ALICE: A tool for automatic localization of intra-cranial electrodes for clinical and high-density grids

Abstract

Background: Electrocorticographic (ECoG) measurements require the accurate localization of implanted electrodes with respect to the subject's neuroanatomy. Electrode localization is particularly relevant to associate structure with function. Several procedures have attempted to solve this problem, namely by co-registering a post-operative computed tomography (CT) scan, with a pre-operative magnetic resonance imaging (MRI) anatomy scan. However, this type of procedure requires a manual and time-consuming detection and transcription of the electrode coordinates from the CT volume scan and restricts the extraction of smaller high-resolution ECoG grid electrodes due to the downsampling of the CT.

New method: ALICE automatically detects electrodes on the post-operative high-resolution CT scan, visualizes them in a combined 2D and 3D volume space using AFNI and SUMA software and then projects the electrodes on the individual's cortical surface rendering. The pipeline integrates the multiple-step method into a user-friendly GUI in Matlab®, thus providing an easy, automated and standard tool for ECoG electrode localization.

Results: ALICE was validated in 13 subjects implanted with clinical ECoG grids by comparing the calculated electrode center-of-mass coordinates with those computed using a commonly used method.

Comparison with existing methods: A novel aspect of ALICE is the combined 2D-3D visualization of the electrodes on the CT scan and the option to also detect high-density ECoG grids. Feasibility was shown in 5 subjects and validated for 2 subjects.

Conclusions: The ALICE pipeline provides a fast and accurate detection, discrimination and localization of ECoG electrodes spaced down to 4 mm apart.

Keywords: ALICE; Clinical grid; ECoG; Electrode localization; High-density grid; Integrated pipeline.

Copyright © 2017 Elsevier B.V. All rights reserved.

Figures

Figure 1. ALICE pipeline back-end procedure

The back-end of the pipeline consists of 5 steps. The first step co-registers the high-resolution post-operative CT scan (post-CT) with the pre-operative anatomical MRI scan (pre-MRI). The second step extracts the center-of-mass for every electrode using a new automated voxel clustering function in AFNI. Then in step 3 one can sort and label each electrode’s center-of-mass according to a pre-defined layout. Step 4 co-registers the center-of-mass of each electrode extracted from the CT space to the MRI anatomy space, using the affine transformation matrix resulting from step 1. Finally, in step 5 the electrodes are projected and visualized on the surface rendering of each individual’s brain by means of the standard orthogonal projection method presented in Hermes et al. (2010).

Figure 2. AFNI-SUMA interface for electrode selection

A) SUMA displays the information in a 3D surface space. B) AFNI displays the CT and the electrode-clusters in the volume space. These two programs are linked and synchronized, such that an electrode selected in (A) (black arrow) immediately updates the position in the volumetric view in (B) (green vertical and horizontal crosshair lines).

Figure 3. Validation

For each subject, the electrode coordinates calculated using the method implemented in Hermes et al. (2010) (white circles) and ALICE (blue crosses) were displayed (in Matlab 9.0®) on each individuals’ FreeSurfer cortical surface rendering of the implanted hemisphere. The mean and standard deviation distance (in mm) across the electrode coordinates, d, obtained by the two methods is displayed beneath each subject’s brain surface. N represents the number of projected electrodes for each subject. The black shaded region in S13 indicates previously resected cortex. Whether the blue cross or the white circle is displayed on top depends on the 3D coordinates of the electrode.

Figure 4. Validation results for clinical grids

Box-plots of the Euclidean distance between projected electrodes obtained using ALICE and Hermes et al. (2010) methods for all subjects (i.e., median distance from all subjects) and per individual subject. In total 744 electrodes. White circles with gray dot in the middle indicate the median distance value; the thick gray bars indicate the 50% of the distribution; the thin black lines indicate the maximum and minimum distance; and the outliers are indicated by empty white circles (i.e., points that are larger than Q3 + 1.5 × (Q3 − Q1) or smaller than Q1 − 1.5 × (Q3 − Q1), with Q1 and Q3 the 25th and 75th percentile, respectively). A uniform random displacement jitter is used in order to make duplicate outlier-points visible.

Figure 5. Detection of overlapped electrodes

Example of 6 overlapping electrodes in subject S12. The 3D visualization of the grids allows the user to identify to which grid the electrode belongs to and by knowing in advance the clinical grid layout. Even though these were not separated in the clustering process, these could be manually detected and added as an extra cluster by placing a sphere with center coincident with the center of the electrode as detected in the volume space (see Figure 2). The small (yellow) electrode indicated with * is the standard platinum marker of the grid.

Figure 6. High-density grid electrode detection

Electrode clustering in SUMA (insets) of high-density ECoG grids. Visualization of the electrodes was achieved using the method described in Kubanek and Schalk (2015) and displayed on the FreeSurfer cortex rendering. A) For a grid with 4mm inter-electrode spacing the high-resolution CT has enough resolution to separate all electrodes automatically. 3D SUMA surface visualization (inset) shows independent clusters for all electrodes. B) Grids with 3 mm electrode distance require some manual interaction (white-circled electrodes). The black shaded region in S13 indicates previously resected cortex.

Figure 7. Validation of high-density grid projection

An intraoperative photo (B) is used to mark the blood vessels (blue, D) and sulci (green, E) for subject S14. The sulci in the brain rendering (C) were also identified (red, H). Linear and non-linear transformation (to account for cortical curvature lost in the photo) were used to map the sulci between (H) and (E), resulting in a match (I) between the sulci in the reference photo (B) and the sulci in (H) using Photoshop®. A similar procedure was used to match the blood vessels (D) to the intraoperative photo with the high-density grid. The electrodes mapped to the reference photo (yellow circles) could be then compared with ones from the CT (blue circles) (J). The gray circles in (J) had no match with intraoperative photo due to lack of visibility.

Figure 8. Validation results for high-density grids

Box-plots of the Euclidean distance between projected electrodes obtained using ALICE and the electrodes mapped to an intraoperative photo. White circles with gray dot in the middle indicate the median distance value; the thick gray bars indicate the 50% of the distribution; the thin black lines indicate the maximum and minimum distance; and the outliers are indicated by empty white circles (i.e., points that are larger than Q3 + 1.5 × (Q3 − Q1) or smaller than Q1 − 1.5 × (Q3 − Q1), with Q1 and Q3 the 25th and 75th percentile, respectively).