Automatic Extraction of Implanted Electrode Grids

Oskar M. Skrinjar, James S. DuncanMICCAI 1999 · DOI

Introduction

In preparation for epilepsy surgery, surgeons often implant grids of electrodes between the skull and the brain to map functional areas. To use these maps effectively, clinicians must know exactly where each electrode sits relative to the patient's brain anatomy.

The standard tool for this is a post-operative MR scan. However, metal electrodes create significant "artifacts"—large, dark, sphere-shaped distortions that obscure nearby tissue. Manually identifying 64+ electrodes across dozens of 2D image slices is tedious, error-prone, and makes it difficult to visualize the grid's overall curvature.

MRI artifacts caused by metallic electrodes. Arrows indicate the dark, spherical distortions that obscure tissue.

Methodology: A Geometry-Driven Approach

The authors propose a three-stage pipeline that leverages the known physical geometry of the electrode grid (e.g., 10mm spacing) to overcome image noise.

  1. Nonlinear Filtering: The algorithm uses a custom 3D kernel to search for dark spheres surrounded by lighter tissue, identifying potential electrode centers based on voxel brightness thresholds.
  2. Predictive Filtering: The system establishes a "low-level topology" by identifying links and squares of centers at the correct distance. It then fits a grid template over these squares. Crucially, if an electrode is missing due to scan noise, the algorithm predicts its location based on its neighbors.
  3. Regularization and Surface Interpolation: The grid is modeled as a deformable but non-stretchable surface. The algorithm iteratively adjusts positions to match the ideal 10mm spacing and calculates surface normals to properly orient the electrodes in a 3D render.

Results and Clinical Impact

The automated extraction was compared to manual positioning, showing an error within 1.5 mm for most electrodes, and often under 1 mm. This level of precision is considered sufficient for clinical applications.

Final extracted electrode grid reconstructed as a 3D mesh and overlaid on the patient's brain anatomy.

The entire procedure takes only a few minutes on an SGI Octane workstation, a significant speedup over manual identification. By moving from 2D slices to a 3D surface model, the authors allow surgeons to see the electrodes "sitting" on the brain folds (gyri and sulci), which is much more intuitive for planning a resection. At the time of publication, the tool was already in active clinical use within the Department of Neurosurgery at Yale University.

Conclusion

By treating the grid as a single geometric structure rather than independent points, this system accurately reconstructs 3D positions even in the presence of heavy metallic artifacts. While currently optimized for rectangular grids, the underlying principles of geometric priors offer a robust path for automating surgical mapping.

Summary updated

BibTeX
@inproceedings{skrinjar1999automatic,
  author = {Oskar M. Skrinjar and James S. Duncan},
  title = {{Automatic Extraction of Implanted Electrode Grids}},
  booktitle = {International Conference on Medical Image Computing and Computer-Assisted Intervention},
  year = {1999},
  doi = {10.1007/10704282\_108},
}
Figure