NiChart_DLWMLS

Run Deep-Learning-based-White-Matter-Lesion-Segmentation on your data (requires FLAIR, optional T1 masks for granular segmentation).

Executing the full pipeline including seperating WMLS mask into Brain ROI level based on the input DLMUSE masks.

Installation

0. Create a conda env (Python >= 3.9)

1. Install DLWMLS (Required dependency)

git clone https://github.com/CBICA/DLWMLS.git
cd DLWMLS
pip install -e .

1. Install wmh_seg (Optional)

git clone https://github.com/euroso97/wmh_seg.git
cd wmh_seg
wget https://huggingface.co/jil202/wmh_seg/resolve/main/ChallengeMatched_Unet_mit_b5.pth

2. Install NiChart_DLWMLS

git clone https://github.com/CBICA/NiChart_DLWMLS.git
cd NiChart_DLWMLS
pip install -e .

Usage

Prerequisits:

• T1 image (.nii.gz)
• FL image (.nii.gz)
• DLMUSE mask (.nii.gz) (refer to: NiChart_DLMUSE)
• (*optional) your own WMH mask (.nii.gz)

Required arguments:

[-fl, --fl_dir] : Name of the input folder with FL scans  (REQUIRED)
[-o, --out_dir] : Name of the output folder for segmentation (REQUIRED)
[--list]        List of MRIDs; first raw (column header) skipped (OPTIONAL)
[--t1_dir]      Name of the input folder with T1 scans  (OPTIONAL)
[--t1_suff]     Suffix of the input T1 scans (OPTIONAL, DEFAULT: _T1.nii.gz)
[--dlmuse_dir]  Name of the input folder with T1 scans  (OPTIONAL)
[--dlmuse_suff] Suffix of the input T1 scans (OPTIONAL, DEFAULT: _T1_LPS_DLMUSE.nii.gz)

Optional arguments:

[-d, --device]  Device to run segmentation ('cuda' (GPU), 'cpu' (CPU) or 
                'mps' (Apple M-series chips supporting 3D CNN))
[-h, --help]    Show this help message and exit.
[-V, --version] Show program's version number and exit.

EXAMPLE USAGE:

Executing the full pipeline including seperating WMLS mask into Brain ROI level 
    based on the input DLMUSE masks:

NiChart_DLWMLS  --list          /path/to/mrid_list.csv \
                --fl_dir        /path/to/flair_images  \
                --fl_suff       _FL_LPS.nii.gz         \
                --t1_dir        /path/to/t1_images     \
                --t1_suff       _T1_LPS.nii.gz         \
                --dlmuse_dir    /path/to/dlmuse_masks  \
                --dlmuse_suff   _T1_LPS_DLMUSE.nii.gz  \
                --out_dir       /path/to/output


Using your own WMH masks (skipping DLWMLS segmentation):

    NiChart_DLWMLS  --list          /path/to/mrid_list.csv \
                    --fl_dir        /path/to/flair_images  \
                    --fl_suff       _FL.nii.gz             \
                    --wmh_dir       /path/to/dlwmls_masks  \
                    --wmh_suff      _FL_LPS_DLWMLS.nii.gz  \
                    --t1_dir        /path/to/t1_images     \
                    --t1_suff       _T1.nii.gz             \
                    --dlmuse_dir    /path/to/dlmuse_masks  \
                    --dlmuse_suff   _T1_LPS_DLMUSE.nii.gz  \
                    --out_dir       /path/to/output        \
                    --remove_intermediate True             \
                    --device cpu/cuda