MELD-PostOp

Preprint: here

MELD-PostOp is an automated tool for segmenting resection cavities on postoperative T1w MRI scans of epilepsy patients, developed using nnU-Net framework. This page provides instructions for installing the pretrained MELD-PostOp model and running inference on new patients.

LICENSE

We kindly ask all MELD-PostOp users to complete the MELD-PostOp registration form. After submission, you will receive a license file, which is required to use MELD-PostOp version v1.0.0.

MELD-PostOp Registration Form

Your email address will also be added to the MELD-PostOp mailing list so we can keep you informed about bug fixes and new releases.

System requirements

• The instruction below was tested on Linux (Ubuntu 22.04.5)
• Python: 3.9 or newer
• Hardware: GPU is strongly recommended

1. MELD-PostOp installation & setup

1.1 Create a virtual environment

We recommend using Anaconda to manage the Python environment and dependencies. Please follow the official installation instructions for Anaconda.

conda create -n meldpostop python=3.13.5
conda activate meldpostop

1.2 Install nnU-Net v2

MELD-PostOp is developed on the nnU-Net framework. For additional details, please refer to the official nnU-Net.

pip install nnunetv2

1.3 Download the pretrained model

Download the pretrained MELD-PostOp model archive (MELD-PostOp.zip) from Figshare and unzip it locally:

• MELD-PostOp Figshare
• version: v1.0.0

1.4 License setup

A valid MELD-PostOp licencse file is required to run inference. After completing the MELD-PostOp registration form, you will receive a license file. Place the license file in: /MELD-PostOp/model/. Inference will fail if no valid license file is found.

1.5 Directory structure

After completing the installation and setup, the MELD-PostOp directory should have the following structure:

MELD-PostOp/
├── input_001/
├── bin/
│   └── run_meldpostop_v1.sh
├── model/
│   ├── (MELD-PostOp license file) 
│   └── Dataset001_MELDPostOp_v1.0.0/
│       └── nnUNetTrainer__nnUNetPlans__3d_fullres/     < (pretrained model weights) 

2. Data preparation

2.1 Input data requirements

• Input images must be postoperative T1-weighted MRI scans
• Images must be in NIfTi format (.nii or .nii.gz)
• NOTE: MELD-PostOp was trained on postoperative T1w MRI scans from patients who had lesionectomy or lobectomy. It may not be appropriate for other procedures (e.g. laser ablation, thermocoagulation etc).

2.2 File naming

• All input files must follow nnU-Net naming conventions
• Each file name must include:
  • a subject id
  • the modality suffix _0000, indicating T1w
• Exmaple: sub-<subject_id>_0000.nii.gz

2.3 Input directory

• Upon download, an inital input directory (input_001) is provided
• To process multiple datasets, create additional input directories named input_<key>, where <key> is a dataset identifier (we recommend a 3-digit number, e.g. input_003,input_027)
• Place all NIfTI files for a given dataset inside the corresponding input directory
• Output directories are generated automatically:
  • input_001 → output_001
  • input_002 → output_002
  • No manual creation of output directories is required

3. Running MELD-PostOp inference

3.1 Directory&command setup

cd path/to/MELD-PostOp
chmod +x bin/run_meldpostop_v1.sh

3.2 Inference

To run inference on new patients, execute:

bin/run_meldpostop_v1.sh -m </path/to/MELD-PostOp> -k <key> 

Additional information

After running inference with MELD-PostOp, you may wish to perform external validation of the model on your dataset. Please refer to the following link for detailed instructions:

• notebooks

If you wish to recerate figures in the manuscript, please refer to the following link for detailed instructions:

• figures