MedNet-CMB

Pretrained nnU-Net model for automated cerebral microbleed segmentation in T2*/SWI MRI.

Install

This project uses Conda for dependency management.

git clone https://github.com/iBrain-Lab/MedNet-CMB.git MedNet-CMB
cd MedNet-CMB

conda env create -f environment.yml
conda activate MedNet-CMB

pip install -e .

Weights

Download the weights from Google Drive and unzip them so MedNet-CMB_weights/nnUNet_results/ exists. The directory should contain:

MedNet-CMB_weights/
└── nnUNet_results/
    └── Dataset012_ASPREE_VALDO_CSIRO_stratified_1x1x1_relabeled_v2/
        └── nnUNetTrainer__nnUNetPlans__3d_fullres/
            ├── fold_0/checkpoint_final.pth
            ├── fold_1/checkpoint_final.pth
            ├── fold_2/checkpoint_final.pth
            ├── fold_3/checkpoint_final.pth
            └── fold_4/checkpoint_final.pth

Inference Usage

1. Prepare your input images:

• Place your NIfTI images named *_0000.nii.gz into a single input folder.

2. Open the script and edit the path variables to match your setup:

• WEIGHTS_DIR: full path to the folder containing MedNet-CMB_weights
• INPUT_DIR: folder containing the input images
• OUTPUT_DIR: folder where predictions will be saved

3. Run the script!

bash run_MedNet-CMB_nnunet.sh

Results

Cohort	Cluster Dice	Precision	Sensitivity	False Positives / scan
Internal (n=264)	0.82	0.88	0.77	0.58
AIBL	0.89	0.87	0.94	-
ASPREE	0.58	0.68	0.55	-
VALDO	0.70	0.76	0.69	-
External DOU (n=20)	0.76	0.83	0.79	0.95