Image Registration

Automatic Co-Registration of MEG-MRI Data using Multiple RGB-D Cameras

Abstract

Integration of functional and structural modalities is essential to functional brain mapping. This paper presents an automatic co-registration system for aligning the coordinate systems between magnetoencephalography/ electroencephalography (MEG/EEG) and magnetic resonance image (MRI) using multiple off-the-shelf RGBD cameras.

The system was constructed by using multiple Kinects for Windows v2, which were calibrated for the integration of the captured data of subjects’ heads from multiple views. The integrated point clouds of the head surface captured by Kinects played an intermediate role between MEG/EEG and MRI. MEG/EEG-to-Kinect co-registration was conducted by using 3D locations of three anatomical landmarks, whereas Kinect-to-MRI co-registration was performed by using Gaussian mixture model to align facial part of points automatically segmented from both Kinect data and MRI. Combination of these two co-registration results yields the MEG/EEG-to-MRI transformation.

In our experiments, evaluation results showed that the proposed system can achieve coordinate system alignment with high accuracy.

 

Figure 1 : System flowchart.

 

Results

Figure 2 : Setting of the proposed (a) EEG and (b) MEG- dedicated co-registration system. (c) illustrates the graphical user interface of the proposed system.

 

Test Mean error(mm) SD
1 1.12 0.79
2 1.07 0.81
3 1.14 0.82
4 1.06 0.93
5 1.05 0.94
6 1.44 1.032
7 0.98 1.06
8 1.10 1.05
9 1.08 1.02
10 1.03 1.12
Table 1 : Accuracy of Kinect-to-MRI co-registration: Error was defined as the shortest Euclidean distances between a Kinect facial point co-registered to the MRI coordinate system and the facial surface defined by MRI facial points.

 

Figure 3 : The error distribution of Kinect-MRI registration of Subject 3.

 

Figure 4 : Repeatability of EEG/MEG marker localization of the proposed method, compared with Polhemus digitizer (10 repetitions were used ) (a) Mean and Median error (b) 3D-visualized comparison between two methods. (b) box plot of error measures.

 

 

Conclusions

We have proposed an automatic system for MEG–MRI and EEG-MRI co-registration based on multiple off-the-shelf Kinects for Windows v2. Our system was much more affordable compared to other dedicated co-registration systems.

The amount of manual operations needed is greatly reduced, resulting in higher time-efficiency as well as lesser errors due to manual operations.