Statistical segmentation model for accurate electrode positioning in Parkinson's deep brain stimulation based on clinical low-resolution image data and electrophysiology

Abstract

Background: Deep Brain Stimulation (DBS), applying chronic electrical stimulation of subcortical structures, is a clinical intervention applied in major neurologic disorders. In order to achieve a good clinical effect, accurate electrode placement is necessary. The primary localisation is typically based on presurgical MRI imaging, often followed by intra-operative electrophysiology recording to increase the accuracy and to compensate for brain shift, especially in cases where the surgical target is small, and there is low contrast: e.g., in Parkinson's disease (PD) and in its common target, the subthalamic nucleus (STN).

Methods: We propose a novel, fully automatic method for intra-operative surgical navigation. First, the surgical target is segmented in presurgical MRI images using a statistical shape-intensity model. Next, automated alignment with intra-operatively recorded microelectrode recordings is performed using a probabilistic model of STN electrophysiology. We apply the method to a dataset of 120 PD patients with clinical T2 1.5T images, of which 48 also had available microelectrode recordings (MER).

Results: The proposed segmentation method achieved STN segmentation accuracy around dice = 0.60 compared to manual segmentation. This is comparable to the state-of-the-art on low-resolution clinical MRI data. When combined with electrophysiology-based alignment, we achieved an accuracy of 0.85 for correctly including recording sites of STN-labelled MERs in the final STN volume.

Conclusion: The proposed method combines image-based segmentation of the subthalamic nucleus with microelectrode recordings to estimate their mutual location during the surgery in a fully automated process. Apart from its potential use in clinical targeting, the method can be used to map electrophysiological properties to specific parts of the basal ganglia structures and their vicinity.

Fig 1. Schematic overview of the operating principle of the pre-trained model when applied to a novel patient’s data.

Overview of model pre-training, as well as MRI and MER preprocessing are detailed in Fig 2.

Fig 2. Pipeline for the pre-training phase of the model, including MRI and MER data processing.

Fig 3. Representation of the shrinkage process in a patient native space.

The purple mesh is an ellipsoid that resulted from sphere initialisation on the MNI coordinates. A red mesh–resulting mesh representation of a 3D STN label.

Fig 4

LEFT: Sigmoid transition function (red), modelling the smooth increase in signal energy (NRMS) around the STN border (red dashed vertical line at 0 mm) overlaid on recorded values around the STN entry (blue dots) and mirrored data around the STN exit (green dots). It is apparent that the fitting of separate sigmoids to the STN entry (dashed blue) and the STN exit (dashed green) data separately results in a considerably steeper rise at the STN entry. However, for simplicity, we use only the mean trend (red line) for all points on the STN surface. RIGHT: Probability distribution of the normalised signal energy (NRMS) in the two states: inside the STN (red) and outside the STN (blue), each modelled using a log-normal distribution.

Fig 5. Boxplots of the dice overlap between the segmented structures, namely, Subthalamic nucleus (STN), Substantia Nigra (SN), Red Nucleus (RN) and the manual labels of 120 subjects.

All structures were segmented using the statistical segmentation method presented here. For segmentation of STN and SN we merged their components as the boundary between them is barely visible.

Fig 6. Segmentation result for STN (green) in comparison with manual label (purple) overlaid on an axial T2-weighed MRI slice.

For illustration purposes Red Nucleus is also included (red). The contours represent cross-sections of the 3D mesh labels.

Fig 7

Clustering of T2 MRI intensity profiles around the STN border into 4 clusters: their localisation on the mean right STN surface (LEFT) and the corresponding intensity profiles–mean per cluster (RIGHT) 3mm along the vertex norm. Steeper changes in image intensity around the STN border are apparent in the anterodorsal part of the STN (red and blue clusters). The mean STN shape is shifted arbitrarily, coordinates in milimeters.

Fig 8. Jet plot of likelihood of permanent electrode entry and exit, visualized on a STN mesh of an individual patient.

The calculation was based on trajectories determined using the electrode artefact in postoperative CT images. After coregistration to patient native MRI space, intersection with individual segmented STN 3D mesh was calculated for each patient. Owing to the point correspondence maintained by the proposed model, normal distribution of the electrode entry and exit points was calculated on the STN surface and transferred back to individual patient’s native space.

Fig 9. Electrode positioning within STN initialised (grey) and fitted position (purple).

Individual 5 MER electrodes are represented by the cyllinders. Width of the cyllinders represents normalised MER signal energy (NRMS) at individual recording sites. MER sites, annotated as within STN are marked yellow, outside STN in grey. Units in milimeters in individual patient’s native space.

Fig 10. Overview of transformation parameters, applied to the MRI-based automatic STN segmentation in the MER-fitting stage: Translations (Tx, Ty, Tz) and scalings (Sx, Sy, Sz).

The panel a) shows initial translation according to the maximum-likely entry and exit projected to patient native space, which is further refined by MER-fit in panel b) Panel c) shows MER-fit results initialized using surgical plan. Apparently, the overall character is retained, while eliminating outliers.