool trained using a multimodal and multiethnic MRI dataset
- A 3D segmentation model based on SegResNet
- Trained using multi-ethnic thigh MRIs from both myopathy patients and healthy controls (metadata available in the repository)
- Suitable for training on any one of the multimodal MRI series, including IDEAL fat, IDEAL water, T1, T2, and STIR
- Targets 11 thigh muscles: BL - Biceps Femoris Long Head; BB - Biceps Femoris Short Head; ST - Semitendinosus; SM - Semimembranosus; AM - Adductor Magnus; VI - Vastus Intermedius; VL - Vastus Lateralis; VM - Vastus Medialis; RF - Rectus Femoris; GR - Gracilis; SA - Sartorius.
You can download the anonymized Chinese and Finns MRI data for your research at the "Releases" section, and the Germany data from this link.
- Han Chinese Thigh MRIs (HuashanMyo): 262 MRIs (LGMDR1, LGMDR2, BMD, DM1, Control)
- Finnish Thigh MRIs (Folkhälsan Research Center): 54 MRIs (TMD, IBM, DM2, HMERF)
- Germany Thigh MRIs (MyoSegmenTUM): 38 MRIs (DM2, LGMDR1, ALS, Control)
- The Belgian BMD IDEAL In-phase MRI samples used for the model demo were download from LHuysmans/3d-muscle-segmentation
- Install MONAI dependencies first.
pip install "monai-weekly[all]" "monailabel-weekly"
-
Then, copy the contents (including scripts and the model) of the "MONAI label implementation" folder into the MONAI label's config file located at "./apps/radiology".
-
Start MONAI Label server
# OHIF
monailabel start_server --app apps/radiology --studies http://127.0.0.1:8042/dicom-web --conf models segmentation_thighmuscles
# 3D Slicer
monailabel start_server --app apps/radiology --studies datasets/Thigh_muscles --conf models segmentation_muscle --conf skip_trainers true
- Start "OHIF viewer" or "3D slicer" to run the model on your own MRI data. Press the "Auto segmentation" button to start.
We would like to express our heartfelt gratitude to the patients and their families for their invaluable contributions to the MRI dataset. We also extend special thanks to Dr. Sarah Schlaeger and Dr. Lotte Huysmans for generously sharing their MRI data for further research.