DBS-ElecNet

A method for automated localization and segmentation of DBS electrodes in clinical MRI

Vanessa H. Yu, Edward Chen, Jürgen Germann, Alexandre Boutet, Andres M. Lozano, Kâmil Uludağ, and Sriranga Kashyap

V. H. Yu et al., "DBS-ElecNet: Automated Localization and Segmentation of DBS Electrodes in Clinical MRI," 2026 IEEE 23rd International Symposium on Biomedical Imaging (ISBI), London, United Kingdom, 2026, pp. 1-4, doi:10.1109/ISBI61048.2026.11515562

Poster available at: https://zenodo.org/records/19430997

Table of Contents

1. Requirements
2. SALINE Electrode Segmentation
3. DBS-ElecNet Inference

Requirements

Python

• Python 3.9+
• Install Python dependencies:

python3 -m pip install -r requirements.txt

External command-line tools

The run_saline and run_elecnet scripts call external neuroimaging tools that must be available on your PATH:

• ANTs (ResampleImage, N4BiasFieldCorrection)
• FreeSurfer / SynthStrip / SynthSeg (mri_synthstrip, mri_synthseg)

SALINE electrode segmentation:

This pipeline performs single- or dual-electrode localization using the SALINE framework.

Command:

./run_saline <nifti_file> <num_elec>

Required arguments:

• nifti_file: Path to the input MRI volume (NIfTI format).

• num_elec: Number of implanted electrodes (1 or 2).

Output:

• subject_saline_elec.nii.gz: Electrode segmentation in the native image space.

• subject_1mm_iso.nii.gz: Input MRI resampled to 1 mm isotropic resolution.

• subject_1mm_iso_saline_elec.nii.gz: Electrode segmentation in 1 mm isotropic space.

DBS-ElecNet Inference:

This pipeline performs electrode segmentation using DBS-ElecNet.

Command:

./run_elecnet <nifti_file> <device>

Required arguments:

• nifti_file: Path to the input MRI volume (NIfTI format).

• device: Inference device, following PyTorch conventions (cpu, cuda, cuda:0, etc.).

Output:

• subject_elec.nii.gz: Electrode segmentation in the native image space.

• subject_1mm_iso.nii.gz: Input MRI resampled to 1 mm isotropic resolution.

• subject_1mm_iso_elec.nii.gz: Electrode segmentation in 1 mm isotropic space.

Note:

• All preprocessing steps are performed automatically, including resampling to 1 mm isotropic resolution, skull stripping, N4 bias field correction, and (for SALINE) SynthSeg.
• Intermediate files are removed upon completion except for the 1 mm isotropic resampled MRI (subject_1mm_iso.nii.gz), which is retained for reference.
• Final outputs are provided in both 1 mm isotropic space and the original native image space.